idnits 2.17.1 draft-dt-teas-rfc3272bis-05.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (December 6, 2019) is 1600 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Obsolete informational reference (is this intentional?): RFC 3272 (Obsoleted by RFC 9522) -- Obsolete informational reference (is this intentional?): RFC 7752 (Obsoleted by RFC 9552) Summary: 0 errors (**), 0 flaws (~~), 1 warning (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 TEAS Working Group A. Farrel, Ed. 3 Internet-Draft Old Dog Consulting 4 Obsoletes: 3272 (if approved) December 6, 2019 5 Intended status: Informational 6 Expires: June 8, 2020 8 Overview and Principles of Internet Traffic Engineering 9 draft-dt-teas-rfc3272bis-05 11 Abstract 13 This memo describes the principles of Traffic Engineering (TE) in the 14 Internet. The document is intended to promote better understanding 15 of the issues surrounding traffic engineering in IP networks, and to 16 provide a common basis for the development of traffic engineering 17 capabilities for the Internet. The principles, architectures, and 18 methodologies for performance evaluation and performance optimization 19 of operational IP networks are discussed throughout this document. 21 This work was first published as RFC 3272 in May 2002. This document 22 obsoletes RFC 3272 by making a complete update to bring the text in 23 line with current best practices for Internet traffic engineering and 24 to include references to the latest relevant work in the IETF. 26 Status of This Memo 28 This Internet-Draft is submitted in full conformance with the 29 provisions of BCP 78 and BCP 79. 31 Internet-Drafts are working documents of the Internet Engineering 32 Task Force (IETF). Note that other groups may also distribute 33 working documents as Internet-Drafts. The list of current Internet- 34 Drafts is at https://datatracker.ietf.org/drafts/current/. 36 Internet-Drafts are draft documents valid for a maximum of six months 37 and may be updated, replaced, or obsoleted by other documents at any 38 time. It is inappropriate to use Internet-Drafts as reference 39 material or to cite them other than as "work in progress." 41 This Internet-Draft will expire on June 8, 2020. 43 Copyright Notice 45 Copyright (c) 2019 IETF Trust and the persons identified as the 46 document authors. All rights reserved. 48 This document is subject to BCP 78 and the IETF Trust's Legal 49 Provisions Relating to IETF Documents 50 (https://trustee.ietf.org/license-info) in effect on the date of 51 publication of this document. Please review these documents 52 carefully, as they describe your rights and restrictions with respect 53 to this document. Code Components extracted from this document must 54 include Simplified BSD License text as described in Section 4.e of 55 the Trust Legal Provisions and are provided without warranty as 56 described in the Simplified BSD License. 58 Table of Contents 60 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 61 1.1. What is Internet Traffic Engineering? . . . . . . . . . . 4 62 1.2. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 7 63 1.3. Terminology . . . . . . . . . . . . . . . . . . . . . . . 7 64 2. Background . . . . . . . . . . . . . . . . . . . . . . . . . 10 65 2.1. Context of Internet Traffic Engineering . . . . . . . . . 11 66 2.2. Network Context . . . . . . . . . . . . . . . . . . . . . 12 67 2.3. Problem Context . . . . . . . . . . . . . . . . . . . . . 13 68 2.3.1. Congestion and its Ramifications . . . . . . . . . . 15 69 2.4. Solution Context . . . . . . . . . . . . . . . . . . . . 15 70 2.4.1. Combating the Congestion Problem . . . . . . . . . . 17 71 2.5. Implementation and Operational Context . . . . . . . . . 20 72 2.6. High-Level Objectives . . . . . . . . . . . . . . . . . . 21 73 3. Traffic Engineering Process Models . . . . . . . . . . . . . 23 74 3.1. Components of the Traffic Engineering Process Model . . . 24 75 3.2. Measurement . . . . . . . . . . . . . . . . . . . . . . . 24 76 3.3. Modeling, Analysis, and Simulation . . . . . . . . . . . 25 77 3.4. Optimization . . . . . . . . . . . . . . . . . . . . . . 26 78 4. Review of TE Techniques . . . . . . . . . . . . . . . . . . . 27 79 4.1. Historic Overview . . . . . . . . . . . . . . . . . . . . 28 80 4.1.1. Traffic Engineering in Classical Telephone Networks . 28 81 4.1.2. Evolution of Traffic Engineering in Packet Networks . 29 82 4.2. Development of Internet Traffic Engineering . . . . . . . 32 83 4.2.1. Overlay Model . . . . . . . . . . . . . . . . . . . . 32 84 4.2.2. Constraint-Based Routing . . . . . . . . . . . . . . 33 85 4.3. Overview of IETF Projects Related to Traffic Engineering 33 86 4.3.1. Integrated Services . . . . . . . . . . . . . . . . . 34 87 4.3.2. RSVP . . . . . . . . . . . . . . . . . . . . . . . . 34 88 4.3.3. Differentiated Services . . . . . . . . . . . . . . . 35 89 4.3.4. MPLS . . . . . . . . . . . . . . . . . . . . . . . . 36 90 4.3.5. Generalized MPLS . . . . . . . . . . . . . . . . . . 37 91 4.3.6. IP Performance Metrics . . . . . . . . . . . . . . . 37 92 4.3.7. Flow Measurement . . . . . . . . . . . . . . . . . . 38 93 4.3.8. Endpoint Congestion Management . . . . . . . . . . . 38 94 4.3.9. TE Extensions to the IGPs . . . . . . . . . . . . . . 39 95 4.3.10. Link-State BGP . . . . . . . . . . . . . . . . . . . 39 96 4.3.11. Path Computation Element . . . . . . . . . . . . . . 39 97 4.3.12. Application-Layer Traffic Optimization . . . . . . . 40 98 4.3.13. Segment Routing . . . . . . . . . . . . . . . . . . . 40 99 4.3.14. Network Virtualization and Abstraction . . . . . . . 40 100 4.3.15. Deterministic Networking . . . . . . . . . . . . . . 40 101 4.3.16. Network TE State Definition and Presentation . . . . 40 102 4.3.17. System Management and Control Interfaces . . . . . . 40 103 4.4. Overview of ITU Activities Related to Traffic Engineering 41 104 4.5. Content Distribution . . . . . . . . . . . . . . . . . . 42 105 5. Taxonomy of Traffic Engineering Systems . . . . . . . . . . . 43 106 5.1. Time-Dependent Versus State-Dependent Versus Event 107 Dependent . . . . . . . . . . . . . . . . . . . . . . . . 43 108 5.2. Offline Versus Online . . . . . . . . . . . . . . . . . . 45 109 5.3. Centralized Versus Distributed . . . . . . . . . . . . . 45 110 5.3.1. Hybrid Systems . . . . . . . . . . . . . . . . . . . 45 111 5.3.2. Considerations for Software Defined Networking . . . 45 112 5.4. Local Versus Global . . . . . . . . . . . . . . . . . . . 45 113 5.5. Prescriptive Versus Descriptive . . . . . . . . . . . . . 46 114 5.5.1. Intent-Based Networking . . . . . . . . . . . . . . . 46 115 5.6. Open-Loop Versus Closed-Loop . . . . . . . . . . . . . . 46 116 5.7. Tactical vs Strategic . . . . . . . . . . . . . . . . . . 47 117 6. Objectives for Internet Traffic Engineering . . . . . . . . . 47 118 6.1. Routing . . . . . . . . . . . . . . . . . . . . . . . . . 47 119 6.2. Traffic Mapping . . . . . . . . . . . . . . . . . . . . . 50 120 6.3. Measurement . . . . . . . . . . . . . . . . . . . . . . . 51 121 6.4. Network Survivability . . . . . . . . . . . . . . . . . . 52 122 6.4.1. Survivability in MPLS Based Networks . . . . . . . . 54 123 6.4.2. Protection Option . . . . . . . . . . . . . . . . . . 55 124 6.5. Traffic Engineering in Diffserv Environments . . . . . . 56 125 6.6. Network Controllability . . . . . . . . . . . . . . . . . 58 126 7. Inter-Domain Considerations . . . . . . . . . . . . . . . . . 58 127 8. Overview of Contemporary TE Practices in Operational IP 128 Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 61 129 9. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 65 130 10. Security Considerations . . . . . . . . . . . . . . . . . . . 65 131 11. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 65 132 12. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 65 133 13. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 66 134 14. Informative References . . . . . . . . . . . . . . . . . . . 68 135 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 74 137 1. Introduction 139 This memo describes the principles of Internet traffic engineering. 140 The objective of the document is to articulate the general issues and 141 principles for Internet traffic engineering; and where appropriate to 142 provide recommendations, guidelines, and options for the development 143 of online and offline Internet traffic engineering capabilities and 144 support systems. 146 This document can aid service providers in devising and implementing 147 traffic engineering solutions for their networks. Networking 148 hardware and software vendors will also find this document helpful in 149 the development of mechanisms and support systems for the Internet 150 environment that support the traffic engineering function. 152 This document provides a terminology for describing and understanding 153 common Internet traffic engineering concepts. This document also 154 provides a taxonomy of known traffic engineering styles. In this 155 context, a traffic engineering style abstracts important aspects from 156 a traffic engineering methodology. Traffic engineering styles can be 157 viewed in different ways depending upon the specific context in which 158 they are used and the specific purpose which they serve. The 159 combination of styles and views results in a natural taxonomy of 160 traffic engineering systems. 162 Even though Internet traffic engineering is most effective when 163 applied end-to-end, the initial focus of this document document is 164 intra-domain traffic engineering (that is, traffic engineering within 165 a given autonomous system). However, because a preponderance of 166 Internet traffic tends to be inter-domain (originating in one 167 autonomous system and terminating in another), this document provides 168 an overview of aspects pertaining to inter-domain traffic 169 engineering. 171 This work was first published as [RFC3272] in May 2002. This 172 document obsoletes [RFC3272] by making a complete update to bring the 173 text in line with current best practices for Internet traffic 174 engineering and to include references to the latest relevant work in 175 the IETF. 177 1.1. What is Internet Traffic Engineering? 179 The Internet exists in order to transfer information from source 180 nodes to destination nodes. Accordingly, one of the most significant 181 functions performed by the Internet is the routing of traffic from 182 ingress nodes to egress nodes. Therefore, one of the most 183 distinctive functions performed by Internet traffic engineering is 184 the control and optimization of the routing function, to steer 185 traffic through the network. 187 Internet traffic engineering is defined as that aspect of Internet 188 network engineering dealing with the issue of performance evaluation 189 and performance optimization of operational IP networks. Traffic 190 Engineering encompasses the application of technology and scientific 191 principles to the measurement, characterization, modeling, and 192 control of Internet traffic [RFC2702], [AWD2]. 194 Ultimately, it is the performance of the network as seen by end users 195 of network services that is truly paramount. The characteristics 196 visible to end users are the emergent properties of the network, 197 which are the characteristics of the network when viewed as a whole. 198 A central goal of the service provider, therefore, is to enhance the 199 emergent properties of the network while taking economic 200 considerations into account. This is accomplished by addressing 201 traffic oriented performance requirements, while utilizing network 202 resources economically and reliably. Traffic oriented performance 203 measures include delay, delay variation, packet loss, and throughput. 205 Internet traffic engineering responds at different temporal 206 resolution to network events. Certain aspects of capacity 207 management, such as capacity planning, respond at very coarse 208 temporal levels, ranging from days to possibly years. The 209 introduction of automatically switched optical transport networks 210 (e.g., based on GMPLS concepts, see Section 4.3.5) could 211 significantly reduce the lifecycle for capacity planning by 212 expediting provisioning of optical bandwidth. Routing control 213 functions operate at intermediate levels of temporal resolution, 214 ranging from milliseconds to days. Finally, the packet level 215 processing functions (e.g., rate shaping, queue management, and 216 scheduling) operate at very fine levels of temporal resolution, 217 ranging from picoseconds to milliseconds while responding to the 218 real-time statistical behavior of traffic. 220 Thus, the optimization aspects of traffic engineering can be viewed 221 from a control perspective and can be pro-active and/or reactive. In 222 the pro-active case, the traffic engineering control system takes 223 preventive action to obviate predicted unfavorable future network 224 states such as e.g. engineering a backup path. It may also take 225 perfective action to induce a more desirable state in the future. In 226 the reactive case, the control system responds correctively and 227 perhaps adaptively to events that have already transpired in the 228 network, such as routing after failure. 230 Another important objective of Internet traffic engineering is to 231 facilitate reliable network operations [RFC2702]. Reliable network 232 operations can be facilitated by providing mechanisms that enhance 233 network integrity and by embracing policies emphasizing network 234 survivability. This results in a minimization of the vulnerability 235 of the network to service outages arising from errors, faults, and 236 failures occurring within the infrastructure. 238 The optimization aspects of traffic engineering can be achieved 239 through capacity management and traffic management. As used in this 240 document, capacity management includes capacity planning, routing 241 control, and resource management. Network resources of particular 242 interest include link bandwidth, buffer space, and computational 243 resources. Likewise, as used in this document, traffic management 244 includes (1) nodal traffic control functions such as traffic 245 conditioning, queue management, scheduling, and (2) other functions 246 that regulate traffic flow through the network or that arbitrate 247 access to network resources between different packets or between 248 different traffic streams. 250 One major challenge of Internet traffic engineering is the 251 realization of automated control capabilities that adapt quickly and 252 cost effectively to significant changes in a network's state, while 253 still maintaining stability of the network. Results from performance 254 evaluation assessing the effectiveness of traffic engineering methods 255 can be used to identify existing problems, guide network re- 256 optimization, and aid in the prediction of potential future problems. 257 However this process can also time consuming and may not be suitable 258 to act on sudden, ephemeral changes in the network. 260 Performance evaluation can be achieved in many different ways. The 261 most notable techniques include analytical methods, simulation, and 262 empirical methods based on measurements. The processs can be quite 263 complicated in practical network contexts. For example, simplifying 264 concepts such as effective bandwidth and effective buffer [ELW95] may 265 be used to approximate nodal behaviors at the packet level and 266 simplify the analysis at the connection level. A set of concepts 267 known as network calculus [CRUZ] based on deterministic bounds may 268 simplify network analysis relative to classical stochastic 269 techniques. 271 In many cases, Internet traffic engineering is about finding the 272 least bad action to take to enhance the performance of the network. 273 For this, it is necessary to reliably predict the impact of potential 274 corrective actions in a networking context. Such a prediction often 275 relies on the accuracy of a simulation model identifying root cause 276 and duration of a bottleneck, as well as the effectiveness of 277 corrective actions and its side-effects over time. 279 In genaral, traffic engineering comes in two flavours. Either as a 280 background process that constantly monitors traffic and optimize the 281 usage of resources to improve performance, or in form of a pre- 282 planned optimized traffic distribution that is considered optimal. 283 In the later case, any deviation from the optimum distribution (e.g., 284 caused by a fiber cut) is reverted upon repair without further 285 optimization. However, this form of traffic engineering heavily 286 relies upon the notion that the planned state of the network is 287 indeed optimal. Hence, in such a mode there are two levels of 288 traffic engineering: the TE-planning task to enable an optimum 289 traffic distribution, and the routing task keeping traffic flows 290 attached to the pre-planned distribution 292 As a general rule, traffic engineering concepts and mechanisms must 293 be sufficiently specific and well defined to address known 294 requirements, but simultaneously flexible and extensible to 295 accommodate unforeseen future demands. Optimizing the wrong measures 296 may achieve certain local objectives, but may have disastrous 297 consequences on the emergent properties of the network. 299 1.2. Scope 301 The scope of this document is intra-domain traffic engineering; that 302 is, traffic engineering within a given autonomous system in the 303 Internet. This document will discuss concepts pertaining to intra- 304 domain traffic control, including such issues as routing control, 305 micro and macro resource allocation, and the control coordination 306 problems that arise consequently. 308 This document will describe and characterize techniques already in 309 use or in advanced development for Internet traffic engineering. The 310 way these techniques fit together will be discussed and scenarios in 311 which they are useful will be identified. 313 While this document considers various intra-domain traffic 314 engineering approaches, it focuses more on traffic engineering with 315 MPLS. Traffic engineering based upon manipulation of IGP metrics is 316 not addressed in detail. This topic may be addressed by other 317 working group document(s). 319 Although the emphasis is on intra-domain traffic engineering, in 320 Section 7, an overview of the high level considerations pertaining to 321 inter-domain traffic engineering will be provided. Inter-domain 322 Internet traffic engineering is crucial to the performance 323 enhancement of the global Internet infrastructure. 325 Whenever possible, relevant requirements from existing IETF documents 326 and other sources will be incorporated by reference. 328 1.3. Terminology 330 This subsection provides terminology which is useful for Internet 331 traffic engineering. The definitions presented apply to this 332 document. These terms may have other meanings elsewhere. 334 Baseline analysis A study conducted to serve as a baseline for 335 comparison to the actual behavior of the network. 337 Busy hour A one hour period within a specified interval of time 338 (typically 24 hours) in which the traffic load in a network or 339 sub-network is greatest. 341 Bottleneck A network element whose input traffic rate tends to be 342 greater than its output rate. 344 Congestion A state of a network resource in which the traffic 345 incident on the resource exceeds its output capacity over an 346 interval of time. 348 Congestion avoidance An approach to congestion management that 349 attempts to obviate the occurrence of congestion. 351 Congestion control An approach to congestion management that 352 attempts to remedy congestion problems that have already occurred. 354 Constraint-based routing A class of routing protocols that take 355 specified traffic attributes, network constraints, and policy 356 constraints into account when making routing decisions. 357 Constraint-based routing is applicable to traffic aggregates as 358 well as flows. It is a generalization of QoS routing. 360 Demand side congestion management A congestion management scheme 361 that addresses congestion problems by regulating or conditioning 362 offered load. 364 Effective bandwidth The minimum amount of bandwidth that can be 365 assigned to a flow or traffic aggregate in order to deliver 366 'acceptable service quality' to the flow or traffic aggregate. 368 Egress traffic Traffic exiting a network or network element. 370 Hot-spot A network element or subsystem which is in a state of 371 congestion. 373 Ingress traffic Traffic entering a network or network element. 375 Inter-domain traffic Traffic that originates in one Autonomous 376 system and terminates in another. 378 Loss network A network that does not provide adequate buffering for 379 traffic, so that traffic entering a busy resource within the 380 network will be dropped rather than queued. 382 Metric A parameter defined in terms of standard units of 383 measurement. 385 Measurement Methodology A repeatable measurement technique used to 386 derive one or more metrics of interest. 388 Network Survivability The capability to provide a prescribed level 389 of QoS for existing services after a given number of failures 390 occur within the network. 392 Offline traffic engineering A traffic engineering system that exists 393 outside of the network. 395 Online traffic engineering A traffic engineering system that exists 396 within the network, typically implemented on or as adjuncts to 397 operational network elements. 399 Performance measures Metrics that provide quantitative or 400 qualitative measures of the performance of systems or subsystems 401 of interest. 403 Performance management A systematic approach to improving 404 effectiveness in the accomplishment of specific networking goals 405 related to performance improvement. 407 Performance Metric A performance parameter defined in terms of 408 standard units of measurement. 410 Provisioning The process of assigning or configuring network 411 resources to meet certain requests. 413 QoS routing Class of routing systems that selects paths to be used 414 by a flow based on the QoS requirements of the flow. 416 Service Level Agreement A contract between a provider and a customer 417 that guarantees specific levels of performance and reliability at 418 a certain cost. 420 Stability An operational state in which a network does not oscillate 421 in a disruptive manner from one mode to another mode. 423 Supply side congestion management A congestion management scheme 424 that provisions additional network resources to address existing 425 and/or anticipated congestion problems. 427 Transit traffic Traffic whose origin and destination are both 428 outside of the network under consideration. 430 Traffic characteristic A description of the temporal behavior or a 431 description of the attributes of a given traffic flow or traffic 432 aggregate. 434 Traffic engineering system A collection of objects, mechanisms, and 435 protocols that are used conjunctively to accomplish traffic 436 engineering objectives. 438 Traffic flow A stream of packets between two end-points that can be 439 characterized in a certain way. A micro-flow has a more specific 440 definition A micro-flow is a stream of packets with the same 441 source and destination addresses, source and destination ports, 442 and protocol ID. 444 Traffic intensity A measure of traffic loading with respect to a 445 resource capacity over a specified period of time. In classical 446 telephony systems, traffic intensity is measured in units of 447 Erlang. 449 Traffic matrix A representation of the traffic demand between a set 450 of origin and destination abstract nodes. An abstract node can 451 consist of one or more network elements. 453 Traffic monitoring The process of observing traffic characteristics 454 at a given point in a network and collecting the traffic 455 information for analysis and further action. 457 Traffic trunk An aggregation of traffic flows belonging to the same 458 class which are forwarded through a common path. A traffic trunk 459 may be characterized by an ingress and egress node, and a set of 460 attributes which determine its behavioral characteristics and 461 requirements from the network. 463 2. Background 465 The Internet has quickly evolved into a very critical communications 466 infrastructure, supporting significant economic, educational, and 467 social activities. Simultaneously, the delivery of Internet 468 communications services has become very competitive and end-users are 469 demanding very high quality service from their service providers. 470 Consequently, performance optimization of large scale IP networks, 471 especially public Internet backbones, have become an important 472 problem. Network performance requirements are multi-dimensional, 473 complex, and sometimes contradictory; making the traffic engineering 474 problem very challenging. 476 The network must convey IP packets from ingress nodes to egress nodes 477 efficiently, expeditiously, and economically. Furthermore, in a 478 multiclass service environment (e.g., Diffserv capable networks), the 479 resource sharing parameters of the network must be appropriately 480 determined and configured according to prevailing policies and 481 service models to resolve resource contention issues arising from 482 mutual interference between packets traversing through the network. 483 Thus, consideration must be given to resolving competition for 484 network resources between traffic streams belonging to the same 485 service class (intra-class contention resolution) and traffic streams 486 belonging to different classes (inter-class contention resolution). 488 2.1. Context of Internet Traffic Engineering 490 The context of Internet traffic engineering pertains to the scenarios 491 where traffic engineering is used. A traffic engineering methodology 492 establishes appropriate rules to resolve traffic performance issues 493 occurring in a specific context. The context of Internet traffic 494 engineering includes: 496 1. A network context defining the universe of discourse, and in 497 particular the situations in which the traffic engineering 498 problems occur. The network context includes network structure, 499 network policies, network characteristics, network constraints, 500 network quality attributes, and network optimization criteria. 502 2. A problem context defining the general and concrete issues that 503 traffic engineering addresses. The problem context includes 504 identification, abstraction of relevant features, representation, 505 formulation, specification of the requirements on the solution 506 space, and specification of the desirable features of acceptable 507 solutions. 509 3. A solution context suggesting how to address the issues 510 identified by the problem context. The solution context includes 511 analysis, evaluation of alternatives, prescription, and 512 resolution. 514 4. An implementation and operational context in which the solutions 515 are methodologically instantiated. The implementation and 516 operational context includes planning, organization, and 517 execution. 519 The context of Internet traffic engineering and the different problem 520 scenarios are discussed in the following subsections. 522 2.2. Network Context 524 IP networks range in size from small clusters of routers situated 525 within a given location, to thousands of interconnected routers, 526 switches, and other components distributed all over the world. 528 Conceptually, at the most basic level of abstraction, an IP network 529 can be represented as a distributed dynamical system consisting of: 530 (1) a set of interconnected resources which provide transport 531 services for IP traffic subject to certain constraints, (2) a demand 532 system representing the offered load to be transported through the 533 network, and (3) a response system consisting of network processes, 534 protocols, and related mechanisms which facilitate the movement of 535 traffic through the network (see also [AWD2]). 537 The network elements and resources may have specific characteristics 538 restricting the manner in which the demand is handled. Additionally, 539 network resources may be equipped with traffic control mechanisms 540 superintending the way in which the demand is serviced. Traffic 541 control mechanisms may, for example, be used to control various 542 packet processing activities within a given resource, arbitrate 543 contention for access to the resource by different packets, and 544 regulate traffic behavior through the resource. A configuration 545 management and provisioning system may allow the settings of the 546 traffic control mechanisms to be manipulated by external or internal 547 entities in order to exercise control over the way in which the 548 network elements respond to internal and external stimuli. 550 The details of how the network provides transport services for 551 packets are specified in the policies of the network administrators 552 and are installed through network configuration management and policy 553 based provisioning systems. Generally, the types of services 554 provided by the network also depends upon the technology and 555 characteristics of the network elements and protocols, the prevailing 556 service and utility models, and the ability of the network 557 administrators to translate policies into network configurations. 559 Contemporary Internet networks have three significant 560 characteristics: (1) they provide real-time services, (2) they have 561 become mission critical, and (3) their operating environments are 562 very dynamic. The dynamic characteristics of IP networks can be 563 attributed in part to fluctuations in demand, to the interaction 564 between various network protocols and processes, to the rapid 565 evolution of the infrastructure which demands the constant inclusion 566 of new technologies and new network elements, and to transient and 567 persistent impairments which occur within the system. 569 Packets contend for the use of network resources as they are conveyed 570 through the network. A network resource is considered to be 571 congested if the arrival rate of packets exceed the output capacity 572 of the resource over an interval of time. Congestion may result in 573 some of the arrival packets being delayed or even dropped. 575 Congestion increases transit delays, delay variation, packet loss, 576 and reduces the predictability of network services. Clearly, 577 congestion is a highly undesirable phenomenon. 579 Combating congestion at a reasonable cost is a major objective of 580 Internet traffic engineering. 582 Efficient sharing of network resources by multiple traffic streams is 583 a basic economic premise for packet switched networks in general and 584 for the Internet in particular. A fundamental challenge in network 585 operation, especially in a large scale public IP network, is to 586 increase the efficiency of resource utilization while minimizing the 587 possibility of congestion. 589 Increasingly, the Internet will have to function in the presence of 590 different classes of traffic with different service requirements. 591 The advent of Differentiated Services [RFC2475] makes this 592 requirement particularly acute. Thus, packets may be grouped into 593 behavior aggregates such that each behavior aggregate may have a 594 common set of behavioral characteristics or a common set of delivery 595 requirements. In practice, the delivery requirements of a specific 596 set of packets may be specified explicitly or implicitly. Two of the 597 most important traffic delivery requirements are capacity constraints 598 and QoS constraints. 600 Capacity constraints can be expressed statistically as peak rates, 601 mean rates, burst sizes, or as some deterministic notion of effective 602 bandwidth. QoS requirements can be expressed in terms of (1) 603 integrity constraints such as packet loss and (2) in terms of 604 temporal constraints such as timing restrictions for the delivery of 605 each packet (delay) and timing restrictions for the delivery of 606 consecutive packets belonging to the same traffic stream (delay 607 variation). 609 2.3. Problem Context 611 Fundamental problems exist in association with the operation of a 612 network described by the simple model of the previous subsection. 613 This subsection reviews the problem context in relation to the 614 traffic engineering function. 616 The identification, abstraction, representation, and measurement of 617 network features relevant to traffic engineering is a significant 618 issue. 620 One particularly important class of problems concerns how to 621 explicitly formulate the problems that traffic engineering attempts 622 to solve, how to identify the requirements on the solution space, how 623 to specify the desirable features of good solutions, how to actually 624 solve the problems, and how to measure and characterize the 625 effectiveness of the solutions. 627 Another class of problems concerns how to measure and estimate 628 relevant network state parameters. Effective traffic engineering 629 relies on a good estimate of the offered traffic load as well as a 630 view of the underlying topology and associated resource constraints. 631 A network-wide view of the topology is also a must for offline 632 planning. 634 Still another class of problems concerns how to characterize the 635 state of the network and how to evaluate its performance under a 636 variety of scenarios. The performance evaluation problem is two- 637 fold. One aspect of this problem relates to the evaluation of the 638 system level performance of the network. The other aspect relates to 639 the evaluation of the resource level performance, which restricts 640 attention to the performance analysis of individual network 641 resources. In this memo, we refer to the system level 642 characteristics of the network as the "macro-states" and the resource 643 level characteristics as the "micro-states." The system level 644 characteristics are also known as the emergent properties of the 645 network as noted earlier. Correspondingly, we shall refer to the 646 traffic engineering schemes dealing with network performance 647 optimization at the systems level as "macro-TE" and the schemes that 648 optimize at the individual resource level as "micro-TE." Under 649 certain circumstances, the system level performance can be derived 650 from the resource level performance using appropriate rules of 651 composition, depending upon the particular performance measures of 652 interest. 654 Another fundamental class of problems concerns how to effectively 655 optimize network performance. Performance optimization may entail 656 translating solutions to specific traffic engineering problems into 657 network configurations. Optimization may also entail some degree of 658 resource management control, routing control, and/or capacity 659 augmentation. 661 As noted previously, congestion is an undesirable phenomena in 662 operational networks. Therefore, the next subsection addresses the 663 issue of congestion and its ramifications within the problem context 664 of Internet traffic engineering. 666 2.3.1. Congestion and its Ramifications 668 Congestion is one of the most significant problems in an operational 669 IP context. A network element is said to be congested if it 670 experiences sustained overload over an interval of time. Congestion 671 almost always results in degradation of service quality to end users. 672 Congestion control schemes can include demand side policies and 673 supply side policies. Demand side policies may restrict access to 674 congested resources and/or dynamically regulate the demand to 675 alleviate the overload situation. Supply side policies may expand or 676 augment network capacity to better accommodate offered traffic. 677 Supply side policies may also re-allocate network resources by 678 redistributing traffic over the infrastructure. Traffic 679 redistribution and resource re-allocation serve to increase the 680 'effective capacity' seen by the demand. 682 The emphasis of this memo is primarily on congestion management 683 schemes falling within the scope of the network, rather than on 684 congestion management systems dependent upon sensitivity and 685 adaptivity from end-systems. That is, the aspects that are 686 considered in this memo with respect to congestion management are 687 those solutions that can be provided by control entities operating on 688 the network and by the actions of network administrators and network 689 operations systems. 691 2.4. Solution Context 693 The solution context for Internet traffic engineering involves 694 analysis, evaluation of alternatives, and choice between alternative 695 courses of action. Generally the solution context is predicated on 696 making reasonable inferences about the current or future state of the 697 network, and subsequently making appropriate decisions that may 698 involve a preference between alternative sets of action. More 699 specifically, the solution context demands reasonable estimates of 700 traffic workload, characterization of network state, deriving 701 solutions to traffic engineering problems which may be implicitly or 702 explicitly formulated, and possibly instantiating a set of control 703 actions. Control actions may involve the manipulation of parameters 704 associated with routing, control over tactical capacity acquisition, 705 and control over the traffic management functions. 707 The following list of instruments may be applicable to the solution 708 context of Internet traffic engineering. 710 1. A set of policies, objectives, and requirements (which may be 711 context dependent) for network performance evaluation and 712 performance optimization. 714 2. A collection of online and possibly offline tools and mechanisms 715 for measurement, characterization, modeling, and control of 716 Internet traffic and control over the placement and allocation of 717 network resources, as well as control over the mapping or 718 distribution of traffic onto the infrastructure. 720 3. A set of constraints on the operating environment, the network 721 protocols, and the traffic engineering system itself. 723 4. A set of quantitative and qualitative techniques and 724 methodologies for abstracting, formulating, and solving traffic 725 engineering problems. 727 5. A set of administrative control parameters which may be 728 manipulated through a Configuration Management (CM) system. The 729 CM system itself may include a configuration control subsystem, a 730 configuration repository, a configuration accounting subsystem, 731 and a configuration auditing subsystem. 733 6. A set of guidelines for network performance evaluation, 734 performance optimization, and performance improvement. 736 Derivation of traffic characteristics through measurement and/or 737 estimation is very useful within the realm of the solution space for 738 traffic engineering. Traffic estimates can be derived from customer 739 subscription information, traffic projections, traffic models, and 740 from actual empirical measurements. The empirical measurements may 741 be performed at the traffic aggregate level or at the flow level in 742 order to derive traffic statistics at various levels of detail. 743 Measurements at the flow level or on small traffic aggregates may be 744 performed at edge nodes, where traffic enters and leaves the network. 745 Measurements at large traffic aggregate levels may be performed 746 within the core of the network where potentially numerous traffic 747 flows may be in transit concurrently. 749 To conduct performance studies and to support planning of existing 750 and future networks, a routing analysis may be performed to determine 751 the path(s) the routing protocols will choose for various traffic 752 demands, and to ascertain the utilization of network resources as 753 traffic is routed through the network. The routing analysis should 754 capture the selection of paths through the network, the assignment of 755 traffic across multiple feasible routes, and the multiplexing of IP 756 traffic over traffic trunks (if such constructs exists) and over the 757 underlying network infrastructure. A network topology model is a 758 necessity for routing analysis. A network topology model may be 759 extracted from network architecture documents, from network designs, 760 from information contained in router configuration files, from 761 routing databases, from routing tables, or from automated tools that 762 discover and depict network topology information. Topology 763 information may also be derived from servers that monitor network 764 state, and from servers that perform provisioning functions. 766 Routing in operational IP networks can be administratively controlled 767 at various levels of abstraction including the manipulation of BGP 768 attributes and manipulation of IGP metrics. For path oriented 769 technologies such as MPLS, routing can be further controlled by the 770 manipulation of relevant traffic engineering parameters, resource 771 parameters, and administrative policy constraints. Within the 772 context of MPLS, the path of an explicit label switched path (LSP) 773 can be computed and established in various ways including: (1) 774 manually, (2) automatically online using constraint-based routing 775 processes implemented on label switching routers, and (3) 776 automatically offline using constraint-based routing entities 777 implemented on external traffic engineering support systems. 779 2.4.1. Combating the Congestion Problem 781 Minimizing congestion is a significant aspect of Internet traffic 782 engineering. This subsection gives an overview of the general 783 approaches that have been used or proposed to combat congestion 784 problems. 786 Congestion management policies can be categorized based upon the 787 following criteria (see e.g., [YARE95] for a more detailed taxonomy 788 of congestion control schemes): (1) Response time scale which can be 789 characterized as long, medium, or short; (2) reactive versus 790 preventive which relates to congestion control and congestion 791 avoidance; and (3) supply side versus demand side congestion 792 management schemes. These aspects are discussed in the following 793 paragraphs. 795 1. Congestion Management based on Response Time Scales 797 * Long (weeks to months): Capacity planning works over a 798 relatively long time scale to expand network capacity based on 799 estimates or forecasts of future traffic demand and traffic 800 distribution. Since router and link provisioning take time 801 and are generally expensive, these upgrades are typically 802 carried out in the weeks-to-months or even years time scale. 804 * Medium (minutes to days): Several control policies fall within 805 the medium time scale category. Examples include: (1) 806 Adjusting IGP and/or BGP parameters to route traffic away or 807 towards certain segments of the network; (2) Setting up and/or 808 adjusting some explicitly routed label switched paths (ER- 809 LSPs) in MPLS networks to route some traffic trunks away from 810 possibly congested resources or towards possibly more 811 favorable routes; (3) re-configuring the logical topology of 812 the network to make it correlate more closely with the spatial 813 traffic distribution using for example some underlying path- 814 oriented technology such as MPLS LSPs, ATM PVCs, or optical 815 channel trails. Many of these adaptive medium time scale 816 response schemes rely on a measurement system that monitors 817 changes in traffic distribution, traffic shifts, and network 818 resource utilization and subsequently provides feedback to the 819 online and/or offline traffic engineering mechanisms and tools 820 which employ this feedback information to trigger certain 821 control actions to occur within the network. The traffic 822 engineering mechanisms and tools can be implemented in a 823 distributed fashion or in a centralized fashion, and may have 824 a hierarchical structure or a flat structure. The comparative 825 merits of distributed and centralized control structures for 826 networks are well known. A centralized scheme may have global 827 visibility into the network state and may produce potentially 828 more optimal solutions. However, centralized schemes are 829 prone to single points of failure and may not scale as well as 830 distributed schemes. Moreover, the information utilized by a 831 centralized scheme may be stale and may not reflect the actual 832 state of the network. It is not an objective of this memo to 833 make a recommendation between distributed and centralized 834 schemes. This is a choice that network administrators must 835 make based on their specific needs. 837 * Short (picoseconds to minutes): This category includes packet 838 level processing functions and events on the order of several 839 round trip times. It includes router mechanisms such as 840 passive and active buffer management. These mechanisms are 841 used to control congestion and/or signal congestion to end 842 systems so that they can adaptively regulate the rate at which 843 traffic is injected into the network. One of the most popular 844 active queue management schemes, especially for TCP traffic, 845 is Random Early Detection (RED) [FLJA93], which supports 846 congestion avoidance by controlling the average queue size. 847 During congestion (but before the queue is filled), the RED 848 scheme chooses arriving packets to "mark" according to a 849 probabilistic algorithm which takes into account the average 850 queue size. For a router that does not utilize explicit 851 congestion notification (ECN) see e.g., [FLOY94], the marked 852 packets can simply be dropped to signal the inception of 853 congestion to end systems. On the other hand, if the router 854 supports ECN, then it can set the ECN field in the packet 855 header. Several variations of RED have been proposed to 856 support different drop precedence levels in multi-class 857 environments [RFC2597], e.g., RED with In and Out (RIO) and 858 Weighted RED. There is general consensus that RED provides 859 congestion avoidance performance which is not worse than 860 traditional Tail-Drop (TD) queue management (drop arriving 861 packets only when the queue is full). Importantly, however, 862 RED reduces the possibility of global synchronization and 863 improves fairness among different TCP sessions. However, RED 864 by itself can not prevent congestion and unfairness caused by 865 sources unresponsive to RED, e.g., UDP traffic and some 866 misbehaved greedy connections. Other schemes have been 867 proposed to improve the performance and fairness in the 868 presence of unresponsive traffic. Some of these schemes were 869 proposed as theoretical frameworks and are typically not 870 available in existing commercial products. Two such schemes 871 are Longest Queue Drop (LQD) and Dynamic Soft Partitioning 872 with Random Drop (RND) [SLDC98]. 874 2. Congestion Management: Reactive versus Preventive Schemes 876 * Reactive: reactive (recovery) congestion management policies 877 react to existing congestion problems to improve it. All the 878 policies described in the long and medium time scales above 879 can be categorized as being reactive especially if the 880 policies are based on monitoring and identifying existing 881 congestion problems, and on the initiation of relevant actions 882 to ease a situation. 884 * Preventive: preventive (predictive/avoidance) policies take 885 proactive action to prevent congestion based on estimates and 886 predictions of future potential congestion problems. Some of 887 the policies described in the long and medium time scales fall 888 into this category. They do not necessarily respond 889 immediately to existing congestion problems. Instead 890 forecasts of traffic demand and workload distribution are 891 considered and action may be taken to prevent potential 892 congestion problems in the future. The schemes described in 893 the short time scale (e.g., RED and its variations, ECN, LQD, 894 and RND) are also used for congestion avoidance since dropping 895 or marking packets before queues actually overflow would 896 trigger corresponding TCP sources to slow down. 898 3. Congestion Management: Supply Side versus Demand Side Schemes 900 * Supply side: supply side congestion management policies 901 increase the effective capacity available to traffic in order 902 to control or obviate congestion. This can be accomplished by 903 augmenting capacity. Another way to accomplish this is to 904 minimize congestion by having a relatively balanced 905 distribution of traffic over the network. For example, 906 capacity planning should aim to provide a physical topology 907 and associated link bandwidths that match estimated traffic 908 workload and traffic distribution based on forecasting 909 (subject to budgetary and other constraints). However, if 910 actual traffic distribution does not match the topology 911 derived from capacity panning (due to forecasting errors or 912 facility constraints for example), then the traffic can be 913 mapped onto the existing topology using routing control 914 mechanisms, using path oriented technologies (e.g., MPLS LSPs 915 and optical channel trails) to modify the logical topology, or 916 by using some other load redistribution mechanisms. 918 * Demand side: demand side congestion management policies 919 control or regulate the offered traffic to alleviate 920 congestion problems. For example, some of the short time 921 scale mechanisms described earlier (such as RED and its 922 variations, ECN, LQD, and RND) as well as policing and rate 923 shaping mechanisms attempt to regulate the offered load in 924 various ways. Tariffs may also be applied as a demand side 925 instrument. To date, however, tariffs have not been used as a 926 means of demand side congestion management within the 927 Internet. 929 In summary, a variety of mechanisms can be used to address congestion 930 problems in IP networks. These mechanisms may operate at multiple 931 time-scales. 933 2.5. Implementation and Operational Context 935 The operational context of Internet traffic engineering is 936 characterized by constant change which occur at multiple levels of 937 abstraction. The implementation context demands effective planning, 938 organization, and execution. The planning aspects may involve 939 determining prior sets of actions to achieve desired objectives. 940 Organizing involves arranging and assigning responsibility to the 941 various components of the traffic engineering system and coordinating 942 the activities to accomplish the desired TE objectives. Execution 943 involves measuring and applying corrective or perfective actions to 944 attain and maintain desired TE goals. 946 2.6. High-Level Objectives 948 The high-level objectives for Internet traffic engineering include: 949 usability, automation, scalability, stability, visibility, 950 simplicity, efficiency, reliability, correctness, maintainability, 951 extensibility, interoperability, and security. In a given context, 952 some of these recommendations may be critical while others may be 953 optional. Therefore, prioritization may be required during the 954 development phase of a traffic engineering system (or components 955 thereof) to tailor it to a specific operational context. 957 In the following paragraphs, some of the aspects of the high-level 958 objectives for Internet traffic engineering are summarized. 960 Usability: Usability is a human factor aspect of traffic engineering 961 systems. Usability refers to the ease with which a traffic 962 engineering system can be deployed and operated. In general, it is 963 desirable to have a TE system that can be readily deployed in an 964 existing network. It is also desirable to have a TE system that is 965 easy to operate and maintain. 967 Automation: Whenever feasible, a traffic engineering system should 968 automate as many traffic engineering functions as possible to 969 minimize the amount of human effort needed to control and analyze 970 operational networks. Automation is particularly imperative in large 971 scale public networks because of the high cost of the human aspects 972 of network operations and the high risk of network problems caused by 973 human errors. Automation may entail the incorporation of automatic 974 feedback and intelligence into some components of the traffic 975 engineering system. 977 Scalability: Contemporary public networks are growing very fast with 978 respect to network size and traffic volume. Therefore, a TE system 979 should be scalable to remain applicable as the network evolves. In 980 particular, a TE system should remain functional as the network 981 expands with regard to the number of routers and links, and with 982 respect to the traffic volume. A TE system should have a scalable 983 architecture, should not adversely impair other functions and 984 processes in a network element, and should not consume too much 985 network resources when collecting and distributing state information 986 or when exerting control. 988 Stability: Stability is a very important consideration in traffic 989 engineering systems that respond to changes in the state of the 990 network. State-dependent traffic engineering methodologies typically 991 mandate a tradeoff between responsiveness and stability. It is 992 strongly recommended that when tradeoffs are warranted between 993 responsiveness and stability, that the tradeoff should be made in 994 favor of stability (especially in public IP backbone networks). 996 Flexibility: A TE system should be flexible to allow for changes in 997 optimization policy. In particular, a TE system should provide 998 sufficient configuration options so that a network administrator can 999 tailor the TE system to a particular environment. It may also be 1000 desirable to have both online and offline TE subsystems which can be 1001 independently enabled and disabled. TE systems that are used in 1002 multi-class networks should also have options to support class based 1003 performance evaluation and optimization. 1005 Visibility: As part of the TE system, mechanisms should exist to 1006 collect statistics from the network and to analyze these statistics 1007 to determine how well the network is functioning. Derived statistics 1008 such as traffic matrices, link utilization, latency, packet loss, and 1009 other performance measures of interest which are determined from 1010 network measurements can be used as indicators of prevailing network 1011 conditions. Other examples of status information which should be 1012 observed include existing functional routing information 1013 (additionally, in the context of MPLS existing LSP routes), etc. 1015 Simplicity: Generally, a TE system should be as simple as possible. 1016 More importantly, the TE system should be relatively easy to use 1017 (i.e., clean, convenient, and intuitive user interfaces). Simplicity 1018 in user interface does not necessarily imply that the TE system will 1019 use naive algorithms. When complex algorithms and internal 1020 structures are used, such complexities should be hidden as much as 1021 possible from the network administrator through the user interface. 1023 Interoperability: Whenever feasible, traffic engineering systems and 1024 their components should be developed with open standards based 1025 interfaces to allow interoperation with other systems and components. 1027 Security: Security is a critical consideration in traffic engineering 1028 systems. Such traffic engineering systems typically exert control 1029 over certain functional aspects of the network to achieve the desired 1030 performance objectives. Therefore, adequate measures must be taken 1031 to safeguard the integrity of the traffic engineering system. 1032 Adequate measures must also be taken to protect the network from 1033 vulnerabilities that originate from security breaches and other 1034 impairments within the traffic engineering system. 1036 The remainder of this section will focus on some of the high level 1037 functional recommendations for traffic engineering. 1039 3. Traffic Engineering Process Models 1041 This section describes a generic process model that captures the high 1042 level practical aspects of Internet traffic engineering in an 1043 operational context. The process model is described as a sequence of 1044 actions that a traffic engineer, or more generally a traffic 1045 engineering system, must perform to optimize the performance of an 1046 operational network (see also [RFC2702], [AWD2]). The process model 1047 described here represents the broad activities common to most traffic 1048 engineering methodologies although the details regarding how traffic 1049 engineering is executed may differ from network to network. This 1050 process model may be enacted explicitly or implicitly, by an 1051 automaton and/or by a human. 1053 The traffic engineering process model is iterative [AWD2]. The four 1054 phases of the process model described below are repeated continually. 1056 The first phase of the TE process model is to define the relevant 1057 control policies that govern the operation of the network. These 1058 policies may depend upon many factors including the prevailing 1059 business model, the network cost structure, the operating 1060 constraints, the utility model, and optimization criteria. 1062 The second phase of the process model is a feedback mechanism 1063 involving the acquisition of measurement data from the operational 1064 network. If empirical data is not readily available from the 1065 network, then synthetic workloads may be used instead which reflect 1066 either the prevailing or the expected workload of the network. 1067 Synthetic workloads may be derived by estimation or extrapolation 1068 using prior empirical data. Their derivation may also be obtained 1069 using mathematical models of traffic characteristics or other means. 1071 The third phase of the process model is to analyze the network state 1072 and to characterize traffic workload. Performance analysis may be 1073 proactive and/or reactive. Proactive performance analysis identifies 1074 potential problems that do not exist, but could manifest in the 1075 future. Reactive performance analysis identifies existing problems, 1076 determines their cause through diagnosis, and evaluates alternative 1077 approaches to remedy the problem, if necessary. A number of 1078 quantitative and qualitative techniques may be used in the analysis 1079 process, including modeling based analysis and simulation. The 1080 analysis phase of the process model may involve investigating the 1081 concentration and distribution of traffic across the network or 1082 relevant subsets of the network, identifying the characteristics of 1083 the offered traffic workload, identifying existing or potential 1084 bottlenecks, and identifying network pathologies such as ineffective 1085 link placement, single points of failures, etc. Network pathologies 1086 may result from many factors including inferior network architecture, 1087 inferior network design, and configuration problems. A traffic 1088 matrix may be constructed as part of the analysis process. Network 1089 analysis may also be descriptive or prescriptive. 1091 The fourth phase of the TE process model is the performance 1092 optimization of the network. The performance optimization phase 1093 involves a decision process which selects and implements a set of 1094 actions from a set of alternatives. Optimization actions may include 1095 the use of appropriate techniques to either control the offered 1096 traffic or to control the distribution of traffic across the network. 1097 Optimization actions may also involve adding additional links or 1098 increasing link capacity, deploying additional hardware such as 1099 routers and switches, systematically adjusting parameters associated 1100 with routing such as IGP metrics and BGP attributes, and adjusting 1101 traffic management parameters. Network performance optimization may 1102 also involve starting a network planning process to improve the 1103 network architecture, network design, network capacity, network 1104 technology, and the configuration of network elements to accommodate 1105 current and future growth. 1107 3.1. Components of the Traffic Engineering Process Model 1109 The key components of the traffic engineering process model include a 1110 measurement subsystem, a modeling and analysis subsystem, and an 1111 optimization subsystem. The following subsections examine these 1112 components as they apply to the traffic engineering process model. 1114 3.2. Measurement 1116 Measurement is crucial to the traffic engineering function. The 1117 operational state of a network can be conclusively determined only 1118 through measurement. Measurement is also critical to the 1119 optimization function because it provides feedback data which is used 1120 by traffic engineering control subsystems. This data is used to 1121 adaptively optimize network performance in response to events and 1122 stimuli originating within and outside the network. Measurement is 1123 also needed to determine the quality of network services and to 1124 evaluate the effectiveness of traffic engineering policies. 1125 Experience suggests that measurement is most effective when acquired 1126 and applied systematically. 1128 When developing a measurement system to support the traffic 1129 engineering function in IP networks, the following questions should 1130 be carefully considered: Why is measurement needed in this particular 1131 context? What parameters are to be measured? How should the 1132 measurement be accomplished? Where should the measurement be 1133 performed? When should the measurement be performed? How frequently 1134 should the monitored variables be measured? What level of 1135 measurement accuracy and reliability is desirable? What level of 1136 measurement accuracy and reliability is realistically attainable? To 1137 what extent can the measurement system permissibly interfere with the 1138 monitored network components and variables? What is the acceptable 1139 cost of measurement? The answers to these questions will determine 1140 the measurement tools and methodologies appropriate in any given 1141 traffic engineering context. 1143 It should also be noted that there is a distinction between 1144 measurement and evaluation. Measurement provides raw data concerning 1145 state parameters and variables of monitored network elements. 1146 Evaluation utilizes the raw data to make inferences regarding the 1147 monitored system. 1149 Measurement in support of the TE function can occur at different 1150 levels of abstraction. For example, measurement can be used to 1151 derive packet level characteristics, flow level characteristics, user 1152 or customer level characteristics, traffic aggregate characteristics, 1153 component level characteristics, and network wide characteristics. 1155 3.3. Modeling, Analysis, and Simulation 1157 Modeling and analysis are important aspects of Internet traffic 1158 engineering. Modeling involves constructing an abstract or physical 1159 representation which depicts relevant traffic characteristics and 1160 network attributes. 1162 A network model is an abstract representation of the network which 1163 captures relevant network features, attributes, and characteristics, 1164 such as link and nodal attributes and constraints. A network model 1165 may facilitate analysis and/or simulation which can be used to 1166 predict network performance under various conditions as well as to 1167 guide network expansion plans. 1169 In general, Internet traffic engineering models can be classified as 1170 either structural or behavioral. Structural models focus on the 1171 organization of the network and its components. Behavioral models 1172 focus on the dynamics of the network and the traffic workload. 1173 Modeling for Internet traffic engineering may also be formal or 1174 informal. 1176 Accurate behavioral models for traffic sources are particularly 1177 useful for analysis. Development of behavioral traffic source models 1178 that are consistent with empirical data obtained from operational 1179 networks is a major research topic in Internet traffic engineering. 1180 These source models should also be tractable and amenable to 1181 analysis. The topic of source models for IP traffic is a research 1182 topic and is therefore outside the scope of this document. Its 1183 importance, however, must be emphasized. 1185 Network simulation tools are extremely useful for traffic 1186 engineering. Because of the complexity of realistic quantitative 1187 analysis of network behavior, certain aspects of network performance 1188 studies can only be conducted effectively using simulation. A good 1189 network simulator can be used to mimic and visualize network 1190 characteristics under various conditions in a safe and non-disruptive 1191 manner. For example, a network simulator may be used to depict 1192 congested resources and hot spots, and to provide hints regarding 1193 possible solutions to network performance problems. A good simulator 1194 may also be used to validate the effectiveness of planned solutions 1195 to network issues without the need to tamper with the operational 1196 network, or to commence an expensive network upgrade which may not 1197 achieve the desired objectives. Furthermore, during the process of 1198 network planning, a network simulator may reveal pathologies such as 1199 single points of failure which may require additional redundancy, and 1200 potential bottlenecks and hot spots which may require additional 1201 capacity. 1203 Routing simulators are especially useful in large networks. A 1204 routing simulator may identify planned links which may not actually 1205 be used to route traffic by the existing routing protocols. 1206 Simulators can also be used to conduct scenario based and 1207 perturbation based analysis, as well as sensitivity studies. 1208 Simulation results can be used to initiate appropriate actions in 1209 various ways. For example, an important application of network 1210 simulation tools is to investigate and identify how best to make the 1211 network evolve and grow, in order to accommodate projected future 1212 demands. 1214 3.4. Optimization 1216 Network performance optimization involves resolving network issues by 1217 transforming such issues into concepts that enable a solution, 1218 identification of a solution, and implementation of the solution. 1219 Network performance optimization can be corrective or perfective. In 1220 corrective optimization, the goal is to remedy a problem that has 1221 occurred or that is incipient. In perfective optimization, the goal 1222 is to improve network performance even when explicit problems do not 1223 exist and are not anticipated. 1225 Network performance optimization is a continual process, as noted 1226 previously. Performance optimization iterations may consist of real- 1227 time optimization sub-processes and non-real-time network planning 1228 sub-processes. The difference between real-time optimization and 1229 network planning is primarily in the relative time- scale in which 1230 they operate and in the granularity of actions. One of the 1231 objectives of a real-time optimization sub-process is to control the 1232 mapping and distribution of traffic over the existing network 1233 infrastructure to avoid and/or relieve congestion, to assure 1234 satisfactory service delivery, and to optimize resource utilization. 1235 Real-time optimization is needed because random incidents such as 1236 fiber cuts or shifts in traffic demand will occur irrespective of how 1237 well a network is designed. These incidents can cause congestion and 1238 other problems to manifest in an operational network. Real-time 1239 optimization must solve such problems in small to medium time-scales 1240 ranging from micro-seconds to minutes or hours. Examples of real- 1241 time optimization include queue management, IGP/BGP metric tuning, 1242 and using technologies such as MPLS explicit LSPs to change the paths 1243 of some traffic trunks [XIAO]. 1245 One of the functions of the network planning sub-process is to 1246 initiate actions to systematically evolve the architecture, 1247 technology, topology, and capacity of a network. When a problem 1248 exists in the network, real-time optimization should provide an 1249 immediate remedy. Because a prompt response is necessary, the real- 1250 time solution may not be the best possible solution. Network 1251 planning may subsequently be needed to refine the solution and 1252 improve the situation. Network planning is also required to expand 1253 the network to support traffic growth and changes in traffic 1254 distribution over time. As previously noted, a change in the 1255 topology and/or capacity of the network may be the outcome of network 1256 planning. 1258 Clearly, network planning and real-time performance optimization are 1259 mutually complementary activities. A well-planned and designed 1260 network makes real-time optimization easier, while a systematic 1261 approach to real-time network performance optimization allows network 1262 planning to focus on long term issues rather than tactical 1263 considerations. Systematic real-time network performance 1264 optimization also provides valuable inputs and insights toward 1265 network planning. 1267 Stability is an important consideration in real-time network 1268 performance optimization. This aspect will be repeatedly addressed 1269 throughout this memo. 1271 4. Review of TE Techniques 1273 This section briefly reviews different traffic engineering approaches 1274 proposed and implemented in telecommunications and computer networks. 1275 The discussion is not intended to be comprehensive. It is primarily 1276 intended to illuminate pre-existing perspectives and prior art 1277 concerning traffic engineering in the Internet and in legacy 1278 telecommunications networks. 1280 4.1. Historic Overview 1282 4.1.1. Traffic Engineering in Classical Telephone Networks 1284 This subsection presents a brief overview of traffic engineering in 1285 telephone networks which often relates to the way user traffic is 1286 steered from an originating node to the terminating node. This 1287 subsection presents a brief overview of this topic. A detailed 1288 description of the various routing strategies applied in telephone 1289 networks is included in the book by G. Ash [ASH2]. 1291 The early telephone network relied on static hierarchical routing, 1292 whereby routing patterns remained fixed independent of the state of 1293 the network or time of day. The hierarchy was intended to 1294 accommodate overflow traffic, improve network reliability via 1295 alternate routes, and prevent call looping by employing strict 1296 hierarchical rules. The network was typically over-provisioned since 1297 a given fixed route had to be dimensioned so that it could carry user 1298 traffic during a busy hour of any busy day. Hierarchical routing in 1299 the telephony network was found to be too rigid upon the advent of 1300 digital switches and stored program control which were able to manage 1301 more complicated traffic engineering rules. 1303 Dynamic routing was introduced to alleviate the routing inflexibility 1304 in the static hierarchical routing so that the network would operate 1305 more efficiently. This resulted in significant economic gains 1306 [HUSS87]. Dynamic routing typically reduces the overall loss 1307 probability by 10 to 20 percent (compared to static hierarchical 1308 routing). Dynamic routing can also improve network resilience by 1309 recalculating routes on a per-call basis and periodically updating 1310 routes. 1312 There are three main types of dynamic routing in the telephone 1313 network. They are time-dependent routing, state-dependent routing 1314 (SDR), and event dependent routing (EDR). 1316 In time-dependent routing, regular variations in traffic loads (such 1317 as time of day or day of week) are exploited in pre-planned routing 1318 tables. In state-dependent routing, routing tables are updated 1319 online according to the current state of the network (e.g., traffic 1320 demand, utilization, etc.). In event dependent routing, routing 1321 changes are incepted by events (such as call setups encountering 1322 congested or blocked links) whereupon new paths are searched out 1323 using learning models. EDR methods are real-time adaptive, but they 1324 do not require global state information as does SDR. Examples of EDR 1325 schemes include the dynamic alternate routing (DAR) from BT, the 1326 state-and-time dependent routing (STR) from NTT, and the success-to- 1327 the-top (STT) routing from AT&T. 1329 Dynamic non-hierarchical routing (DNHR) is an example of dynamic 1330 routing that was introduced in the AT&T toll network in the 1980's to 1331 respond to time-dependent information such as regular load variations 1332 as a function of time. Time-dependent information in terms of load 1333 may be divided into three time scales: hourly, weekly, and yearly. 1334 Correspondingly, three algorithms are defined to pre-plan the routing 1335 tables. The network design algorithm operates over a year-long 1336 interval while the demand servicing algorithm operates on a weekly 1337 basis to fine tune link sizes and routing tables to correct forecast 1338 errors on the yearly basis. At the smallest time scale, the routing 1339 algorithm is used to make limited adjustments based on daily traffic 1340 variations. Network design and demand servicing are computed using 1341 offline calculations. Typically, the calculations require extensive 1342 searches on possible routes. On the other hand, routing may need 1343 online calculations to handle crankback. DNHR adopts a "two-link" 1344 approach whereby a path can consist of two links at most. The 1345 routing algorithm presents an ordered list of route choices between 1346 an originating switch and a terminating switch. If a call overflows, 1347 a via switch (a tandem exchange between the originating switch and 1348 the terminating switch) would send a crankback signal to the 1349 originating switch. This switch would then select the next route, 1350 and so on, until there are no alternative routes available in which 1351 the call is blocked. 1353 4.1.2. Evolution of Traffic Engineering in Packet Networks 1355 This subsection reviews related prior work that was intended to 1356 improve the performance of data networks. Indeed, optimization of 1357 the performance of data networks started in the early days of the 1358 ARPANET. Other early commercial networks such as SNA also recognized 1359 the importance of performance optimization and service 1360 differentiation. 1362 In terms of traffic management, the Internet has been a best effort 1363 service environment until recently. In particular, very limited 1364 traffic management capabilities existed in IP networks to provide 1365 differentiated queue management and scheduling services to packets 1366 belonging to different classes. 1368 In terms of routing control, the Internet has employed distributed 1369 protocols for intra-domain routing. These protocols are highly 1370 scalable and resilient. However, they are based on simple algorithms 1371 for path selection which have very limited functionality to allow 1372 flexible control of the path selection process. 1374 In the following subsections, the evolution of practical traffic 1375 engineering mechanisms in IP networks and its predecessors are 1376 reviewed. 1378 4.1.2.1. Adaptive Routing in the ARPANET 1380 The early ARPANET recognized the importance of adaptive routing where 1381 routing decisions were based on the current state of the network 1382 [MCQ80]. Early minimum delay routing approaches forwarded each 1383 packet to its destination along a path for which the total estimated 1384 transit time was the smallest. Each node maintained a table of 1385 network delays, representing the estimated delay that a packet would 1386 experience along a given path toward its destination. The minimum 1387 delay table was periodically transmitted by a node to its neighbors. 1388 The shortest path, in terms of hop count, was also propagated to give 1389 the connectivity information. 1391 One drawback to this approach is that dynamic link metrics tend to 1392 create "traffic magnets" causing congestion to be shifted from one 1393 location of a network to another location, resulting in oscillation 1394 and network instability. 1396 4.1.2.2. Dynamic Routing in the Internet 1398 The Internet evolved from the ARPANET and adopted dynamic routing 1399 algorithms with distributed control to determine the paths that 1400 packets should take en-route to their destinations. The routing 1401 algorithms are adaptations of shortest path algorithms where costs 1402 are based on link metrics. The link metric can be based on static or 1403 dynamic quantities. The link metric based on static quantities may 1404 be assigned administratively according to local criteria. The link 1405 metric based on dynamic quantities may be a function of a network 1406 congestion measure such as delay or packet loss. 1408 It was apparent early that static link metric assignment was 1409 inadequate because it can easily lead to unfavorable scenarios in 1410 which some links become congested while others remain lightly loaded. 1411 One of the many reasons for the inadequacy of static link metrics is 1412 that link metric assignment was often done without considering the 1413 traffic matrix in the network. Also, the routing protocols did not 1414 take traffic attributes and capacity constraints into account when 1415 making routing decisions. This results in traffic concentration 1416 being localized in subsets of the network infrastructure and 1417 potentially causing congestion. Even if link metrics are assigned in 1418 accordance with the traffic matrix, unbalanced loads in the network 1419 can still occur due to a number factors including: 1421 o Resources may not be deployed in the most optimal locations from a 1422 routing perspective. 1424 o Forecasting errors in traffic volume and/or traffic distribution. 1426 o Dynamics in traffic matrix due to the temporal nature of traffic 1427 patterns, BGP policy change from peers, etc. 1429 The inadequacy of the legacy Internet interior gateway routing system 1430 is one of the factors motivating the interest in path oriented 1431 technology with explicit routing and constraint-based routing 1432 capability such as MPLS. 1434 4.1.2.3. ToS Routing 1436 Type-of-Service (ToS) routing involves different routes going to the 1437 same destination with selection dependent upon the ToS field of an IP 1438 packet [RFC2474]. The ToS classes may be classified as low delay and 1439 high throughput. Each link is associated with multiple link costs 1440 and each link cost is used to compute routes for a particular ToS. A 1441 separate shortest path tree is computed for each ToS. The shortest 1442 path algorithm must be run for each ToS resulting in very expensive 1443 computation. Classical ToS-based routing is now outdated as the IP 1444 header field has been replaced by a Diffserv field. Effective 1445 traffic engineering is difficult to perform in classical ToS-based 1446 routing because each class still relies exclusively on shortest path 1447 routing which results in localization of traffic concentration within 1448 the network. 1450 4.1.2.4. Equal Cost Multi-Path 1452 Equal Cost Multi-Path (ECMP) is another technique that attempts to 1453 address the deficiency in the Shortest Path First (SPF) interior 1454 gateway routing systems [RFC2328]. In the classical SPF algorithm, 1455 if two or more shortest paths exist to a given destination, the 1456 algorithm will choose one of them. The algorithm is modified 1457 slightly in ECMP so that if two or more equal cost shortest paths 1458 exist between two nodes, the traffic between the nodes is distributed 1459 among the multiple equal-cost paths. Traffic distribution across the 1460 equal-cost paths is usually performed in one of two ways: (1) packet- 1461 based in a round-robin fashion, or (2) flow-based using hashing on 1462 source and destination IP addresses and possibly other fields of the 1463 IP header. The first approach can easily cause out- of-order packets 1464 while the second approach is dependent upon the number and 1465 distribution of flows. Flow-based load sharing may be unpredictable 1466 in an enterprise network where the number of flows is relatively 1467 small and less heterogeneous (for example, hashing may not be 1468 uniform), but it is generally effective in core public networks where 1469 the number of flows is large and heterogeneous. 1471 In ECMP, link costs are static and bandwidth constraints are not 1472 considered, so ECMP attempts to distribute the traffic as equally as 1473 possible among the equal-cost paths independent of the congestion 1474 status of each path. As a result, given two equal-cost paths, it is 1475 possible that one of the paths will be more congested than the other. 1476 Another drawback of ECMP is that load sharing cannot be achieved on 1477 multiple paths which have non-identical costs. 1479 4.1.2.5. Nimrod 1481 Nimrod was a routing system developed to provide heterogeneous 1482 service specific routing in the Internet, while taking multiple 1483 constraints into account [RFC1992]. Essentially, Nimrod was a link 1484 state routing protocol to support path oriented packet forwarding. 1485 It used the concept of maps to represent network connectivity and 1486 services at multiple levels of abstraction. Mechanisms allowed 1487 restriction of the distribution of routing information. 1489 Even though Nimrod did not enjoy deployment in the public Internet, a 1490 number of key concepts incorporated into the Nimrod architecture, 1491 such as explicit routing which allows selection of paths at 1492 originating nodes, are beginning to find applications in some recent 1493 constraint-based routing initiatives. 1495 4.2. Development of Internet Traffic Engineering 1497 4.2.1. Overlay Model 1499 In the overlay model, a virtual-circuit network, such as ATM, frame 1500 relay, or WDM, provides virtual-circuit connectivity between routers 1501 that are located at the edges of a virtual-circuit cloud. In this 1502 mode, two routers that are connected through a virtual circuit see a 1503 direct adjacency between themselves independent of the physical route 1504 taken by the virtual circuit through the ATM, frame relay, or WDM 1505 network. Thus, the overlay model essentially decouples the logical 1506 topology that routers see from the physical topology that the ATM, 1507 frame relay, or WDM network manages. The overlay model based on ATM 1508 or frame relay enables a network administrator or an automaton to 1509 employ traffic engineering concepts to perform path optimization by 1510 re-configuring or rearranging the virtual circuits so that a virtual 1511 circuit on a congested or sub-optimal physical link can be re-routed 1512 to a less congested or more optimal one. In the overlay model, 1513 traffic engineering is also employed to establish relationships 1514 between the traffic management parameters (e.g., PCR, SCR, and MBS 1515 for ATM) of the virtual-circuit technology and the actual traffic 1516 that traverses each circuit. These relationships can be established 1517 based upon known or projected traffic profiles, and some other 1518 factors. 1520 The overlay model using IP over ATM requires the management of two 1521 separate networks with different technologies (IP and ATM) resulting 1522 in increased operational complexity and cost. In the fully-meshed 1523 overlay model, each router would peer to every other router in the 1524 network, so that the total number of adjacencies is a quadratic 1525 function of the number of routers. Some of the issues with the 1526 overlay model are discussed in [AWD2]. 1528 4.2.2. Constraint-Based Routing 1530 Constraint-based routing refers to a class of routing systems that 1531 compute routes through a network subject to the satisfaction of a set 1532 of constraints and requirements. In the most general setting, 1533 constraint-based routing may also seek to optimize overall network 1534 performance while minimizing costs. 1536 The constraints and requirements may be imposed by the network itself 1537 or by administrative policies. Constraints may include bandwidth, 1538 hop count, delay, and policy instruments such as resource class 1539 attributes. Constraints may also include domain specific attributes 1540 of certain network technologies and contexts which impose 1541 restrictions on the solution space of the routing function. Path 1542 oriented technologies such as MPLS have made constraint-based routing 1543 feasible and attractive in public IP networks. 1545 The concept of constraint-based routing within the context of MPLS 1546 traffic engineering requirements in IP networks was first described 1547 in [RFC2702] and led to developments such as MPLS-TE [RFC3209] as 1548 described in Section 4.3.4. 1550 Unlike QoS routing (for example, see [RFC2386] and [MA]) which 1551 generally addresses the issue of routing individual traffic flows to 1552 satisfy prescribed flow based QoS requirements subject to network 1553 resource availability, constraint-based routing is applicable to 1554 traffic aggregates as well as flows and may be subject to a wide 1555 variety of constraints which may include policy restrictions. 1557 4.3. Overview of IETF Projects Related to Traffic Engineering 1559 This subsection reviews a number of IETF activities pertinent to 1560 Internet traffic engineering. These activities are primarily 1561 intended to evolve the IP architecture to support new service 1562 definitions which allow preferential or differentiated treatment to 1563 be accorded to certain types of traffic. 1565 4.3.1. Integrated Services 1567 The IETF Integrated Services working group developed the integrated 1568 services (Intserv) model. This model requires resources, such as 1569 bandwidth and buffers, to be reserved a priori for a given traffic 1570 flow to ensure that the quality of service requested by the traffic 1571 flow is satisfied. The integrated services model includes additional 1572 components beyond those used in the best-effort model such as packet 1573 classifiers, packet schedulers, and admission control. A packet 1574 classifier is used to identify flows that are to receive a certain 1575 level of service. A packet scheduler handles the scheduling of 1576 service to different packet flows to ensure that QoS commitments are 1577 met. Admission control is used to determine whether a router has the 1578 necessary resources to accept a new flow. 1580 The main issue with the Integrated Services model has been 1581 scalability [RFC2998], especially in large public IP networks which 1582 may potentially have millions of active micro-flows in transit 1583 concurrently. 1585 A notable feature of the Integrated Services model is that it 1586 requires explicit signaling of QoS requirements from end systems to 1587 routers [RFC2753]. The Resource Reservation Protocol (RSVP) performs 1588 this signaling function and is a critical component of the Integrated 1589 Services model. RSVP is described next. 1591 4.3.2. RSVP 1593 RSVP is a soft state signaling protocol [RFC2205]. It supports 1594 receiver initiated establishment of resource reservations for both 1595 multicast and unicast flows. RSVP was originally developed as a 1596 signaling protocol within the integrated services framework for 1597 applications to communicate QoS requirements to the network and for 1598 the network to reserve relevant resources to satisfy the QoS 1599 requirements [RFC2205]. 1601 Under RSVP, the sender or source node sends a PATH message to the 1602 receiver with the same source and destination addresses as the 1603 traffic which the sender will generate. The PATH message contains: 1604 (1) a sender Tspec specifying the characteristics of the traffic, (2) 1605 a sender Template specifying the format of the traffic, and (3) an 1606 optional Adspec which is used to support the concept of one pass with 1607 advertising (OPWA) [RFC2205]. Every intermediate router along the 1608 path forwards the PATH Message to the next hop determined by the 1609 routing protocol. Upon receiving a PATH Message, the receiver 1610 responds with a RESV message which includes a flow descriptor used to 1611 request resource reservations. The RESV message travels to the 1612 sender or source node in the opposite direction along the path that 1613 the PATH message traversed. Every intermediate router along the path 1614 can reject or accept the reservation request of the RESV message. If 1615 the request is rejected, the rejecting router will send an error 1616 message to the receiver and the signaling process will terminate. If 1617 the request is accepted, link bandwidth and buffer space are 1618 allocated for the flow and the related flow state information is 1619 installed in the router. 1621 One of the issues with the original RSVP specification was 1622 Scalability. This is because reservations were required for micro- 1623 flows, so that the amount of state maintained by network elements 1624 tends to increase linearly with the number of micro-flows. These 1625 issues are described in [RFC2961]. 1627 Recently, RSVP has been modified and extended in several ways to 1628 mitigate the scaling problems. As a result, it is becoming a 1629 versatile signaling protocol for the Internet. For example, RSVP has 1630 been extended to reserve resources for aggregation of flows, to set 1631 up MPLS explicit label switched paths, and to perform other signaling 1632 functions within the Internet. There are also a number of proposals 1633 to reduce the amount of refresh messages required to maintain 1634 established RSVP sessions [RFC2961]. 1636 A number of IETF working groups have been engaged in activities 1637 related to the RSVP protocol. These include the original RSVP 1638 working group, the MPLS working group, the Resource Allocation 1639 Protocol working group, and the Policy Framework working group. 1641 4.3.3. Differentiated Services 1643 The goal of the Differentiated Services (Diffserv) effort within the 1644 IETF is to devise scalable mechanisms for categorization of traffic 1645 into behavior aggregates, which ultimately allows each behavior 1646 aggregate to be treated differently, especially when there is a 1647 shortage of resources such as link bandwidth and buffer space 1648 [RFC2475]. One of the primary motivations for the Diffserv effort 1649 was to devise alternative mechanisms for service differentiation in 1650 the Internet that mitigate the scalability issues encountered with 1651 the Intserv model. 1653 The IETF Diffserv working group has defined a Differentiated Services 1654 field in the IP header (DS field). The DS field consists of six bits 1655 of the part of the IP header formerly known as TOS octet. The DS 1656 field is used to indicate the forwarding treatment that a packet 1657 should receive at a node [RFC2474]. The Diffserv working group has 1658 also standardized a number of Per-Hop Behavior (PHB) groups. Using 1659 the PHBs, several classes of services can be defined using different 1660 classification, policing, shaping, and scheduling rules. 1662 For an end-user of network services to receive Differentiated 1663 Services from its Internet Service Provider (ISP), it may be 1664 necessary for the user to have a Service Level Agreement (SLA) with 1665 the ISP. An SLA may explicitly or implicitly specify a Traffic 1666 Conditioning Agreement (TCA) which defines classifier rules as well 1667 as metering, marking, discarding, and shaping rules. 1669 Packets are classified, and possibly policed and shaped at the 1670 ingress to a Diffserv network. When a packet traverses the boundary 1671 between different Diffserv domains, the DS field of the packet may be 1672 re-marked according to existing agreements between the domains. 1674 Differentiated Services allows only a finite number of service 1675 classes to be indicated by the DS field. The main advantage of the 1676 Diffserv approach relative to the Intserv model is scalability. 1677 Resources are allocated on a per-class basis and the amount of state 1678 information is proportional to the number of classes rather than to 1679 the number of application flows. 1681 It should be obvious from the previous discussion that the Diffserv 1682 model essentially deals with traffic management issues on a per hop 1683 basis. The Diffserv control model consists of a collection of micro- 1684 TE control mechanisms. Other traffic engineering capabilities, such 1685 as capacity management (including routing control), are also required 1686 in order to deliver acceptable service quality in Diffserv networks. 1687 The concept of Per Domain Behaviors has been introduced to better 1688 capture the notion of differentiated services across a complete 1689 domain [RFC3086]. 1691 4.3.4. MPLS 1693 MPLS is an advanced forwarding scheme which also includes extensions 1694 to conventional IP control plane protocols. MPLS extends the 1695 Internet routing model and enhances packet forwarding and path 1696 control [RFC3031]. 1698 At the ingress to an MPLS domain, Label Switching Routers (LSRs) 1699 classify IP packets into forwarding equivalence classes (FECs) based 1700 on a variety of factors, including, e.g., a combination of the 1701 information carried in the IP header of the packets and the local 1702 routing information maintained by the LSRs. An MPLS label stack 1703 entry is then prepended to each packet according to their forwarding 1704 equivalence classes. The MPLS label stack entry is 32 bits long and 1705 contains a 20-bit label field. 1707 An LSR makes forwarding decisions by using the label prepended to 1708 packets as the index into a local next hop label forwarding entry 1709 (NHLFE). The packet is then processed as specified in the NHLFE. 1711 The incoming label may be replaced by an outgoing label (label swap), 1712 and the packet may be forwarded to the next LSR. Before a packet 1713 leaves an MPLS domain, its MPLS label may be removed (label pop). A 1714 Label Switched Path (LSP) is the path between an ingress LSRs and an 1715 egress LSRs through which a labeled packet traverses. The path of an 1716 explicit LSP is defined at the originating (ingress) node of the LSP. 1717 MPLS can use a signaling protocol such as RSVP or LDP to set up LSPs. 1719 MPLS is a very powerful technology for Internet traffic engineering 1720 because it supports explicit LSPs which allow constraint-based 1721 routing to be implemented efficiently in IP networks [AWD2]. The 1722 requirements for traffic engineering over MPLS are described in 1723 [RFC2702]. Extensions to RSVP to support instantiation of explicit 1724 LSP are discussed in [RFC3209]. 1726 4.3.5. Generalized MPLS 1728 GMPLS extends MPLS to encompass time-division (e.g., SONET/SDH, PDH, 1729 G.709), wavelength (lambdas), and spatial switching (e.g., incoming 1730 port or fiber to outgoing port or fiber) as well as continuing to 1731 support packet switching. GMPLS provides a common control plane for 1732 all of these layers each of which has a diverse data or forwarding 1733 plane. GMPLS covers both the signaling and the routing part of that 1734 control plane and is based on the Traffic Engineering extensions to 1735 MPLS (see Section 4.3.4). 1737 In GMPLS, the original MPLS architecture is extended to include LSRs 1738 whose forwarding planes recognize neither packet, nor cell 1739 boundaries, and therefore, cannot forward data based on the 1740 information carried in either packet or cell headers. Specifically, 1741 such LSRs include devices where the switching decision is based on 1742 time slots, wavelengths, or physical ports. These additions impact 1743 basic LSP properties: how labels are requested and communicated, the 1744 unidirectional nature of LSPs, how errors are propagated, and 1745 information provided for synchronizing the ingress and egress LSRs. 1747 4.3.6. IP Performance Metrics 1749 The IETF IP Performance Metrics (IPPM) working group has been 1750 developing a set of standard metrics that can be used to monitor the 1751 quality, performance, and reliability of Internet services. These 1752 metrics can be applied by network operators, end-users, and 1753 independent testing groups to provide users and service providers 1754 with a common understanding of the performance and reliability of the 1755 Internet component 'clouds' they use/provide [RFC2330]. The criteria 1756 for performance metrics developed by the IPPM WG are described in 1757 [RFC2330]. Examples of performance metrics include one-way packet 1758 loss [RFC7680], one-way delay [RFC7679], and connectivity measures 1759 between two nodes [RFC2678]. Other metrics include second-order 1760 measures of packet loss and delay. 1762 Some of the performance metrics specified by the IPPM WG are useful 1763 for specifying Service Level Agreements (SLAs). SLAs are sets of 1764 service level objectives negotiated between users and service 1765 providers, wherein each objective is a combination of one or more 1766 performance metrics, possibly subject to certain constraints. 1768 4.3.7. Flow Measurement 1770 The IETF Real Time Flow Measurement (RTFM) working group has produced 1771 an architecture document defining a method to specify traffic flows 1772 as well as a number of components for flow measurement (meters, meter 1773 readers, manager) [RFC2722]. A flow measurement system enables 1774 network traffic flows to be measured and analyzed at the flow level 1775 for a variety of purposes. As noted in RFC 2722, a flow measurement 1776 system can be very useful in the following contexts: (1) 1777 understanding the behavior of existing networks, (2) planning for 1778 network development and expansion, (3) quantification of network 1779 performance, (4) verifying the quality of network service, and (5) 1780 attribution of network usage to users. 1782 A flow measurement system consists of meters, meter readers, and 1783 managers. A meter observes packets passing through a measurement 1784 point, classifies them into certain groups, accumulates certain usage 1785 data (such as the number of packets and bytes for each group), and 1786 stores the usage data in a flow table. A group may represent a user 1787 application, a host, a network, a group of networks, etc. A meter 1788 reader gathers usage data from various meters so it can be made 1789 available for analysis. A manager is responsible for configuring and 1790 controlling meters and meter readers. The instructions received by a 1791 meter from a manager include flow specification, meter control 1792 parameters, and sampling techniques. The instructions received by a 1793 meter reader from a manager include the address of the meter whose 1794 date is to be collected, the frequency of data collection, and the 1795 types of flows to be collected. 1797 4.3.8. Endpoint Congestion Management 1799 [RFC3124] is intended to provide a set of congestion control 1800 mechanisms that transport protocols can use. It is also intended to 1801 develop mechanisms for unifying congestion control across a subset of 1802 an endpoint's active unicast connections (called a congestion group). 1803 A congestion manager continuously monitors the state of the path for 1804 each congestion group under its control. The manager uses that 1805 information to instruct a scheduler on how to partition bandwidth 1806 among the connections of that congestion group. 1808 4.3.9. TE Extensions to the IGPs 1810 TBD 1812 4.3.10. Link-State BGP 1814 In a number of environments, a component external to a network is 1815 called upon to perform computations based on the network topology and 1816 current state of the connections within the network, including 1817 traffic engineering information. This is information typically 1818 distributed by IGP routing protocols within the network (see 1819 Section 4.3.9. 1821 The Border Gateway Protocol (BGP) Section 7 is one of the essential 1822 routing protocols that glue the Internet together. BGP Link State 1823 (BGP-LS) [RFC7752] is a mechanism by which link-state and traffic 1824 engineering information can be collected from networks and shared 1825 with external components using the BGP routing protocol. The 1826 mechanism is applicable to physical and virtual IGP links, and is 1827 subject to policy control. 1829 Information collected by BGP-LS can be used to construct the Traffic 1830 Engineering Database (TED, see Section 4.3.16) for use by the Path 1831 Computation Element (PCE, see Section 4.3.11), or may be used by 1832 Application-Layer Traffic Optimization (ALTO) servers (see 1833 Section 4.3.12). 1835 4.3.11. Path Computation Element 1837 Constraint-based path computation is a fundamental building block for 1838 traffic engineering in MPLS and GMPLS networks. Path computation in 1839 large, multi-domain networks is complex and may require special 1840 computational components and cooperation between the elements in 1841 different domains. The Path Computation Element (PCE) [RFC4655] is 1842 an entity (component, application, or network node) that is capable 1843 of computing a network path or route based on a network graph and 1844 applying computational constraints. 1846 Thus, a PCE can provide a central component in a traffic engineering 1847 system operating on the Traffic Engineering Database (TED, see 1848 Section 4.3.16) with delegated responsibility for determining paths 1849 in MPLS, GMPLS, or Segment Routing networks. The PCE uses the Path 1850 Computation Element Communication Protocol (PCEP) [RFC5440] to 1851 communicate with Path Computation Clients (PCCs), such as MPLS LSRs, 1852 to answer their requests for computed paths or to instruct them to 1853 initiate new paths [RFC8281] and maintain state about paths already 1854 installed in the network [RFC8231] 1855 PCEs form key components of a number of traffic engineering systems, 1856 such as the Application of the Path Computation Element Architecture 1857 [RFC6805], Abstraction and Control of TE Networks (ACTN) 1858 Section 4.3.14, Centralized Network Control [RFC8283], and Software 1859 Defined Networking (SDN) Section 5.3.2 1861 4.3.12. Application-Layer Traffic Optimization 1863 TBD 1865 4.3.13. Segment Routing 1867 TBD 1869 4.3.14. Network Virtualization and Abstraction 1871 ACTN goes here : TBD 1873 4.3.15. Deterministic Networking 1875 TBD 1877 4.3.16. Network TE State Definition and Presentation 1879 The network states that are relevant to the traffic engineering need 1880 to be stored in the system and presented to the user. The Traffic 1881 Engineering Database (TED) is a collection of all TE information 1882 about all TE nodes and TE links in the network, which is an essential 1883 component of a TE system, such as MPLS-TE [RFC2702] and GMPLS 1884 [RFC3945]. In order to formally define the data in the TED and to 1885 present the data to the user with high usability, the data modeling 1886 language YANG [RFC7950] can be used as described in 1887 [I-D.ietf-teas-yang-te-topo]. 1889 4.3.17. System Management and Control Interfaces 1891 The traffic engineering control system needs to have a management 1892 interface that is human-friendly and a control interfaces that is 1893 programable for automation. The Network Configuration Protocol 1894 (NETCONF) [RFC6241] or the RESTCONF Protocol [RFC8040] provide 1895 programmable interfaces that are also human-friendly. These 1896 protocols use XML or JSON encoded messages. When message compactness 1897 or protocol bandwidth consumption needs to be optimized for the 1898 control interface, other protocols, such as Group Communication for 1899 the Constrained Application Protocol (CoAP) [RFC7390] or gRPC, are 1900 available, especially when the protocol messages are encoded in a 1901 binary format. Along with any of these protocols, the data modeling 1902 language YANG [RFC7950] can be used to formally and precisely define 1903 the interface data. 1905 The Path Computation Element Communication Protocol (PCEP) [RFC5440] 1906 is another protocol that has evolved to be an option for the TE 1907 system control interface. The messages of PCEP are TLV-based, not 1908 defined by a data modeling language such as YANG. 1910 4.4. Overview of ITU Activities Related to Traffic Engineering 1912 This section provides an overview of prior work within the ITU-T 1913 pertaining to traffic engineering in traditional telecommunications 1914 networks. 1916 ITU-T Recommendations E.600 [ITU-E600], E.701 [ITU-E701], and E.801 1917 [ITU-E801] address traffic engineering issues in traditional 1918 telecommunications networks. Recommendation E.600 provides a 1919 vocabulary for describing traffic engineering concepts, while E.701 1920 defines reference connections, Grade of Service (GOS), and traffic 1921 parameters for ISDN. Recommendation E.701 uses the concept of a 1922 reference connection to identify representative cases of different 1923 types of connections without describing the specifics of their actual 1924 realizations by different physical means. As defined in 1925 Recommendation E.600, "a connection is an association of resources 1926 providing means for communication between two or more devices in, or 1927 attached to, a telecommunication network." Also, E.600 defines "a 1928 resource as any set of physically or conceptually identifiable 1929 entities within a telecommunication network, the use of which can be 1930 unambiguously determined" [ITU-E600]. There can be different types 1931 of connections as the number and types of resources in a connection 1932 may vary. 1934 Typically, different network segments are involved in the path of a 1935 connection. For example, a connection may be local, national, or 1936 international. The purposes of reference connections are to clarify 1937 and specify traffic performance issues at various interfaces between 1938 different network domains. Each domain may consist of one or more 1939 service provider networks. 1941 Reference connections provide a basis to define grade of service 1942 (GoS) parameters related to traffic engineering within the ITU-T 1943 framework. As defined in E.600, "GoS refers to a number of traffic 1944 engineering variables which are used to provide a measure of the 1945 adequacy of a group of resources under specified conditions." These 1946 GoS variables may be probability of loss, dial tone, delay, etc. 1947 They are essential for network internal design and operation as well 1948 as for component performance specification. 1950 GoS is different from quality of service (QoS) in the ITU framework. 1951 QoS is the performance perceivable by a telecommunication service 1952 user and expresses the user's degree of satisfaction of the service. 1953 QoS parameters focus on performance aspects observable at the service 1954 access points and network interfaces, rather than their causes within 1955 the network. GoS, on the other hand, is a set of network oriented 1956 measures which characterize the adequacy of a group of resources 1957 under specified conditions. For a network to be effective in serving 1958 its users, the values of both GoS and QoS parameters must be related, 1959 with GoS parameters typically making a major contribution to the QoS. 1961 Recommendation E.600 stipulates that a set of GoS parameters must be 1962 selected and defined on an end-to-end basis for each major service 1963 category provided by a network to assist the network provider with 1964 improving efficiency and effectiveness of the network. Based on a 1965 selected set of reference connections, suitable target values are 1966 assigned to the selected GoS parameters under normal and high load 1967 conditions. These end-to-end GoS target values are then apportioned 1968 to individual resource components of the reference connections for 1969 dimensioning purposes. 1971 4.5. Content Distribution 1973 The Internet is dominated by client-server interactions, especially 1974 Web traffic (in the future, more sophisticated media servers may 1975 become dominant). The location and performance of major information 1976 servers has a significant impact on the traffic patterns within the 1977 Internet as well as on the perception of service quality by end 1978 users. 1980 A number of dynamic load balancing techniques have been devised to 1981 improve the performance of replicated information servers. These 1982 techniques can cause spatial traffic characteristics to become more 1983 dynamic in the Internet because information servers can be 1984 dynamically picked based upon the location of the clients, the 1985 location of the servers, the relative utilization of the servers, the 1986 relative performance of different networks, and the relative 1987 performance of different parts of a network. This process of 1988 assignment of distributed servers to clients is called Traffic 1989 Directing. It functions at the application layer. 1991 Traffic Directing schemes that allocate servers in multiple 1992 geographically dispersed locations to clients may require empirical 1993 network performance statistics to make more effective decisions. In 1994 the future, network measurement systems may need to provide this type 1995 of information. The exact parameters needed are not yet defined. 1997 When congestion exists in the network, Traffic Directing and Traffic 1998 Engineering systems should act in a coordinated manner. This topic 1999 is for further study. 2001 The issues related to location and replication of information 2002 servers, particularly web servers, are important for Internet traffic 2003 engineering because these servers contribute a substantial proportion 2004 of Internet traffic. 2006 5. Taxonomy of Traffic Engineering Systems 2008 This section presents a short taxonomy of traffic engineering 2009 systems. A taxonomy of traffic engineering systems can be 2010 constructed based on traffic engineering styles and views as listed 2011 below: 2013 o Time-dependent vs State-dependent vs Event-dependent 2015 o Offline vs Online 2017 o Centralized vs Distributed 2019 o Local vs Global Information 2021 o Prescriptive vs Descriptive 2023 o Open Loop vs Closed Loop 2025 o Tactical vs Strategic 2027 These classification systems are described in greater detail in the 2028 following subsections of this document. 2030 5.1. Time-Dependent Versus State-Dependent Versus Event Dependent 2032 Traffic engineering methodologies can be classified as time- 2033 dependent, or state-dependent, or event-dependent. All TE schemes 2034 are considered to be dynamic in this document. Static TE implies 2035 that no traffic engineering methodology or algorithm is being 2036 applied. 2038 In the time-dependent TE, historical information based on periodic 2039 variations in traffic, (such as time of day), is used to pre-program 2040 routing plans and other TE control mechanisms. Additionally, 2041 customer subscription or traffic projection may be used. Pre- 2042 programmed routing plans typically change on a relatively long time 2043 scale (e.g., diurnal). Time-dependent algorithms do not attempt to 2044 adapt to random variations in traffic or changing network conditions. 2046 An example of a time-dependent algorithm is a global centralized 2047 optimizer where the input to the system is a traffic matrix and 2048 multi-class QoS requirements as described [MR99]. 2050 State-dependent TE adapts the routing plans for packets based on the 2051 current state of the network. The current state of the network 2052 provides additional information on variations in actual traffic 2053 (i.e., perturbations from regular variations) that could not be 2054 predicted using historical information. Constraint-based routing is 2055 an example of state-dependent TE operating in a relatively long time 2056 scale. An example operating in a relatively short time scale is a 2057 load-balancing algorithm described in [MATE]. 2059 The state of the network can be based on parameters such as 2060 utilization, packet delay, packet loss, etc. These parameters can be 2061 obtained in several ways. For example, each router may flood these 2062 parameters periodically or by means of some kind of trigger to other 2063 routers. Another approach is for a particular router performing 2064 adaptive TE to send probe packets along a path to gather the state of 2065 that path. Still another approach is for a management system to 2066 gather relevant information from network elements. 2068 Expeditious and accurate gathering and distribution of state 2069 information is critical for adaptive TE due to the dynamic nature of 2070 network conditions. State-dependent algorithms may be applied to 2071 increase network efficiency and resilience. Time-dependent 2072 algorithms are more suitable for predictable traffic variations. On 2073 the other hand, state-dependent algorithms are more suitable for 2074 adapting to the prevailing network state. 2076 Event-dependent TE methods can also be used for TE path selection. 2077 Event-dependent TE methods are distinct from time-dependent and 2078 state-dependent TE methods in the manner in which paths are selected. 2079 These algorithms are adaptive and distributed in nature and typically 2080 use learning models to find good paths for TE in a network. While 2081 state-dependent TE models typically use available-link-bandwidth 2082 (ALB) flooding for TE path selection, event-dependent TE methods do 2083 not require ALB flooding. Rather, event-dependent TE methods 2084 typically search out capacity by learning models, as in the success- 2085 to-the-top (STT) method. ALB flooding can be resource intensive, 2086 since it requires link bandwidth to carry LSAs, processor capacity to 2087 process LSAs, and the overhead can limit area/autonomous system (AS) 2088 size. Modeling results suggest that event-dependent TE methods could 2089 lead to a reduction in ALB flooding overhead without loss of network 2090 throughput performance [I-D.ietf-tewg-qos-routing]. 2092 5.2. Offline Versus Online 2094 Traffic engineering requires the computation of routing plans. The 2095 computation may be performed offline or online. The computation can 2096 be done offline for scenarios where routing plans need not be 2097 executed in real-time. For example, routing plans computed from 2098 forecast information may be computed offline. Typically, offline 2099 computation is also used to perform extensive searches on multi- 2100 dimensional solution spaces. 2102 Online computation is required when the routing plans must adapt to 2103 changing network conditions as in state-dependent algorithms. Unlike 2104 offline computation (which can be computationally demanding), online 2105 computation is geared toward relative simple and fast calculations to 2106 select routes, fine-tune the allocations of resources, and perform 2107 load balancing. 2109 5.3. Centralized Versus Distributed 2111 Centralized control has a central authority which determines routing 2112 plans and perhaps other TE control parameters on behalf of each 2113 router. The central authority collects the network-state information 2114 from all routers periodically and returns the routing information to 2115 the routers. The routing update cycle is a critical parameter 2116 directly impacting the performance of the network being controlled. 2117 Centralized control may need high processing power and high bandwidth 2118 control channels. 2120 Distributed control determines route selection by each router 2121 autonomously based on the routers view of the state of the network. 2122 The network state information may be obtained by the router using a 2123 probing method or distributed by other routers on a periodic basis 2124 using link state advertisements. Network state information may also 2125 be disseminated under exceptional conditions. 2127 5.3.1. Hybrid Systems 2129 TBD 2131 5.3.2. Considerations for Software Defined Networking 2133 TBD 2135 5.4. Local Versus Global 2137 Traffic engineering algorithms may require local or global network- 2138 state information. 2140 Local information pertains to the state of a portion of the domain. 2141 Examples include the bandwidth and packet loss rate of a particular 2142 path. Local state information may be sufficient for certain 2143 instances of distributed-controlled TEs. 2145 Global information pertains to the state of the entire domain 2146 undergoing traffic engineering. Examples include a global traffic 2147 matrix and loading information on each link throughout the domain of 2148 interest. Global state information is typically required with 2149 centralized control. Distributed TE systems may also need global 2150 information in some cases. 2152 5.5. Prescriptive Versus Descriptive 2154 TE systems may also be classified as prescriptive or descriptive. 2156 Prescriptive traffic engineering evaluates alternatives and 2157 recommends a course of action. Prescriptive traffic engineering can 2158 be further categorized as either corrective or perfective. 2159 Corrective TE prescribes a course of action to address an existing or 2160 predicted anomaly. Perfective TE prescribes a course of action to 2161 evolve and improve network performance even when no anomalies are 2162 evident. 2164 Descriptive traffic engineering, on the other hand, characterizes the 2165 state of the network and assesses the impact of various policies 2166 without recommending any particular course of action. 2168 5.5.1. Intent-Based Networking 2170 TBD 2172 5.6. Open-Loop Versus Closed-Loop 2174 Open-loop traffic engineering control is where control action does 2175 not use feedback information from the current network state. The 2176 control action may use its own local information for accounting 2177 purposes, however. 2179 Closed-loop traffic engineering control is where control action 2180 utilizes feedback information from the network state. The feedback 2181 information may be in the form of historical information or current 2182 measurement. 2184 5.7. Tactical vs Strategic 2186 Tactical traffic engineering aims to address specific performance 2187 problems (such as hot-spots) that occur in the network from a 2188 tactical perspective, without consideration of overall strategic 2189 imperatives. Without proper planning and insights, tactical TE tends 2190 to be ad hoc in nature. 2192 Strategic traffic engineering approaches the TE problem from a more 2193 organized and systematic perspective, taking into consideration the 2194 immediate and longer term consequences of specific policies and 2195 actions. 2197 6. Objectives for Internet Traffic Engineering 2199 This section describes high-level objectives for traffic engineering 2200 in the Internet. These objectives are presented in general terms and 2201 some advice is given as to how to meet the objectives. 2203 Broadly speaking, these objectives can be categorized as either 2204 functional or non-functional. 2206 Functional objectives for Internet traffic engineering describe the 2207 functions that a traffic engineering system should perform. These 2208 functions are needed to realize traffic engineering objectives by 2209 addressing traffic engineering problems. 2211 Non-functional objectives for Internet traffic engineering relate to 2212 the quality attributes or state characteristics of a traffic 2213 engineering system. These objectives may contain conflicting 2214 assertions and may sometimes be difficult to quantify precisely. 2216 6.1. Routing 2218 Routing control is a significant aspect of Internet traffic 2219 engineering. Routing impacts many of the key performance measures 2220 associated with networks, such as throughput, delay, and utilization. 2221 Generally, it is very difficult to provide good service quality in a 2222 wide area network without effective routing control. A desirable 2223 routing system is one that takes traffic characteristics and network 2224 constraints into account during route selection while maintaining 2225 stability. 2227 Traditional shortest path first (SPF) interior gateway protocols are 2228 based on shortest path algorithms and have limited control 2229 capabilities for traffic engineering [RFC2702], [AWD2]. These 2230 limitations include : 2232 1. The well known issues with pure SPF protocols, which do not take 2233 network constraints and traffic characteristics into account 2234 during route selection. For example, since IGPs always use the 2235 shortest paths (based on administratively assigned link metrics) 2236 to forward traffic, load sharing cannot be accomplished among 2237 paths of different costs. Using shortest paths to forward 2238 traffic conserves network resources, but may cause the following 2239 problems: 1) If traffic from a source to a destination exceeds 2240 the capacity of a link along the shortest path, the link (hence 2241 the shortest path) becomes congested while a longer path between 2242 these two nodes may be under-utilized; 2) the shortest paths from 2243 different sources can overlap at some links. If the total 2244 traffic from the sources exceeds the capacity of any of these 2245 links, congestion will occur. Problems can also occur because 2246 traffic demand changes over time but network topology and routing 2247 configuration cannot be changed as rapidly. This causes the 2248 network topology and routing configuration to become sub-optimal 2249 over time, which may result in persistent congestion problems. 2251 2. The Equal-Cost Multi-Path (ECMP) capability of SPF IGPs supports 2252 sharing of traffic among equal cost paths between two nodes. 2253 However, ECMP attempts to divide the traffic as equally as 2254 possible among the equal cost shortest paths. Generally, ECMP 2255 does not support configurable load sharing ratios among equal 2256 cost paths. The result is that one of the paths may carry 2257 significantly more traffic than other paths because it may also 2258 carry traffic from other sources. This situation can result in 2259 congestion along the path that carries more traffic. 2261 3. Modifying IGP metrics to control traffic routing tends to have 2262 network-wide effect. Consequently, undesirable and unanticipated 2263 traffic shifts can be triggered as a result. Recent work 2264 described in Section 8 may be capable of better control [FT00], 2265 [FT01]. 2267 Because of these limitations, new capabilities are needed to enhance 2268 the routing function in IP networks. Some of these capabilities have 2269 been described elsewhere and are summarized below. 2271 Constraint-based routing is desirable to evolve the routing 2272 architecture of IP networks, especially public IP backbones with 2273 complex topologies [RFC2702]. Constraint-based routing computes 2274 routes to fulfill requirements subject to constraints. Constraints 2275 may include bandwidth, hop count, delay, and administrative policy 2276 instruments such as resource class attributes [RFC2702], [RFC2386]. 2277 This makes it possible to select routes that satisfy a given set of 2278 requirements subject to network and administrative policy 2279 constraints. Routes computed through constraint-based routing are 2280 not necessarily the shortest paths. Constraint-based routing works 2281 best with path oriented technologies that support explicit routing, 2282 such as MPLS. 2284 Constraint-based routing can also be used as a way to redistribute 2285 traffic onto the infrastructure (even for best effort traffic). For 2286 example, if the bandwidth requirements for path selection and 2287 reservable bandwidth attributes of network links are appropriately 2288 defined and configured, then congestion problems caused by uneven 2289 traffic distribution may be avoided or reduced. In this way, the 2290 performance and efficiency of the network can be improved. 2292 A number of enhancements are needed to conventional link state IGPs, 2293 such as OSPF and IS-IS, to allow them to distribute additional state 2294 information required for constraint-based routing. These extensions 2295 to OSPF were described in [RFC3630] and to IS-IS in [RFC5305]. 2296 Essentially, these enhancements require the propagation of additional 2297 information in link state advertisements. Specifically, in addition 2298 to normal link-state information, an enhanced IGP is required to 2299 propagate topology state information needed for constraint-based 2300 routing. Some of the additional topology state information include 2301 link attributes such as reservable bandwidth and link resource class 2302 attribute (an administratively specified property of the link). The 2303 resource class attribute concept was defined in [RFC2702]. The 2304 additional topology state information is carried in new TLVs and sub- 2305 TLVs in IS-IS, or in the Opaque LSA in OSPF [RFC5305], [RFC3630]. 2307 An enhanced link-state IGP may flood information more frequently than 2308 a normal IGP. This is because even without changes in topology, 2309 changes in reservable bandwidth or link affinity can trigger the 2310 enhanced IGP to initiate flooding. A tradeoff is typically required 2311 between the timeliness of the information flooded and the flooding 2312 frequency to avoid excessive consumption of link bandwidth and 2313 computational resources, and more importantly, to avoid instability. 2315 In a TE system, it is also desirable for the routing subsystem to 2316 make the load splitting ratio among multiple paths (with equal cost 2317 or different cost) configurable. This capability gives network 2318 administrators more flexibility in the control of traffic 2319 distribution across the network. It can be very useful for avoiding/ 2320 relieving congestion in certain situations. Examples can be found in 2321 [XIAO]. 2323 The routing system should also have the capability to control the 2324 routes of subsets of traffic without affecting the routes of other 2325 traffic if sufficient resources exist for this purpose. This 2326 capability allows a more refined control over the distribution of 2327 traffic across the network. For example, the ability to move traffic 2328 from a source to a destination away from its original path to another 2329 path (without affecting other traffic paths) allows traffic to be 2330 moved from resource-poor network segments to resource-rich segments. 2331 Path oriented technologies such as MPLS inherently support this 2332 capability as discussed in [AWD2]. 2334 Additionally, the routing subsystem should be able to select 2335 different paths for different classes of traffic (or for different 2336 traffic behavior aggregates) if the network supports multiple classes 2337 of service (different behavior aggregates). 2339 6.2. Traffic Mapping 2341 Traffic mapping pertains to the assignment of traffic workload onto 2342 pre-established paths to meet certain requirements. Thus, while 2343 constraint-based routing deals with path selection, traffic mapping 2344 deals with the assignment of traffic to established paths which may 2345 have been selected by constraint-based routing or by some other 2346 means. Traffic mapping can be performed by time-dependent or state- 2347 dependent mechanisms, as described in Section 5.1. 2349 An important aspect of the traffic mapping function is the ability to 2350 establish multiple paths between an originating node and a 2351 destination node, and the capability to distribute the traffic 2352 between the two nodes across the paths according to some policies. A 2353 pre-condition for this scheme is the existence of flexible mechanisms 2354 to partition traffic and then assign the traffic partitions onto the 2355 parallel paths. This requirement was noted in [RFC2702]. When 2356 traffic is assigned to multiple parallel paths, it is recommended 2357 that special care should be taken to ensure proper ordering of 2358 packets belonging to the same application (or micro-flow) at the 2359 destination node of the parallel paths. 2361 As a general rule, mechanisms that perform the traffic mapping 2362 functions should aim to map the traffic onto the network 2363 infrastructure to minimize congestion. If the total traffic load 2364 cannot be accommodated, or if the routing and mapping functions 2365 cannot react fast enough to changing traffic conditions, then a 2366 traffic mapping system may rely on short time scale congestion 2367 control mechanisms (such as queue management, scheduling, etc.) to 2368 mitigate congestion. Thus, mechanisms that perform the traffic 2369 mapping functions should complement existing congestion control 2370 mechanisms. In an operational network, it is generally desirable to 2371 map the traffic onto the infrastructure such that intra-class and 2372 inter-class resource contention are minimized. 2374 When traffic mapping techniques that depend on dynamic state feedback 2375 (e.g., MATE and such like) are used, special care must be taken to 2376 guarantee network stability. 2378 6.3. Measurement 2380 The importance of measurement in traffic engineering has been 2381 discussed throughout this document. Mechanisms should be provided to 2382 measure and collect statistics from the network to support the 2383 traffic engineering function. Additional capabilities may be needed 2384 to help in the analysis of the statistics. The actions of these 2385 mechanisms should not adversely affect the accuracy and integrity of 2386 the statistics collected. The mechanisms for statistical data 2387 acquisition should also be able to scale as the network evolves. 2389 Traffic statistics may be classified according to long-term or short- 2390 term time scales. Long-term time scale traffic statistics are very 2391 useful for traffic engineering. Long-term time scale traffic 2392 statistics may capture or reflect periodicity in network workload 2393 (such as hourly, daily, and weekly variations in traffic profiles) as 2394 well as traffic trends. Aspects of the monitored traffic statistics 2395 may also depict class of service characteristics for a network 2396 supporting multiple classes of service. Analysis of the long-term 2397 traffic statistics may yield secondary statistics such as busy hour 2398 characteristics, traffic growth patterns, persistent congestion 2399 problems, hot-spot, and imbalances in link utilization caused by 2400 routing anomalies. 2402 A mechanism for constructing traffic matrices for both long-term and 2403 short-term traffic statistics should be in place. In multi-service 2404 IP networks, the traffic matrices may be constructed for different 2405 service classes. Each element of a traffic matrix represents a 2406 statistic of traffic flow between a pair of abstract nodes. An 2407 abstract node may represent a router, a collection of routers, or a 2408 site in a VPN. 2410 Measured traffic statistics should provide reasonable and reliable 2411 indicators of the current state of the network on the short-term 2412 scale. Some short term traffic statistics may reflect link 2413 utilization and link congestion status. Examples of congestion 2414 indicators include excessive packet delay, packet loss, and high 2415 resource utilization. Examples of mechanisms for distributing this 2416 kind of information include SNMP, probing techniques, FTP, IGP link 2417 state advertisements, etc. 2419 6.4. Network Survivability 2421 Network survivability refers to the capability of a network to 2422 maintain service continuity in the presence of faults. This can be 2423 accomplished by promptly recovering from network impairments and 2424 maintaining the required QoS for existing services after recovery. 2425 Survivability has become an issue of great concern within the 2426 Internet community due to the increasing demands to carry mission 2427 critical traffic, real-time traffic, and other high priority traffic 2428 over the Internet. Survivability can be addressed at the device 2429 level by developing network elements that are more reliable; and at 2430 the network level by incorporating redundancy into the architecture, 2431 design, and operation of networks. It is recommended that a 2432 philosophy of robustness and survivability should be adopted in the 2433 architecture, design, and operation of traffic engineering that 2434 control IP networks (especially public IP networks). Because 2435 different contexts may demand different levels of survivability, the 2436 mechanisms developed to support network survivability should be 2437 flexible so that they can be tailored to different needs. 2439 Failure protection and restoration capabilities have become available 2440 from multiple layers as network technologies have continued to 2441 improve. At the bottom of the layered stack, optical networks are 2442 now capable of providing dynamic ring and mesh restoration 2443 functionality at the wavelength level as well as traditional 2444 protection functionality. At the SONET/SDH layer survivability 2445 capability is provided with Automatic Protection Switching (APS) as 2446 well as self-healing ring and mesh architectures. Similar 2447 functionality is provided by layer 2 technologies such as ATM 2448 (generally with slower mean restoration times). Rerouting is 2449 traditionally used at the IP layer to restore service following link 2450 and node outages. Rerouting at the IP layer occurs after a period of 2451 routing convergence which may require seconds to minutes to complete. 2452 Some new developments in the MPLS context make it possible to achieve 2453 recovery at the IP layer prior to convergence [RFC3469]. 2455 To support advanced survivability requirements, path-oriented 2456 technologies such a MPLS can be used to enhance the survivability of 2457 IP networks in a potentially cost effective manner. The advantages 2458 of path oriented technologies such as MPLS for IP restoration becomes 2459 even more evident when class based protection and restoration 2460 capabilities are required. 2462 Recently, a common suite of control plane protocols has been proposed 2463 for both MPLS and optical transport networks under the acronym Multi- 2464 protocol Lambda Switching [AWD1]. This new paradigm of Multi- 2465 protocol Lambda Switching will support even more sophisticated mesh 2466 restoration capabilities at the optical layer for the emerging IP 2467 over WDM network architectures. 2469 Another important aspect regarding multi-layer survivability is that 2470 technologies at different layers provide protection and restoration 2471 capabilities at different temporal granularities (in terms of time 2472 scales) and at different bandwidth granularity (from packet-level to 2473 wavelength level). Protection and restoration capabilities can also 2474 be sensitive to different service classes and different network 2475 utility models. 2477 The impact of service outages varies significantly for different 2478 service classes depending upon the effective duration of the outage. 2479 The duration of an outage can vary from milliseconds (with minor 2480 service impact) to seconds (with possible call drops for IP telephony 2481 and session time-outs for connection oriented transactions) to 2482 minutes and hours (with potentially considerable social and business 2483 impact). 2485 Coordinating different protection and restoration capabilities across 2486 multiple layers in a cohesive manner to ensure network survivability 2487 is maintained at reasonable cost is a challenging task. Protection 2488 and restoration coordination across layers may not always be 2489 feasible, because networks at different layers may belong to 2490 different administrative domains. 2492 The following paragraphs present some of the general recommendations 2493 for protection and restoration coordination. 2495 o Protection and restoration capabilities from different layers 2496 should be coordinated whenever feasible and appropriate to provide 2497 network survivability in a flexible and cost effective manner. 2498 Minimization of function duplication across layers is one way to 2499 achieve the coordination. Escalation of alarms and other fault 2500 indicators from lower to higher layers may also be performed in a 2501 coordinated manner. A temporal order of restoration trigger 2502 timing at different layers is another way to coordinate multi- 2503 layer protection/restoration. 2505 o Spare capacity at higher layers is often regarded as working 2506 traffic at lower layers. Placing protection/restoration functions 2507 in many layers may increase redundancy and robustness, but it 2508 should not result in significant and avoidable inefficiencies in 2509 network resource utilization. 2511 o It is generally desirable to have protection and restoration 2512 schemes that are bandwidth efficient. 2514 o Failure notification throughout the network should be timely and 2515 reliable. 2517 o Alarms and other fault monitoring and reporting capabilities 2518 should be provided at appropriate layers. 2520 6.4.1. Survivability in MPLS Based Networks 2522 MPLS is an important emerging technology that enhances IP networks in 2523 terms of features, capabilities, and services. Because MPLS is path- 2524 oriented, it can potentially provide faster and more predictable 2525 protection and restoration capabilities than conventional hop by hop 2526 routed IP systems. This subsection describes some of the basic 2527 aspects and recommendations for MPLS networks regarding protection 2528 and restoration. See [RFC3469] for a more comprehensive discussion 2529 on MPLS based recovery. 2531 Protection types for MPLS networks can be categorized as link 2532 protection, node protection, path protection, and segment protection. 2534 o Link Protection: The objective for link protection is to protect 2535 an LSP from a given link failure. Under link protection, the path 2536 of the protection or backup LSP (the secondary LSP) is disjoint 2537 from the path of the working or operational LSP at the particular 2538 link over which protection is required. When the protected link 2539 fails, traffic on the working LSP is switched over to the 2540 protection LSP at the head-end of the failed link. This is a 2541 local repair method which can be fast. It might be more 2542 appropriate in situations where some network elements along a 2543 given path are less reliable than others. 2545 o Node Protection: The objective of LSP node protection is to 2546 protect an LSP from a given node failure. Under node protection, 2547 the path of the protection LSP is disjoint from the path of the 2548 working LSP at the particular node to be protected. The secondary 2549 path is also disjoint from the primary path at all links 2550 associated with the node to be protected. When the node fails, 2551 traffic on the working LSP is switched over to the protection LSP 2552 at the upstream LSR directly connected to the failed node. 2554 o Path Protection: The goal of LSP path protection is to protect an 2555 LSP from failure at any point along its routed path. Under path 2556 protection, the path of the protection LSP is completely disjoint 2557 from the path of the working LSP. The advantage of path 2558 protection is that the backup LSP protects the working LSP from 2559 all possible link and node failures along the path, except for 2560 failures that might occur at the ingress and egress LSRs, or for 2561 correlated failures that might impact both working and backup 2562 paths simultaneously. Additionally, since the path selection is 2563 end-to-end, path protection might be more efficient in terms of 2564 resource usage than link or node protection. However, path 2565 protection may be slower than link and node protection in general. 2567 o Segment Protection: An MPLS domain may be partitioned into 2568 multiple protection domains whereby a failure in a protection 2569 domain is rectified within that domain. In cases where an LSP 2570 traverses multiple protection domains, a protection mechanism 2571 within a domain only needs to protect the segment of the LSP that 2572 lies within the domain. Segment protection will generally be 2573 faster than path protection because recovery generally occurs 2574 closer to the fault. 2576 6.4.2. Protection Option 2578 Another issue to consider is the concept of protection options. The 2579 protection option uses the notation m:n protection, where m is the 2580 number of protection LSPs used to protect n working LSPs. Feasible 2581 protection options follow. 2583 o 1:1: one working LSP is protected/restored by one protection LSP. 2585 o 1:n: one protection LSP is used to protect/restore n working LSPs. 2587 o n:1: one working LSP is protected/restored by n protection LSPs, 2588 possibly with configurable load splitting ratio. When more than 2589 one protection LSP is used, it may be desirable to share the 2590 traffic across the protection LSPs when the working LSP fails to 2591 satisfy the bandwidth requirement of the traffic trunk associated 2592 with the working LSP. This may be especially useful when it is 2593 not feasible to find one path that can satisfy the bandwidth 2594 requirement of the primary LSP. 2596 o 1+1: traffic is sent concurrently on both the working LSP and the 2597 protection LSP. In this case, the egress LSR selects one of the 2598 two LSPs based on a local traffic integrity decision process, 2599 which compares the traffic received from both the working and the 2600 protection LSP and identifies discrepancies. It is unlikely that 2601 this option would be used extensively in IP networks due to its 2602 resource utilization inefficiency. However, if bandwidth becomes 2603 plentiful and cheap, then this option might become quite viable 2604 and attractive in IP networks. 2606 6.5. Traffic Engineering in Diffserv Environments 2608 This section provides an overview of the traffic engineering features 2609 and recommendations that are specifically pertinent to Differentiated 2610 Services (Diffserv) [RFC2475] capable IP networks. 2612 Increasing requirements to support multiple classes of traffic, such 2613 as best effort and mission critical data, in the Internet calls for 2614 IP networks to differentiate traffic according to some criteria, and 2615 to accord preferential treatment to certain types of traffic. Large 2616 numbers of flows can be aggregated into a few behavior aggregates 2617 based on some criteria in terms of common performance requirements in 2618 terms of packet loss ratio, delay, and jitter; or in terms of common 2619 fields within the IP packet headers. 2621 As Diffserv evolves and becomes deployed in operational networks, 2622 traffic engineering will be critical to ensuring that SLAs defined 2623 within a given Diffserv service model are met. Classes of service 2624 (CoS) can be supported in a Diffserv environment by concatenating 2625 per-hop behaviors (PHBs) along the routing path, using service 2626 provisioning mechanisms, and by appropriately configuring edge 2627 functionality such as traffic classification, marking, policing, and 2628 shaping. PHB is the forwarding behavior that a packet receives at a 2629 DS node (a Diffserv-compliant node). This is accomplished by means 2630 of buffer management and packet scheduling mechanisms. In this 2631 context, packets belonging to a class are those that are members of a 2632 corresponding ordering aggregate. 2634 Traffic engineering can be used as a compliment to Diffserv 2635 mechanisms to improve utilization of network resources, but not as a 2636 necessary element in general. When traffic engineering is used, it 2637 can be operated on an aggregated basis across all service classes 2638 [RFC3270] or on a per service class basis. The former is used to 2639 provide better distribution of the aggregate traffic load over the 2640 network resources. (See [RFC3270] for detailed mechanisms to support 2641 aggregate traffic engineering.) The latter case is discussed below 2642 since it is specific to the Diffserv environment, with so called 2643 Diffserv-aware traffic engineering [RFC4124]. 2645 For some Diffserv networks, it may be desirable to control the 2646 performance of some service classes by enforcing certain 2647 relationships between the traffic workload contributed by each 2648 service class and the amount of network resources allocated or 2649 provisioned for that service class. Such relationships between 2650 demand and resource allocation can be enforced using a combination 2651 of, for example: (1) traffic engineering mechanisms on a per service 2652 class basis that enforce the desired relationship between the amount 2653 of traffic contributed by a given service class and the resources 2654 allocated to that class, and (2) mechanisms that dynamically adjust 2655 the resources allocated to a given service class to relate to the 2656 amount of traffic contributed by that service class. 2658 It may also be desirable to limit the performance impact of high 2659 priority traffic on relatively low priority traffic. This can be 2660 achieved by, for example, controlling the percentage of high priority 2661 traffic that is routed through a given link. Another way to 2662 accomplish this is to increase link capacities appropriately so that 2663 lower priority traffic can still enjoy adequate service quality. 2664 When the ratio of traffic workload contributed by different service 2665 classes vary significantly from router to router, it may not suffice 2666 to rely exclusively on conventional IGP routing protocols or on 2667 traffic engineering mechanisms that are insensitive to different 2668 service classes. Instead, it may be desirable to perform traffic 2669 engineering, especially routing control and mapping functions, on a 2670 per service class basis. One way to accomplish this in a domain that 2671 supports both MPLS and Diffserv is to define class specific LSPs and 2672 to map traffic from each class onto one or more LSPs that correspond 2673 to that service class. An LSP corresponding to a given service class 2674 can then be routed and protected/restored in a class dependent 2675 manner, according to specific policies. 2677 Performing traffic engineering on a per class basis may require 2678 certain per-class parameters to be distributed. Note that it is 2679 common to have some classes share some aggregate constraint (e.g., 2680 maximum bandwidth requirement) without enforcing the constraint on 2681 each individual class. These classes then can be grouped into a 2682 class-type and per-class-type parameters can be distributed instead 2683 to improve scalability. It also allows better bandwidth sharing 2684 between classes in the same class-type. A class-type is a set of 2685 classes that satisfy the following two conditions: 2687 1) Classes in the same class-type have common aggregate requirements 2688 to satisfy required performance levels. 2690 2) There is no requirement to be enforced at the level of individual 2691 class in the class-type. Note that it is still possible, 2692 nevertheless, to implement some priority policies for classes in the 2693 same class-type to permit preferential access to the class-type 2694 bandwidth through the use of preemption priorities. 2696 An example of the class-type can be a low-loss class-type that 2697 includes both AF1-based and AF2-based Ordering Aggregates. With such 2698 a class-type, one may implement some priority policy which assigns 2699 higher preemption priority to AF1-based traffic trunks over AF2-based 2700 ones, vice versa, or the same priority. 2702 See [RFC4124] for detailed requirements on Diffserv-aware traffic 2703 engineering. 2705 6.6. Network Controllability 2707 Off-line (and on-line) traffic engineering considerations would be of 2708 limited utility if the network could not be controlled effectively to 2709 implement the results of TE decisions and to achieve desired network 2710 performance objectives. Capacity augmentation is a coarse grained 2711 solution to traffic engineering issues. However, it is simple and 2712 may be advantageous if bandwidth is abundant and cheap or if the 2713 current or expected network workload demands it. However, bandwidth 2714 is not always abundant and cheap, and the workload may not always 2715 demand additional capacity. Adjustments of administrative weights 2716 and other parameters associated with routing protocols provide finer 2717 grained control, but is difficult to use and imprecise because of the 2718 routing interactions that occur across the network. In certain 2719 network contexts, more flexible, finer grained approaches which 2720 provide more precise control over the mapping of traffic to routes 2721 and over the selection and placement of routes may be appropriate and 2722 useful. 2724 Control mechanisms can be manual (e.g., administrative 2725 configuration), partially-automated (e.g., scripts) or fully- 2726 automated (e.g., policy based management systems). Automated 2727 mechanisms are particularly required in large scale networks. Multi- 2728 vendor interoperability can be facilitated by developing and 2729 deploying standardized management systems (e.g., standard MIBs) and 2730 policies (PIBs) to support the control functions required to address 2731 traffic engineering objectives such as load distribution and 2732 protection/restoration. 2734 Network control functions should be secure, reliable, and stable as 2735 these are often needed to operate correctly in times of network 2736 impairments (e.g., during network congestion or security attacks). 2738 7. Inter-Domain Considerations 2740 Inter-domain traffic engineering is concerned with the performance 2741 optimization for traffic that originates in one administrative domain 2742 and terminates in a different one. 2744 Traffic exchange between autonomous systems in the Internet occurs 2745 through exterior gateway protocols. Currently, BGP [RFC4271] is the 2746 standard exterior gateway protocol for the Internet. BGP provides a 2747 number of attributes and capabilities (e.g., route filtering) that 2748 can be used for inter-domain traffic engineering. More specifically, 2749 BGP permits the control of routing information and traffic exchange 2750 between Autonomous Systems (AS's) in the Internet. BGP incorporates 2751 a sequential decision process which calculates the degree of 2752 preference for various routes to a given destination network. There 2753 are two fundamental aspects to inter-domain traffic engineering using 2754 BGP: 2756 o Route Redistribution: controlling the import and export of routes 2757 between AS's, and controlling the redistribution of routes between 2758 BGP and other protocols within an AS. 2760 o Best path selection: selecting the best path when there are 2761 multiple candidate paths to a given destination network. Best 2762 path selection is performed by the BGP decision process based on a 2763 sequential procedure, taking a number of different considerations 2764 into account. Ultimately, best path selection under BGP boils 2765 down to selecting preferred exit points out of an AS towards 2766 specific destination networks. The BGP path selection process can 2767 be influenced by manipulating the attributes associated with the 2768 BGP decision process. These attributes include: NEXT-HOP, WEIGHT 2769 (Cisco proprietary which is also implemented by some other 2770 vendors), LOCAL-PREFERENCE, AS-PATH, ROUTE-ORIGIN, MULTI-EXIT- 2771 DESCRIMINATOR (MED), IGP METRIC, etc. 2773 Route-maps provide the flexibility to implement complex BGP policies 2774 based on pre-configured logical conditions. In particular, Route- 2775 maps can be used to control import and export policies for incoming 2776 and outgoing routes, control the redistribution of routes between BGP 2777 and other protocols, and influence the selection of best paths by 2778 manipulating the attributes associated with the BGP decision process. 2779 Very complex logical expressions that implement various types of 2780 policies can be implemented using a combination of Route-maps, BGP- 2781 attributes, Access-lists, and Community attributes. 2783 When looking at possible strategies for inter-domain TE with BGP, it 2784 must be noted that the outbound traffic exit point is controllable, 2785 whereas the interconnection point where inbound traffic is received 2786 from an EBGP peer typically is not, unless a special arrangement is 2787 made with the peer sending the traffic. Therefore, it is up to each 2788 individual network to implement sound TE strategies that deal with 2789 the efficient delivery of outbound traffic from one's customers to 2790 one's peering points. The vast majority of TE policy is based upon a 2791 "closest exit" strategy, which offloads interdomain traffic at the 2792 nearest outbound peer point towards the destination autonomous 2793 system. Most methods of manipulating the point at which inbound 2794 traffic enters a network from an EBGP peer (inconsistent route 2795 announcements between peering points, AS pre-pending, and sending 2796 MEDs) are either ineffective, or not accepted in the peering 2797 community. 2799 Inter-domain TE with BGP is generally effective, but it is usually 2800 applied in a trial-and-error fashion. A systematic approach for 2801 inter-domain traffic engineering is yet to be devised. 2803 Inter-domain TE is inherently more difficult than intra-domain TE 2804 under the current Internet architecture. The reasons for this are 2805 both technical and administrative. Technically, while topology and 2806 link state information are helpful for mapping traffic more 2807 effectively, BGP does not propagate such information across domain 2808 boundaries for stability and scalability reasons. Administratively, 2809 there are differences in operating costs and network capacities 2810 between domains. Generally, what may be considered a good solution 2811 in one domain may not necessarily be a good solution in another 2812 domain. Moreover, it would generally be considered inadvisable for 2813 one domain to permit another domain to influence the routing and 2814 management of traffic in its network. 2816 MPLS TE-tunnels (explicit LSPs) can potentially add a degree of 2817 flexibility in the selection of exit points for inter-domain routing. 2818 The concept of relative and absolute metrics can be applied to this 2819 purpose. The idea is that if BGP attributes are defined such that 2820 the BGP decision process depends on IGP metrics to select exit points 2821 for inter-domain traffic, then some inter-domain traffic destined to 2822 a given peer network can be made to prefer a specific exit point by 2823 establishing a TE-tunnel between the router making the selection to 2824 the peering point via a TE-tunnel and assigning the TE-tunnel a 2825 metric which is smaller than the IGP cost to all other peering 2826 points. If a peer accepts and processes MEDs, then a similar MPLS 2827 TE-tunnel based scheme can be applied to cause certain entrance 2828 points to be preferred by setting MED to be an IGP cost, which has 2829 been modified by the tunnel metric. 2831 Similar to intra-domain TE, inter-domain TE is best accomplished when 2832 a traffic matrix can be derived to depict the volume of traffic from 2833 one autonomous system to another. 2835 Generally, redistribution of inter-domain traffic requires 2836 coordination between peering partners. An export policy in one 2837 domain that results in load redistribution across peer points with 2838 another domain can significantly affect the local traffic matrix 2839 inside the domain of the peering partner. This, in turn, will affect 2840 the intra-domain TE due to changes in the spatial distribution of 2841 traffic. Therefore, it is mutually beneficial for peering partners 2842 to coordinate with each other before attempting any policy changes 2843 that may result in significant shifts in inter-domain traffic. In 2844 certain contexts, this coordination can be quite challenging due to 2845 technical and non- technical reasons. 2847 It is a matter of speculation as to whether MPLS, or similar 2848 technologies, can be extended to allow selection of constrained paths 2849 across domain boundaries. 2851 8. Overview of Contemporary TE Practices in Operational IP Networks 2853 This section provides an overview of some contemporary traffic 2854 engineering practices in IP networks. The focus is primarily on the 2855 aspects that pertain to the control of the routing function in 2856 operational contexts. The intent here is to provide an overview of 2857 the commonly used practices. The discussion is not intended to be 2858 exhaustive. 2860 Currently, service providers apply many of the traffic engineering 2861 mechanisms discussed in this document to optimize the performance of 2862 their IP networks. These techniques include capacity planning for 2863 long time scales, routing control using IGP metrics and MPLS for 2864 medium time scales, the overlay model also for medium time scales, 2865 and traffic management mechanisms for short time scale. 2867 When a service provider plans to build an IP network, or expand the 2868 capacity of an existing network, effective capacity planning should 2869 be an important component of the process. Such plans may take the 2870 following aspects into account: location of new nodes if any, 2871 existing and predicted traffic patterns, costs, link capacity, 2872 topology, routing design, and survivability. 2874 Performance optimization of operational networks is usually an 2875 ongoing process in which traffic statistics, performance parameters, 2876 and fault indicators are continually collected from the network. 2877 This empirical data is then analyzed and used to trigger various 2878 traffic engineering mechanisms. Tools that perform what-if analysis 2879 can also be used to assist the TE process by allowing various 2880 scenarios to be reviewed before a new set of configurations are 2881 implemented in the operational network. 2883 Traditionally, intra-domain real-time TE with IGP is done by 2884 increasing the OSPF or IS-IS metric of a congested link until enough 2885 traffic has been diverted from that link. This approach has some 2886 limitations as discussed in Section 6.1. Recently, some new intra- 2887 domain TE approaches/tools have been proposed [RR94] [FT00] [FT01] 2888 [WANG]. Such approaches/tools take traffic matrix, network topology, 2889 and network performance objective(s) as input, and produce some link 2890 metrics and possibly some unequal load-sharing ratios to be set at 2891 the head-end routers of some ECMPs as output. These new progresses 2892 open new possibility for intra-domain TE with IGP to be done in a 2893 more systematic way. 2895 The overlay model (IP over ATM, or IP over Frame Relay) is another 2896 approach which was commonly used [AWD2], but has been replaced by 2897 MPLS and router hardware technology. 2899 Deployment of MPLS for traffic engineering applications has commenced 2900 in some service provider networks. One operational scenario is to 2901 deploy MPLS in conjunction with an IGP (IS-IS-TE or OSPF-TE) that 2902 supports the traffic engineering extensions, in conjunction with 2903 constraint-based routing for explicit route computations, and a 2904 signaling protocol (e.g., RSVP-TE) for LSP instantiation. 2906 In contemporary MPLS traffic engineering contexts, network 2907 administrators specify and configure link attributes and resource 2908 constraints such as maximum reservable bandwidth and resource class 2909 attributes for links (interfaces) within the MPLS domain. A link 2910 state protocol that supports TE extensions (IS-IS-TE or OSPF-TE) is 2911 used to propagate information about network topology and link 2912 attribute to all routers in the routing area. Network administrators 2913 also specify all the LSPs that are to originate each router. For 2914 each LSP, the network administrator specifies the destination node 2915 and the attributes of the LSP which indicate the requirements that to 2916 be satisfied during the path selection process. Each router then 2917 uses a local constraint-based routing process to compute explicit 2918 paths for all LSPs originating from it. Subsequently, a signaling 2919 protocol is used to instantiate the LSPs. By assigning proper 2920 bandwidth values to links and LSPs, congestion caused by uneven 2921 traffic distribution can generally be avoided or mitigated. 2923 The bandwidth attributes of LSPs used for traffic engineering can be 2924 updated periodically. The basic concept is that the bandwidth 2925 assigned to an LSP should relate in some manner to the bandwidth 2926 requirements of traffic that actually flows through the LSP. The 2927 traffic attribute of an LSP can be modified to accommodate traffic 2928 growth and persistent traffic shifts. If network congestion occurs 2929 due to some unexpected events, existing LSPs can be rerouted to 2930 alleviate the situation or network administrator can configure new 2931 LSPs to divert some traffic to alternative paths. The reservable 2932 bandwidth of the congested links can also be reduced to force some 2933 LSPs to be rerouted to other paths. 2935 In an MPLS domain, a traffic matrix can also be estimated by 2936 monitoring the traffic on LSPs. Such traffic statistics can be used 2937 for a variety of purposes including network planning and network 2938 optimization. Current practice suggests that deploying an MPLS 2939 network consisting of hundreds of routers and thousands of LSPs is 2940 feasible. In summary, recent deployment experience suggests that 2941 MPLS approach is very effective for traffic engineering in IP 2942 networks [XIAO]. 2944 As mentioned previously in Section 7, one usually has no direct 2945 control over the distribution of inbound traffic. Therefore, the 2946 main goal of contemporary inter-domain TE is to optimize the 2947 distribution of outbound traffic between multiple inter-domain links. 2948 When operating a global network, maintaining the ability to operate 2949 the network in a regional fashion where desired, while continuing to 2950 take advantage of the benefits of a global network, also becomes an 2951 important objective. 2953 Inter-domain TE with BGP usually begins with the placement of 2954 multiple peering interconnection points in locations that have high 2955 peer density, are in close proximity to originating/terminating 2956 traffic locations on one's own network, and are lowest in cost. 2957 There are generally several locations in each region of the world 2958 where the vast majority of major networks congregate and 2959 interconnect. Some location-decision problems that arise in 2960 association with inter-domain routing are discussed in [AWD5]. 2962 Once the locations of the interconnects are determined, and circuits 2963 are implemented, one decides how best to handle the routes heard from 2964 the peer, as well as how to propagate the peers' routes within one's 2965 own network. One way to engineer outbound traffic flows on a network 2966 with many EBGP peers is to create a hierarchy of peers. Generally, 2967 the Local Preferences of all peers are set to the same value so that 2968 the shortest AS paths will be chosen to forward traffic. Then, by 2969 over-writing the inbound MED metric (Multi-exit-discriminator metric, 2970 also referred to as "BGP metric". Both terms are used 2971 interchangeably in this document) with BGP metrics to routes received 2972 at different peers, the hierarchy can be formed. For example, all 2973 Local Preferences can be set to 200, preferred private peers can be 2974 assigned a BGP metric of 50, the rest of the private peers can be 2975 assigned a BGP metric of 100, and public peers can be assigned a BGP 2976 metric of 600. "Preferred" peers might be defined as those peers 2977 with whom the most available capacity exists, whose customer base is 2978 larger in comparison to other peers, whose interconnection costs are 2979 the lowest, and with whom upgrading existing capacity is the easiest. 2980 In a network with low utilization at the edge, this works well. The 2981 same concept could be applied to a network with higher edge 2982 utilization by creating more levels of BGP metrics between peers, 2983 allowing for more granularity in selecting the exit points for 2984 traffic bound for a dual homed customer on a peer's network. 2986 By only replacing inbound MED metrics with BGP metrics, only equal 2987 AS-Path length routes' exit points are being changed. (The BGP 2988 decision considers Local Preference first, then AS-Path length, and 2989 then BGP metric). For example, assume a network has two possible 2990 egress points, peer A and peer B. Each peer has 40% of the 2991 Internet's routes exclusively on its network, while the remaining 20% 2992 of the Internet's routes are from customers who dual home between A 2993 and B. Assume that both peers have a Local Preference of 200 and a 2994 BGP metric of 100. If the link to peer A is congested, increasing 2995 its BGP metric while leaving the Local Preference at 200 will ensure 2996 that the 20% of total routes belonging to dual homed customers will 2997 prefer peer B as the exit point. The previous example would be used 2998 in a situation where all exit points to a given peer were close to 2999 congestion levels, and traffic needed to be shifted away from that 3000 peer entirely. 3002 When there are multiple exit points to a given peer, and only one of 3003 them is congested, it is not necessary to shift traffic away from the 3004 peer entirely, but only from the one congested circuit. This can be 3005 achieved by using passive IGP-metrics, AS-path filtering, or prefix 3006 filtering. 3008 Occasionally, more drastic changes are needed, for example, in 3009 dealing with a "problem peer" who is difficult to work with on 3010 upgrades or is charging high prices for connectivity to their 3011 network. In that case, the Local Preference to that peer can be 3012 reduced below the level of other peers. This effectively reduces the 3013 amount of traffic sent to that peer to only originating traffic 3014 (assuming no transit providers are involved). This type of change 3015 can affect a large amount of traffic, and is only used after other 3016 methods have failed to provide the desired results. 3018 Although it is not much of an issue in regional networks, the 3019 propagation of a peer's routes back through the network must be 3020 considered when a network is peering on a global scale. Sometimes, 3021 business considerations can influence the choice of BGP policies in a 3022 given context. For example, it may be imprudent, from a business 3023 perspective, to operate a global network and provide full access to 3024 the global customer base to a small network in a particular country. 3025 However, for the purpose of providing one's own customers with 3026 quality service in a particular region, good connectivity to that in- 3027 country network may still be necessary. This can be achieved by 3028 assigning a set of communities at the edge of the network, which have 3029 a known behavior when routes tagged with those communities are 3030 propagating back through the core. Routes heard from local peers 3031 will be prevented from propagating back to the global network, 3032 whereas routes learned from larger peers may be allowed to propagate 3033 freely throughout the entire global network. By implementing a 3034 flexible community strategy, the benefits of using a single global AS 3035 Number (ASN) can be realized, while the benefits of operating 3036 regional networks can also be taken advantage of. An alternative to 3037 doing this is to use different ASNs in different regions, with the 3038 consequence that the AS path length for routes announced by that 3039 service provider will increase. 3041 9. Conclusion 3043 This document described principles for traffic engineering in the 3044 Internet. It presented an overview of some of the basic issues 3045 surrounding traffic engineering in IP networks. The context of TE 3046 was described, a TE process models and a taxonomy of TE styles were 3047 presented. A brief historical review of pertinent developments 3048 related to traffic engineering was provided. A survey of 3049 contemporary TE techniques in operational networks was presented. 3050 Additionally, the document specified a set of generic requirements, 3051 recommendations, and options for Internet traffic engineering. 3053 10. Security Considerations 3055 This document does not introduce new security issues. 3057 11. IANA Considerations 3059 This draft makes no requests for IANA action. 3061 12. Acknowledgments 3063 The acknowledgements in RFC3272 were as below. All people who helped 3064 in the production of that document also need to be thanked for the 3065 carry-over into this new document. 3067 The authors would like to thank Jim Boyle for inputs on the 3068 recommendations section, Francois Le Faucheur for inputs on Diffserv 3069 aspects, Blaine Christian for inputs on measurement, Gerald Ash for 3070 inputs on routing in telephone networks and for text on event- 3071 dependent TE methods, Steven Wright for inputs on network 3072 controllability, and Jonathan Aufderheide for inputs on inter-domain 3073 TE with BGP. Special thanks to Randy Bush for proposing the TE 3074 taxonomy based on "tactical vs strategic" methods. The subsection 3075 describing an "Overview of ITU Activities Related to Traffic 3076 Engineering" was adapted from a contribution by Waisum Lai. Useful 3077 feedback and pointers to relevant materials were provided by J. Noel 3078 Chiappa. Additional comments were provided by Glenn Grotefeld during 3079 the working last call process. Finally, the authors would like to 3080 thank Ed Kern, the TEWG co-chair, for his comments and support. 3082 The production of this document include a fix to the original text 3083 resulting from an Errata Report by Jean-Michel Grimaldi. 3085 The authors of this document would also like to thank TBD. 3087 13. Contributors 3089 Much of the text in this document is derived from RFC 3272. The 3090 authors of this document would like to express their gratitude to all 3091 involved in that work. Although the source text has been edited in 3092 the production of this document, the orginal authors should be 3093 considered as Contributors to this work. They were: 3095 Daniel O. Awduche 3096 Movaz Networks 3097 7926 Jones Branch Drive, Suite 615 3098 McLean, VA 22102 3100 Phone: 703-298-5291 3101 EMail: awduche@movaz.com 3103 Angela Chiu 3104 Celion Networks 3105 1 Sheila Dr., Suite 2 3106 Tinton Falls, NJ 07724 3108 Phone: 732-747-9987 3109 EMail: angela.chiu@celion.com 3111 Anwar Elwalid 3112 Lucent Technologies 3113 Murray Hill, NJ 07974 3115 Phone: 908 582-7589 3116 EMail: anwar@lucent.com 3118 Indra Widjaja 3119 Bell Labs, Lucent Technologies 3120 600 Mountain Avenue 3121 Murray Hill, NJ 07974 3123 Phone: 908 582-0435 3124 EMail: iwidjaja@research.bell-labs.com 3126 XiPeng Xiao 3127 Redback Networks 3128 300 Holger Way 3129 San Jose, CA 95134 3131 Phone: 408-750-5217 3132 EMail: xipeng@redback.com 3134 The first version of this document was produced by the TEAS Working 3135 Group's RFC3272bis Design Team. The team members are all 3136 Contributors to this document. The full list of contributors to this 3137 document is: 3139 Acee Lindem 3140 EMail: acee@cisco.com 3142 Adrian Farrel 3143 EMail: adrian@olddog.co.uk 3145 Aijun Wang 3146 EMail: wangaijun@tsinghua.org.cn 3148 Daniele Ceccarelli 3149 EMail: daniele.ceccarelli@ericsson.com 3151 Dieter Beller 3152 EMail: dieter.beller@nokia.com 3154 Jeff Tantsura 3155 EMail: jefftant.ietf@gmail.com 3157 Julien Meuric 3158 EMail: julien.meuric@orange.com 3160 Liu Hua 3161 EMail: hliu@ciena.com 3163 Loa Andersson 3164 EMail: loa@pi.nu 3166 Luis Miguel Contreras 3167 EMail: luismiguel.contrerasmurillo@telefonica.com 3169 Martin Horneffer 3170 EMail: Martin.Horneffer@telekom.de 3171 Tarek Saad 3172 EMail: tsaad@cisco.com 3174 Xufeng Liu 3175 EMail: xufeng.liu.ietf@gmail.com 3177 Gert Grammel 3178 EMail: ggrammel@juniper.net 3180 14. Informative References 3182 [ASH2] Ash, J., "Dynamic Routing in Telecommunications Networks", 3183 Book McGraw Hill, 1998. 3185 [AWD1] Awduche, D. and Y. Rekhter, "Multiprocotol Lambda 3186 Switching - Combining MPLS Traffic Engineering Control 3187 with Optical Crossconnects", Article IEEE Communications 3188 Magazine, March 2001. 3190 [AWD2] Awduche, D., "MPLS and Traffic Engineering in IP 3191 Networks", Article IEEE Communications Magazine, December 3192 1999. 3194 [AWD5] Awduche, D., "An Approach to Optimal Peering Between 3195 Autonomous Systems in the Internet", Paper International 3196 Conference on Computer Communications and Networks 3197 (ICCCN'98), October 1998. 3199 [CRUZ] "A Calculus for Network Delay, Part II, Network Analysis", 3200 Transaction IEEE Transactions on Information Theory, vol. 3201 37, pp. 132-141, 1991. 3203 [ELW95] Elwalid, A., Mitra, D., and R. Wentworth, "A New Approach 3204 for Allocating Buffers and Bandwidth to Heterogeneous, 3205 Regulated Traffic in an ATM Node", Article IEEE Journal on 3206 Selected Areas in Communications, 13.6, pp. 1115-1127, 3207 August 1995. 3209 [FLJA93] Floyd, S. and V. Jacobson, "Random Early Detection 3210 Gateways for Congestion Avoidance", Article IEEE/ACM 3211 Transactions on Networking, Vol. 1, p. 387-413, November 3212 1993. 3214 [FLOY94] Floyd, S., "TCP and Explicit Congestion Notification", 3215 Article ACM Computer Communication Review, V. 24, No. 5, 3216 p. 10-23, October 1994. 3218 [FT00] Fortz, B. and M. Thorup, "Internet Traffic Engineering by 3219 Optimizing OSPF Weights", Article IEEE INFOCOM 2000, March 3220 2000. 3222 [FT01] Fortz, B. and M. Thorup, "Optimizing OSPF/IS-IS Weights in 3223 a Changing World", n.d., 3224 . 3226 [HUSS87] Hurley, B., Seidl, C., and W. Sewel, "A Survey of Dynamic 3227 Routing Methods for Circuit-Switched Traffic", 3228 Article IEEE Communication Magazine, September 1987. 3230 [I-D.ietf-teas-yang-te-topo] 3231 Liu, X., Bryskin, I., Beeram, V., Saad, T., Shah, H., and 3232 O. Dios, "YANG Data Model for Traffic Engineering (TE) 3233 Topologies", draft-ietf-teas-yang-te-topo-22 (work in 3234 progress), June 2019. 3236 [I-D.ietf-tewg-qos-routing] 3237 Ash, G., "Traffic Engineering & QoS Methods for IP-, ATM-, 3238 & Based Multiservice Networks", draft-ietf-tewg-qos- 3239 routing-04 (work in progress), October 2001. 3241 [ITU-E600] 3242 "Terms and Definitions of Traffic Engineering", 3243 Recommendation ITU-T Recommendation E.600, March 1993. 3245 [ITU-E701] 3246 "Reference Connections for Traffic Engineering", 3247 Recommendation ITU-T Recommendation E.701, October 1993. 3249 [ITU-E801] 3250 "Framework for Service Quality Agreement", 3251 Recommendation ITU-T Recommendation E.801, October 1996. 3253 [MA] Ma, Q., "Quality of Service Routing in Integrated Services 3254 Networks", Ph.D. PhD Dissertation, CMU-CS-98-138, CMU, 3255 1998. 3257 [MATE] Elwalid, A., Jin, C., Low, S., and I. Widjaja, "MATE - 3258 MPLS Adaptive Traffic Engineering", 3259 Proceedings INFOCOM'01, April 2001. 3261 [MCQ80] McQuillan, J., Richer, I., and E. Rosen, "The New Routing 3262 Algorithm for the ARPANET", Transaction IEEE Transactions 3263 on Communications, vol. 28, no. 5, p. 711-719, May 1980. 3265 [MR99] Mitra, D. and K. Ramakrishnan, "A Case Study of 3266 Multiservice, Multipriority Traffic Engineering Design for 3267 Data Networks", Proceedings Globecom'99, December 1999. 3269 [RFC1992] Castineyra, I., Chiappa, N., and M. Steenstrup, "The 3270 Nimrod Routing Architecture", RFC 1992, 3271 DOI 10.17487/RFC1992, August 1996, 3272 . 3274 [RFC2205] Braden, R., Ed., Zhang, L., Berson, S., Herzog, S., and S. 3275 Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1 3276 Functional Specification", RFC 2205, DOI 10.17487/RFC2205, 3277 September 1997, . 3279 [RFC2328] Moy, J., "OSPF Version 2", STD 54, RFC 2328, 3280 DOI 10.17487/RFC2328, April 1998, 3281 . 3283 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 3284 "Framework for IP Performance Metrics", RFC 2330, 3285 DOI 10.17487/RFC2330, May 1998, 3286 . 3288 [RFC2386] Crawley, E., Nair, R., Rajagopalan, B., and H. Sandick, "A 3289 Framework for QoS-based Routing in the Internet", 3290 RFC 2386, DOI 10.17487/RFC2386, August 1998, 3291 . 3293 [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, 3294 "Definition of the Differentiated Services Field (DS 3295 Field) in the IPv4 and IPv6 Headers", RFC 2474, 3296 DOI 10.17487/RFC2474, December 1998, 3297 . 3299 [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., 3300 and W. Weiss, "An Architecture for Differentiated 3301 Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, 3302 . 3304 [RFC2597] Heinanen, J., Baker, F., Weiss, W., and J. Wroclawski, 3305 "Assured Forwarding PHB Group", RFC 2597, 3306 DOI 10.17487/RFC2597, June 1999, 3307 . 3309 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring 3310 Connectivity", RFC 2678, DOI 10.17487/RFC2678, September 3311 1999, . 3313 [RFC2702] Awduche, D., Malcolm, J., Agogbua, J., O'Dell, M., and J. 3314 McManus, "Requirements for Traffic Engineering Over MPLS", 3315 RFC 2702, DOI 10.17487/RFC2702, September 1999, 3316 . 3318 [RFC2722] Brownlee, N., Mills, C., and G. Ruth, "Traffic Flow 3319 Measurement: Architecture", RFC 2722, 3320 DOI 10.17487/RFC2722, October 1999, 3321 . 3323 [RFC2753] Yavatkar, R., Pendarakis, D., and R. Guerin, "A Framework 3324 for Policy-based Admission Control", RFC 2753, 3325 DOI 10.17487/RFC2753, January 2000, 3326 . 3328 [RFC2961] Berger, L., Gan, D., Swallow, G., Pan, P., Tommasi, F., 3329 and S. Molendini, "RSVP Refresh Overhead Reduction 3330 Extensions", RFC 2961, DOI 10.17487/RFC2961, April 2001, 3331 . 3333 [RFC2998] Bernet, Y., Ford, P., Yavatkar, R., Baker, F., Zhang, L., 3334 Speer, M., Braden, R., Davie, B., Wroclawski, J., and E. 3335 Felstaine, "A Framework for Integrated Services Operation 3336 over Diffserv Networks", RFC 2998, DOI 10.17487/RFC2998, 3337 November 2000, . 3339 [RFC3031] Rosen, E., Viswanathan, A., and R. Callon, "Multiprotocol 3340 Label Switching Architecture", RFC 3031, 3341 DOI 10.17487/RFC3031, January 2001, 3342 . 3344 [RFC3086] Nichols, K. and B. Carpenter, "Definition of 3345 Differentiated Services Per Domain Behaviors and Rules for 3346 their Specification", RFC 3086, DOI 10.17487/RFC3086, 3347 April 2001, . 3349 [RFC3124] Balakrishnan, H. and S. Seshan, "The Congestion Manager", 3350 RFC 3124, DOI 10.17487/RFC3124, June 2001, 3351 . 3353 [RFC3209] Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan, V., 3354 and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP 3355 Tunnels", RFC 3209, DOI 10.17487/RFC3209, December 2001, 3356 . 3358 [RFC3270] Le Faucheur, F., Wu, L., Davie, B., Davari, S., Vaananen, 3359 P., Krishnan, R., Cheval, P., and J. Heinanen, "Multi- 3360 Protocol Label Switching (MPLS) Support of Differentiated 3361 Services", RFC 3270, DOI 10.17487/RFC3270, May 2002, 3362 . 3364 [RFC3272] Awduche, D., Chiu, A., Elwalid, A., Widjaja, I., and X. 3365 Xiao, "Overview and Principles of Internet Traffic 3366 Engineering", RFC 3272, DOI 10.17487/RFC3272, May 2002, 3367 . 3369 [RFC3469] Sharma, V., Ed. and F. Hellstrand, Ed., "Framework for 3370 Multi-Protocol Label Switching (MPLS)-based Recovery", 3371 RFC 3469, DOI 10.17487/RFC3469, February 2003, 3372 . 3374 [RFC3630] Katz, D., Kompella, K., and D. Yeung, "Traffic Engineering 3375 (TE) Extensions to OSPF Version 2", RFC 3630, 3376 DOI 10.17487/RFC3630, September 2003, 3377 . 3379 [RFC3945] Mannie, E., Ed., "Generalized Multi-Protocol Label 3380 Switching (GMPLS) Architecture", RFC 3945, 3381 DOI 10.17487/RFC3945, October 2004, 3382 . 3384 [RFC4124] Le Faucheur, F., Ed., "Protocol Extensions for Support of 3385 Diffserv-aware MPLS Traffic Engineering", RFC 4124, 3386 DOI 10.17487/RFC4124, June 2005, 3387 . 3389 [RFC4271] Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A 3390 Border Gateway Protocol 4 (BGP-4)", RFC 4271, 3391 DOI 10.17487/RFC4271, January 2006, 3392 . 3394 [RFC4655] Farrel, A., Vasseur, J., and J. Ash, "A Path Computation 3395 Element (PCE)-Based Architecture", RFC 4655, 3396 DOI 10.17487/RFC4655, August 2006, 3397 . 3399 [RFC5305] Li, T. and H. Smit, "IS-IS Extensions for Traffic 3400 Engineering", RFC 5305, DOI 10.17487/RFC5305, October 3401 2008, . 3403 [RFC5440] Vasseur, JP., Ed. and JL. Le Roux, Ed., "Path Computation 3404 Element (PCE) Communication Protocol (PCEP)", RFC 5440, 3405 DOI 10.17487/RFC5440, March 2009, 3406 . 3408 [RFC6241] Enns, R., Ed., Bjorklund, M., Ed., Schoenwaelder, J., Ed., 3409 and A. Bierman, Ed., "Network Configuration Protocol 3410 (NETCONF)", RFC 6241, DOI 10.17487/RFC6241, June 2011, 3411 . 3413 [RFC6805] King, D., Ed. and A. Farrel, Ed., "The Application of the 3414 Path Computation Element Architecture to the Determination 3415 of a Sequence of Domains in MPLS and GMPLS", RFC 6805, 3416 DOI 10.17487/RFC6805, November 2012, 3417 . 3419 [RFC7390] Rahman, A., Ed. and E. Dijk, Ed., "Group Communication for 3420 the Constrained Application Protocol (CoAP)", RFC 7390, 3421 DOI 10.17487/RFC7390, October 2014, 3422 . 3424 [RFC7679] Almes, G., Kalidindi, S., Zekauskas, M., and A. Morton, 3425 Ed., "A One-Way Delay Metric for IP Performance Metrics 3426 (IPPM)", STD 81, RFC 7679, DOI 10.17487/RFC7679, January 3427 2016, . 3429 [RFC7680] Almes, G., Kalidindi, S., Zekauskas, M., and A. Morton, 3430 Ed., "A One-Way Loss Metric for IP Performance Metrics 3431 (IPPM)", STD 82, RFC 7680, DOI 10.17487/RFC7680, January 3432 2016, . 3434 [RFC7752] Gredler, H., Ed., Medved, J., Previdi, S., Farrel, A., and 3435 S. Ray, "North-Bound Distribution of Link-State and 3436 Traffic Engineering (TE) Information Using BGP", RFC 7752, 3437 DOI 10.17487/RFC7752, March 2016, 3438 . 3440 [RFC7950] Bjorklund, M., Ed., "The YANG 1.1 Data Modeling Language", 3441 RFC 7950, DOI 10.17487/RFC7950, August 2016, 3442 . 3444 [RFC8040] Bierman, A., Bjorklund, M., and K. Watsen, "RESTCONF 3445 Protocol", RFC 8040, DOI 10.17487/RFC8040, January 2017, 3446 . 3448 [RFC8231] Crabbe, E., Minei, I., Medved, J., and R. Varga, "Path 3449 Computation Element Communication Protocol (PCEP) 3450 Extensions for Stateful PCE", RFC 8231, 3451 DOI 10.17487/RFC8231, September 2017, 3452 . 3454 [RFC8281] Crabbe, E., Minei, I., Sivabalan, S., and R. Varga, "Path 3455 Computation Element Communication Protocol (PCEP) 3456 Extensions for PCE-Initiated LSP Setup in a Stateful PCE 3457 Model", RFC 8281, DOI 10.17487/RFC8281, December 2017, 3458 . 3460 [RFC8283] Farrel, A., Ed., Zhao, Q., Ed., Li, Z., and C. Zhou, "An 3461 Architecture for Use of PCE and the PCE Communication 3462 Protocol (PCEP) in a Network with Central Control", 3463 RFC 8283, DOI 10.17487/RFC8283, December 2017, 3464 . 3466 [RR94] Rodrigues, M. and K. Ramakrishnan, "Optimal Routing in 3467 Shortest Path Networks", Proceedings ITS'94, Rio de 3468 Janeiro, Brazil, 1994. 3470 [SLDC98] Suter, B., Lakshman, T., Stiliadis, D., and A. Choudhury, 3471 "Design Considerations for Supporting TCP with Per-flow 3472 Queueing", Proceedings INFOCOM'98, p. 299-306, 1998. 3474 [WANG] Wang, Y., Wang, Z., and L. Zhang, "Internet traffic 3475 engineering without full mesh overlaying", 3476 Proceedings INFOCOM'2001, April 2001. 3478 [XIAO] Xiao, X., Hannan, A., Bailey, B., and L. Ni, "Traffic 3479 Engineering with MPLS in the Internet", Article IEEE 3480 Network Magazine, March 2000. 3482 [YARE95] Yang, C. and A. Reddy, "A Taxonomy for Congestion Control 3483 Algorithms in Packet Switching Networks", Article IEEE 3484 Network Magazine, p. 34-45, 1995. 3486 Author's Address 3488 Adrian Farrel (editor) 3489 Old Dog Consulting 3491 Email: adrian@olddog.co.uk