idnits 2.17.1 draft-dt-teas-rfc3272bis-10.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 (April 28, 2020) is 1456 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) -- Obsolete informational reference (is this intentional?): RFC 7810 (Obsoleted by RFC 8570) Summary: 0 errors (**), 0 flaws (~~), 1 warning (==), 4 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) April 28, 2020 5 Intended status: Informational 6 Expires: October 30, 2020 8 Overview and Principles of Internet Traffic Engineering 9 draft-dt-teas-rfc3272bis-10 11 Abstract 13 This document describes the principles of Traffic Engineering (TE) in 14 the Internet. The document is intended to promote better 15 understanding of the issues surrounding traffic engineering in IP 16 networks and the networks that support IP networking, and to provide 17 a common basis for the development of traffic engineering 18 capabilities for the Internet. The principles, architectures, and 19 methodologies for performance evaluation and performance optimization 20 of operational networks are discussed throughout this document. 22 This work was first published as RFC 3272 in May 2002. This document 23 obsoletes RFC 3272 by making a complete update to bring the text in 24 line with best current practices for Internet traffic engineering and 25 to include references to the latest relevant work in the IETF. 27 Status of This Memo 29 This Internet-Draft is submitted in full conformance with the 30 provisions of BCP 78 and BCP 79. 32 Internet-Drafts are working documents of the Internet Engineering 33 Task Force (IETF). Note that other groups may also distribute 34 working documents as Internet-Drafts. The list of current Internet- 35 Drafts is at https://datatracker.ietf.org/drafts/current/. 37 Internet-Drafts are draft documents valid for a maximum of six months 38 and may be updated, replaced, or obsoleted by other documents at any 39 time. It is inappropriate to use Internet-Drafts as reference 40 material or to cite them other than as "work in progress." 42 This Internet-Draft will expire on October 30, 2020. 44 Copyright Notice 46 Copyright (c) 2020 IETF Trust and the persons identified as the 47 document authors. All rights reserved. 49 This document is subject to BCP 78 and the IETF Trust's Legal 50 Provisions Relating to IETF Documents 51 (https://trustee.ietf.org/license-info) in effect on the date of 52 publication of this document. Please review these documents 53 carefully, as they describe your rights and restrictions with respect 54 to this document. Code Components extracted from this document must 55 include Simplified BSD License text as described in Section 4.e of 56 the Trust Legal Provisions and are provided without warranty as 57 described in the Simplified BSD License. 59 Table of Contents 61 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4 62 1.1. What is Internet Traffic Engineering? . . . . . . . . . . 4 63 1.2. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 6 64 1.3. Terminology . . . . . . . . . . . . . . . . . . . . . . . 7 65 2. Background . . . . . . . . . . . . . . . . . . . . . . . . . 10 66 2.1. Context of Internet Traffic Engineering . . . . . . . . . 10 67 2.2. Network Context . . . . . . . . . . . . . . . . . . . . . 11 68 2.3. Problem Context . . . . . . . . . . . . . . . . . . . . . 13 69 2.3.1. Congestion and its Ramifications . . . . . . . . . . 14 70 2.4. Solution Context . . . . . . . . . . . . . . . . . . . . 15 71 2.4.1. Combating the Congestion Problem . . . . . . . . . . 17 72 2.5. Implementation and Operational Context . . . . . . . . . 21 73 3. Traffic Engineering Process Models . . . . . . . . . . . . . 21 74 3.1. Components of the Traffic Engineering Process Model . . . 22 75 4. Review of TE Techniques . . . . . . . . . . . . . . . . . . . 22 76 4.1. Overview of IETF Projects Related to Traffic Engineering 23 77 4.1.1. Constraint-Based Routing . . . . . . . . . . . . . . 23 78 4.1.2. Integrated Services . . . . . . . . . . . . . . . . . 23 79 4.1.3. RSVP . . . . . . . . . . . . . . . . . . . . . . . . 24 80 4.1.4. Differentiated Services . . . . . . . . . . . . . . . 25 81 4.1.5. MPLS . . . . . . . . . . . . . . . . . . . . . . . . 26 82 4.1.6. Generalized MPLS . . . . . . . . . . . . . . . . . . 27 83 4.1.7. IP Performance Metrics . . . . . . . . . . . . . . . 27 84 4.1.8. Flow Measurement . . . . . . . . . . . . . . . . . . 28 85 4.1.9. Endpoint Congestion Management . . . . . . . . . . . 28 86 4.1.10. TE Extensions to the IGPs . . . . . . . . . . . . . . 29 87 4.1.11. Link-State BGP . . . . . . . . . . . . . . . . . . . 29 88 4.1.12. Path Computation Element . . . . . . . . . . . . . . 29 89 4.1.13. Application-Layer Traffic Optimization . . . . . . . 30 90 4.1.14. Segment Routing with MPLS encapsuation (SR-MPLS) . . 30 91 4.1.15. Network Virtualization and Abstraction . . . . . . . 31 92 4.1.16. Deterministic Networking . . . . . . . . . . . . . . 32 93 4.1.17. Network TE State Definition and Presentation . . . . 32 94 4.1.18. System Management and Control Interfaces . . . . . . 32 95 4.2. Content Distribution . . . . . . . . . . . . . . . . . . 33 96 5. Taxonomy of Traffic Engineering Systems . . . . . . . . . . . 33 97 5.1. Time-Dependent Versus State-Dependent Versus Event 98 Dependent . . . . . . . . . . . . . . . . . . . . . . . . 34 99 5.2. Offline Versus Online . . . . . . . . . . . . . . . . . . 35 100 5.3. Centralized Versus Distributed . . . . . . . . . . . . . 36 101 5.3.1. Hybrid Systems . . . . . . . . . . . . . . . . . . . 36 102 5.3.2. Considerations for Software Defined Networking . . . 36 103 5.4. Local Versus Global . . . . . . . . . . . . . . . . . . . 36 104 5.5. Prescriptive Versus Descriptive . . . . . . . . . . . . . 37 105 5.5.1. Intent-Based Networking . . . . . . . . . . . . . . . 37 106 5.6. Open-Loop Versus Closed-Loop . . . . . . . . . . . . . . 37 107 5.7. Tactical vs Strategic . . . . . . . . . . . . . . . . . . 37 108 6. Recommendations for Internet Traffic Engineering . . . . . . 38 109 6.1. Generic Non-functional Recommendations . . . . . . . . . 38 110 6.2. Routing Recommendations . . . . . . . . . . . . . . . . . 40 111 6.3. Traffic Mapping Recommendations . . . . . . . . . . . . . 43 112 6.4. Measurement Recommendations . . . . . . . . . . . . . . . 43 113 6.5. Network Survivability . . . . . . . . . . . . . . . . . . 44 114 6.5.1. Survivability in MPLS Based Networks . . . . . . . . 46 115 6.5.2. Protection Option . . . . . . . . . . . . . . . . . . 48 116 6.6. Traffic Engineering in Diffserv Environments . . . . . . 48 117 6.7. Network Controllability . . . . . . . . . . . . . . . . . 50 118 7. Inter-Domain Considerations . . . . . . . . . . . . . . . . . 51 119 8. Overview of Contemporary TE Practices in Operational IP 120 Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 53 121 9. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 57 122 10. Security Considerations . . . . . . . . . . . . . . . . . . . 58 123 11. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 58 124 12. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 58 125 13. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 60 126 14. Informative References . . . . . . . . . . . . . . . . . . . 61 127 Appendix A. Historic Overview . . . . . . . . . . . . . . . . . 69 128 A.1. Traffic Engineering in Classical Telephone Networks . . . 69 129 A.2. Evolution of Traffic Engineering in Packet Networks . . . 70 130 A.2.1. Adaptive Routing in the ARPANET . . . . . . . . . . . 71 131 A.2.2. Dynamic Routing in the Internet . . . . . . . . . . . 71 132 A.2.3. ToS Routing . . . . . . . . . . . . . . . . . . . . . 72 133 A.2.4. Equal Cost Multi-Path . . . . . . . . . . . . . . . . 72 134 A.2.5. Nimrod . . . . . . . . . . . . . . . . . . . . . . . 73 135 A.3. Development of Internet Traffic Engineering . . . . . . . 73 136 A.3.1. Overlay Model . . . . . . . . . . . . . . . . . . . . 73 137 Appendix B. Overview of Traffic Engineering Related Work in 138 Other SDOs . . . . . . . . . . . . . . . . . . . . . 74 139 B.1. Overview of ITU Activities Related to Traffic Engineering 74 140 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 75 142 1. Introduction 144 This memo describes the principles of Internet traffic engineering. 145 The objective of the document is to articulate the general issues and 146 principles for Internet traffic engineering; and where appropriate to 147 provide recommendations, guidelines, and options for the development 148 of online and offline Internet traffic engineering capabilities and 149 support systems. 151 This document can aid service providers in devising and implementing 152 traffic engineering solutions for their networks. Networking 153 hardware and software vendors will also find this document helpful in 154 the development of mechanisms and support systems for the Internet 155 environment that support the traffic engineering function. 157 This document provides a terminology for describing and understanding 158 common Internet traffic engineering concepts. This document also 159 provides a taxonomy of known traffic engineering styles. In this 160 context, a traffic engineering style abstracts important aspects from 161 a traffic engineering methodology. Traffic engineering styles can be 162 viewed in different ways depending upon the specific context in which 163 they are used and the specific purpose which they serve. The 164 combination of styles and views results in a natural taxonomy of 165 traffic engineering systems. 167 Even though Internet traffic engineering is most effective when 168 applied end-to-end, the focus of this document is traffic engineering 169 within a given domain (such as an autonomous system). However, 170 because a preponderance of Internet traffic tends to originate in one 171 autonomous system and terminate in another, this document provides an 172 overview of aspects pertaining to inter-domain traffic engineering. 174 This work was first published as [RFC3272] in May 2002. This 175 document obsoletes [RFC3272] by making a complete update to bring the 176 text in line with best current practices for Internet traffic 177 engineering and to include references to the latest relevant work in 178 the IETF. 180 1.1. What is Internet Traffic Engineering? 182 The Internet exists in order to transfer information from source 183 nodes to destination nodes. Accordingly, one of the most significant 184 functions performed by the Internet is the routing of traffic from 185 ingress nodes to egress nodes. Therefore, one of the most 186 distinctive functions performed by Internet traffic engineering is 187 the control and optimization of the routing function, to steer 188 traffic through the network. 190 Internet traffic engineering is defined as that aspect of Internet 191 network engineering dealing with the issue of performance evaluation 192 and performance optimization of operational IP networks. Traffic 193 Engineering encompasses the application of technology and scientific 194 principles to the measurement, characterization, modeling, and 195 control of Internet traffic [RFC2702], [AWD2]. 197 Ultimately, it is the performance of the network as seen by end users 198 of network services that is truly paramount. The characteristics 199 visible to end users are the emergent properties of the network, 200 which are the characteristics of the network when viewed as a whole. 201 A central goal of the service provider, therefore, is to enhance the 202 emergent properties of the network while taking economic 203 considerations into account. This is accomplished by addressing 204 traffic oriented performance requirements, while utilizing network 205 resources economically and reliably. Traffic oriented performance 206 measures include delay, delay variation, packet loss, and throughput. 208 Internet traffic engineering responds to network events. Aspects of 209 capacity management respond at intervals ranging from days to years. 210 Routing control functions operate at intervals ranging from 211 milliseconds to days. Packet level processing functions operate at 212 very fine levels of temporal resolution, ranging from picoseconds to 213 milliseconds while responding to the real-time statistical behavior 214 of traffic. 216 Thus, the optimization aspects of traffic engineering can be viewed 217 from a control perspective and can be pro-active and/or reactive. In 218 the pro-active case, the traffic engineering control system takes 219 preventive action to obviate predicted unfavorable future network 220 states such as e.g. engineering a backup path. It may also take 221 perfective action to induce a more desirable state in the future. In 222 the reactive case, the control system responds correctively and 223 perhaps adaptively to events that have already transpired in the 224 network, such as routing after failure. 226 Another important objective of Internet traffic engineering is to 227 facilitate reliable network operations [RFC2702]. Reliable network 228 operations can be facilitated by providing mechanisms that enhance 229 network integrity and by embracing policies emphasizing network 230 survivability. This results in a minimization of the vulnerability 231 of the network to service outages arising from errors, faults, and 232 failures occurring within the infrastructure. 234 The optimization aspects of traffic engineering can be achieved 235 through capacity management and traffic management. As used in this 236 document, capacity management includes capacity planning, routing 237 control, and resource management. Network resources of particular 238 interest include link bandwidth, buffer space, and computational 239 resources. Likewise, as used in this document, traffic management 240 includes (1) nodal traffic control functions such as traffic 241 conditioning, queue management, scheduling, and (2) other functions 242 that regulate traffic flow through the network or that arbitrate 243 access to network resources between different packets or between 244 different traffic streams. 246 One major challenge of Internet traffic engineering is the 247 realization of automated control capabilities that adapt quickly and 248 cost effectively to significant changes in a network's state, while 249 still maintaining stability of the network. Results from performance 250 evaluation assessing the effectiveness of traffic engineering methods 251 can be used to identify existing problems, guide network re- 252 optimization, and aid in the prediction of potential future problems. 253 However this process can also be time consuming and may not be 254 suitable to act on sudden, ephemeral changes in the network. 256 Performance evaluation can be achieved in many different ways. The 257 most notable techniques include analytical methods, simulation, and 258 empirical methods based on measurements. 260 In genaral, traffic engineering comes in two flavors. Either as a 261 background process that constantly monitors traffic and optimize the 262 usage of resources to improve performance, or in form of a pre- 263 planned optimized traffic distribution that is considered optimal. 264 In the later case, any deviation from the optimum distribution (e.g., 265 caused by a fiber cut) is reverted upon repair without further 266 optimization. However, this form of traffic engineering heavily 267 relies upon the notion that the planned state of the network is 268 indeed optimal. Hence, in such a mode there are two levels of 269 traffic engineering: the TE-planning task to enable an optimum 270 traffic distribution, and the routing task keeping traffic flows 271 attached to the pre-planned distribution 273 As a general rule, traffic engineering concepts and mechanisms must 274 be sufficiently specific and well-defined to address known 275 requirements, but simultaneously flexible and extensible to 276 accommodate unforeseen future demands. 278 1.2. Scope 280 The scope of this document is intra-domain traffic engineering; that 281 is, traffic engineering within a given autonomous system in the 282 Internet. This document will discuss concepts pertaining to intra- 283 domain traffic control, including such issues as routing control, 284 micro and macro resource allocation, and the control coordination 285 problems that arise consequently. 287 This document describes and characterize techniques already in use or 288 in advanced development for Internet traffic engineering. The way 289 these techniques fit together will be discussed and scenarios in 290 which they are useful will be identified. 292 While this document considers various intra-domain traffic 293 engineering approaches, it focuses more on traffic engineering with 294 MPLS and GMPLS. Traffic engineering based upon manipulation of IGP 295 metrics is not addressed in detail. This topic may be addressed by 296 other working group documents. 298 Although the emphasis is on intra-domain traffic engineering, in 299 Section 7, an overview of the high level considerations pertaining to 300 inter-domain traffic engineering will be provided. Inter-domain 301 Internet traffic engineering is crucial to the performance 302 enhancement of the global Internet infrastructure. 304 Whenever possible, relevant requirements from existing IETF documents 305 and other sources will be incorporated by reference. 307 1.3. Terminology 309 This subsection provides terminology which is useful for Internet 310 traffic engineering. The definitions presented apply to this 311 document. These terms may have other meanings elsewhere. 313 Baseline analysis: A study conducted to serve as a baseline for 314 comparison to the actual behavior of the network. 316 Busy hour: A one hour period within a specified interval of time 317 (typically 24 hours) in which the traffic load in a network or 318 sub-network is greatest. 320 Bottleneck: A network element whose input traffic rate tends to be 321 greater than its output rate. 323 Congestion: A state of a network resource in which the traffic 324 incident on the resource exceeds its output capacity over an 325 interval of time. 327 Congestion avoidance: An approach to congestion management that 328 attempts to obviate the occurrence of congestion. 330 Congestion control: An approach to congestion management that 331 attempts to remedy congestion problems that have already occurred. 333 Constraint-based routing: A class of routing protocols that take 334 specified traffic attributes, network constraints, and policy 335 constraints into account when making routing decisions. 336 Constraint-based routing is applicable to traffic aggregates as 337 well as flows. It is a generalization of QoS routing. 339 Demand side congestion management: A congestion management scheme 340 that addresses congestion problems by regulating or conditioning 341 offered load. 343 Effective bandwidth: The minimum amount of bandwidth that can be 344 assigned to a flow or traffic aggregate in order to deliver 345 'acceptable service quality' to the flow or traffic aggregate. 347 Egress traffic: Traffic exiting a network or network element. 349 Hot-spot: A network element or subsystem which is in a state of 350 congestion. 352 Ingress traffic: Traffic entering a network or network element. 354 Inter-domain traffic: Traffic that originates in one Autonomous 355 system and terminates in another. 357 Loss network: A network that does not provide adequate buffering for 358 traffic, so that traffic entering a busy resource within the 359 network will be dropped rather than queued. 361 Metric: A parameter defined in terms of standard units of 362 measurement. 364 Measurement methodology: A repeatable measurement technique used to 365 derive one or more metrics of interest. 367 Network survivability: The capability to provide a prescribed level 368 of QoS for existing services after a given number of failures 369 occur within the network. 371 Offline traffic engineering: A traffic engineering system that 372 exists outside of the network. 374 Online traffic engineering: A traffic engineering system that exists 375 within the network, typically implemented on or as adjuncts to 376 operational network elements. 378 Performance measures: Metrics that provide quantitative or 379 qualitative measures of the performance of systems or subsystems 380 of interest. 382 Performance management: A systematic approach to improving 383 effectiveness in the accomplishment of specific networking goals 384 related to performance improvement. 386 Performance metric: A performance parameter defined in terms of 387 standard units of measurement. 389 Provisioning: The process of assigning or configuring network 390 resources to meet certain requests. 392 QoS routing: Class of routing systems that selects paths to be used 393 by a flow based on the QoS requirements of the flow. 395 Service Level Agreement (SLA): A contract between a provider and a 396 customer that guarantees specific levels of performance and 397 reliability at a certain cost. 399 Service Level Objective (SLO): A key element of an SLA between a 400 provider and a customer. SLOs are agreed upon as a means of 401 measuring the performance of the Service Provider and are outlined 402 as a way of avoiding disputes between the two parties based on 403 misunderstanding. 405 Stability: An operational state in which a network does not 406 oscillate in a disruptive manner from one mode to another mode. 408 Supply-side congestion management: A congestion management scheme 409 that provisions additional network resources to address existing 410 and/or anticipated congestion problems. 412 Transit traffic: Traffic whose origin and destination are both 413 outside of the network under consideration. 415 Traffic characteristic: A description of the temporal behavior or a 416 description of the attributes of a given traffic flow or traffic 417 aggregate. 419 Traffic engineering system: A collection of objects, mechanisms, and 420 protocols that are used conjunctively to accomplish traffic 421 engineering objectives. 423 Traffic flow: A stream of packets between two end-points that can be 424 characterized in a certain way. A micro-flow has a more specific 425 definition A micro-flow is a stream of packets with the same 426 source and destination addresses, source and destination ports, 427 and protocol ID. 429 Traffic intensity: A measure of traffic loading with respect to a 430 resource capacity over a specified period of time. In classical 431 telephony systems, traffic intensity is measured in units of 432 Erlangs. 434 Traffic matrix: A representation of the traffic demand between a set 435 of origin and destination abstract nodes. An abstract node can 436 consist of one or more network elements. 438 Traffic monitoring: The process of observing traffic characteristics 439 at a given point in a network and collecting the traffic 440 information for analysis and further action. 442 Traffic trunk: An aggregation of traffic flows belonging to the same 443 class which are forwarded through a common path. A traffic trunk 444 may be characterized by an ingress and egress node, and a set of 445 attributes which determine its behavioral characteristics and 446 requirements from the network. 448 2. Background 450 The Internet must convey IP packets from ingress nodes to egress 451 nodes efficiently, expeditiously, and economically. Furthermore, in 452 a multiclass service environment (e.g., Diffserv capable networks), 453 the resource sharing parameters of the network must be appropriately 454 determined and configured according to prevailing policies and 455 service models to resolve resource contention issues arising from 456 mutual interference between packets traversing through the network. 457 Thus, consideration must be given to resolving competition for 458 network resources between traffic streams belonging to the same 459 service class (intra-class contention resolution) and traffic streams 460 belonging to different classes (inter-class contention resolution). 462 2.1. Context of Internet Traffic Engineering 464 The context of Internet traffic engineering pertains to the scenarios 465 where traffic engineering is used. A traffic engineering methodology 466 establishes appropriate rules to resolve traffic performance issues 467 occurring in a specific context. The context of Internet traffic 468 engineering includes: 470 1. A network context defining the universe of discourse, and in 471 particular the situations in which the traffic engineering 472 problems occur. The network context includes network structure, 473 network policies, network characteristics, network constraints, 474 network quality attributes, and network optimization criteria. 476 2. A problem context defining the general and concrete issues that 477 traffic engineering addresses. The problem context includes 478 identification, abstraction of relevant features, representation, 479 formulation, specification of the requirements on the solution 480 space, and specification of the desirable features of acceptable 481 solutions. 483 3. A solution context suggesting how to address the issues 484 identified by the problem context. The solution context includes 485 analysis, evaluation of alternatives, prescription, and 486 resolution. 488 4. An implementation and operational context in which the solutions 489 are methodologically instantiated. The implementation and 490 operational context includes planning, organization, and 491 execution. 493 The context of Internet traffic engineering and the different problem 494 scenarios are discussed in the following subsections. 496 2.2. Network Context 498 IP networks range in size from small clusters of routers situated 499 within a given location, to thousands of interconnected routers, 500 switches, and other components distributed all over the world. 502 Conceptually, at the most basic level of abstraction, an IP network 503 can be represented as a distributed dynamical system consisting of: 505 o a set of interconnected resources which provide transport services 506 for IP traffic subject to certain constraints 508 o a demand system representing the offered load to be transported 509 through the network 511 o a response system consisting of network processes, protocols, and 512 related mechanisms which facilitate the movement of traffic 513 through the network (see also [AWD2]). 515 The network elements and resources may have specific characteristics 516 restricting the manner in which the demand is handled. Additionally, 517 network resources may be equipped with traffic control mechanisms 518 superintending the way in which the demand is serviced. Traffic 519 control mechanisms may, for example, be used to: 521 o control various packet processing activities within a given 522 resource 524 o arbitrate contention for access to the resource by different 525 packets 527 o regulate traffic behavior through the resource. 529 A configuration management and provisioning system may allow the 530 settings of the traffic control mechanisms to be manipulated by 531 external or internal entities in order to exercise control over the 532 way in which the network elements respond to internal and external 533 stimuli. 535 The details of how the network provides transport services for 536 packets are specified in the policies of the network administrators 537 and are installed through network configuration management and policy 538 based provisioning systems. Generally, the types of services 539 provided by the network also depends upon the technology and 540 characteristics of the network elements and protocols, the prevailing 541 service and utility models, and the ability of the network 542 administrators to translate policies into network configurations. 544 Contemporary Internet networks have three significant 545 characteristics: 547 o they provide real-time services 549 o they have become mission critical 551 o their operating environments are very dynamic. 553 The dynamic characteristics of IP and IP/MPLS networks can be 554 attributed in part to fluctuations in demand, to the interaction 555 between various network protocols and processes, to the rapid 556 evolution of the infrastructure which demands the constant inclusion 557 of new technologies and new network elements, and to transient and 558 persistent impairments which occur within the system. 560 Packets contend for the use of network resources as they are conveyed 561 through the network. A network resource is considered to be 562 congested if, for an interval of time, the arrival rate of packets 563 exceed the output capacity of the resource. Congestion may result in 564 some of the arrival packets being delayed or even dropped. 566 Congestion increases transit delays, delay variation, packet loss, 567 and reduces the predictability of network services. Clearly, 568 congestion is highly undesirable. 570 Combating congestion at a reasonable cost is a major objective of 571 Internet traffic engineering. 573 Efficient sharing of network resources by multiple traffic streams is 574 a basic operatoinal premise for packet switched networks in general 575 and for the Internet in particular. A fundamental challenge in 576 network operation, especially in a large scale public IP network, is 577 to increase the efficiency of resource utilization while minimizing 578 the possibility of congestion. 580 The Internet will have to function in the presence of different 581 classes of traffic with different service requirements. RFC 2475 582 provides an architecture for Differentiated Services and makes this 583 requirement clear [RFC2475]. The RFC allows packets to be grouped 584 into behavior aggregates such that each aggregate has a common set of 585 behavioral characteristics or a common set of delivery requirements. 586 Delivery requirements of a specific set of packets may be specified 587 explicitly or implicitly. Two of the most important traffic delivery 588 requirements are capacity constraints and QoS constraints. 590 Capacity constraints can be expressed statistically as peak rates, 591 mean rates, burst sizes, or as some deterministic notion of effective 592 bandwidth. QoS requirements can be expressed in terms of: 594 o integrity constraints such as packet loss 596 o in terms of temporal constraints such as timing restrictions for 597 the delivery of each packet (delay) and timing restrictions for 598 the delivery of consecutive packets belonging to the same traffic 599 stream (delayvariation). 601 2.3. Problem Context 603 There are several large problems in association with operating a 604 network described by the simple model of the previous subsection. 605 This subsection analyze the problem context in relation to traffic 606 engineering. 608 The identification, abstraction, representation, and measurement of 609 network features relevant to traffic engineering are significant 610 issues. 612 A particular challenge is to explicitly formulate the problems that 613 traffic engineering attempts to solve. For example: 615 o how to identify the requirements on the solution space 617 o how to specify the desirable features of good solutions 619 o how to actually solve the problems 620 o how to measure and characterize the effectiveness of the 621 solutions. 623 Another class of problems is how to measure and estimate relevant 624 network state parameters. Effective traffic engineering relies on a 625 good estimate of the offered traffic load as well as a view of the 626 underlying topology and associated resource constraints. A network- 627 wide view of the topology is also a must for offline planning. 629 Still another class of problems is how to characterize the state of 630 the network and how to evaluate its performance under a variety of 631 scenarios. The performance evaluation problem is two- fold. One 632 aspect of this problem relates to the evaluation of the system-level 633 performance of the network. The other aspect relates to the 634 evaluation of the resource-level performance, which restricts 635 attention to the performance analysis of individual network 636 resources. 638 In this document, we refer to the system-level characteristics of the 639 network as the "macro-states" and the resource-level characteristics 640 as the "micro-states." The system-level characteristics are also 641 known as the emergent properties of the network. Correspondingly, we 642 shall refer to the traffic engineering schemes dealing with network 643 performance optimization at the systems level as "macro-TE" and the 644 schemes that optimize at the individual resource level as "micro-TE." 645 Under certain circumstances, the system-level performance can be 646 derived from the resource-level performance using appropriate rules 647 of composition, depending upon the particular performance measures of 648 interest. 650 Another fundamental class of problems concerns how to effectively 651 optimize network performance. Performance optimization may entail 652 translating solutions to specific traffic engineering problems into 653 network configurations. Optimization may also entail some degree of 654 resource management control, routing control, and/or capacity 655 augmentation. 657 As noted previously, congestion is an undesirable phenomena in 658 operational networks. Therefore, the next subsection addresses the 659 issue of congestion and its ramifications within the problem context 660 of Internet traffic engineering. 662 2.3.1. Congestion and its Ramifications 664 Congestion is one of the most significant problems in an operational 665 IP context. A network element is said to be congested if it 666 experiences sustained overload over an interval of time. Congestion 667 almost always results in degradation of service quality to end users. 669 Congestion control schemes can include demand-side policies and 670 supply-side policies. Demand-side policies may restrict access to 671 congested resources and/or dynamically regulate the demand to 672 alleviate the overload situation. Supply-side policies may expand or 673 augment network capacity to better accommodate offered traffic. 674 Supply-side policies may also re-allocate network resources by 675 redistributing traffic over the infrastructure. Traffic 676 redistribution and resource re-allocation serve to increase the 677 'effective capacity' seen by the demand. 679 The emphasis of this document is primarily on congestion management 680 schemes falling within the scope of the network, rather than on 681 congestion management systems dependent upon sensitivity and 682 adaptivity from end-systems. That is, the aspects that are 683 considered in this document with respect to congestion management are 684 those solutions that can be provided by control entities operating on 685 the network and by the actions of network administrators and network 686 operations systems. 688 2.4. Solution Context 690 The solution context for Internet traffic engineering involves 691 analysis, evaluation of alternatives, and choice between alternative 692 courses of action. Generally the solution context is based on making 693 reasonable inferences about the current or future state of the 694 network, and subsequently making appropriate decisions that may 695 involve a preference between alternative sets of action. More 696 specifically, the solution context demands reasonable estimates of 697 traffic workload, characterization of network state, deriving 698 solutions to traffic engineering problems which may be implicitly or 699 explicitly formulated, and possibly instantiating a set of control 700 actions. Control actions may involve the manipulation of parameters 701 associated with routing, control over tactical capacity acquisition, 702 and control over the traffic management functions. 704 The following list of instruments may be applicable to the solution 705 context of Internet traffic engineering. 707 o A set of policies, objectives, and requirements (which may be 708 context dependent) for network performance evaluation and 709 performance optimization. 711 o A collection of online and possibly offline tools and mechanisms 712 for measurement, characterization, modeling, and control of 713 Internet traffic and control over the placement and allocation of 714 network resources, as well as control over the mapping or 715 distribution of traffic onto the infrastructure. 717 o A set of constraints on the operating environment, the network 718 protocols, and the traffic engineering system itself. 720 o A set of quantitative and qualitative techniques and methodologies 721 for abstracting, formulating, and solving traffic engineering 722 problems. 724 o A set of administrative control parameters which may be 725 manipulated through a Configuration Management (CM) system. The 726 CM system itself may include a configuration control subsystem, a 727 configuration repository, a configuration accounting subsystem, 728 and a configuration auditing subsystem. 730 o A set of guidelines for network performance evaluation, 731 performance optimization, and performance improvement. 733 Determining traffic characteristics through measurement and/or 734 estimation is very useful within the realm the traffic engineering 735 solution space. Traffic estimates can be derived from customer 736 subscription information, traffic projections, traffic models, and 737 from actual measurements. The measurements may be performed at 738 different levels, e.g. the traffic-aggregate level or at the flow 739 level. Measuring at different levels is done in order to aquire 740 traffic statistics at more or less detail. Measurements at the flow 741 level or on small traffic aggregates may be performed at edge nodes, 742 when traffic enters and leaves the network. Measurements for large 743 traffic-aggregates may be performed within the core of the network. 745 To conduct performance studies and to support planning of existing 746 and future networks, a routing analysis may be performed to determine 747 the paths the routing protocols will choose for various traffic 748 demands, and to ascertain the utilization of network resources as 749 traffic is routed through the network. The routing analysis should 750 capture the selection of paths through the network, the assignment of 751 traffic across multiple feasible routes, and the multiplexing of IP 752 traffic over traffic trunks (if such constructs exists) and over the 753 underlying network infrastructure. A network topology model is a 754 necessity for routing analysis. A network topology model may be 755 extracted from: 757 o network architecture documents 759 o network designs 761 o information contained in router configuration files 763 o routing databases 764 o routing tables 766 o automated tools that discover and depict network topology 767 information. 769 Topology information may also be derived from servers that monitor 770 network state, and from servers that perform provisioning functions. 772 Routing in operational IP networks can be administratively controlled 773 at various levels of abstraction including the manipulation of BGP 774 attributes and IGP metrics. For path oriented technologies such as 775 MPLS, routing can be further controlled by the manipulation of 776 relevant traffic engineering parameters, resource parameters, and 777 administrative policy constraints. Within the context of MPLS, the 778 path of an explicitly routed label switched path (LSP) can be 779 computed and established in various ways including: 781 o manually 783 o automatically online using constraint-based routing processes 784 implemented on label switching routers 786 o automatically offline using constraint-based routing entities 787 implemented on external traffic engineering support systems. 789 2.4.1. Combating the Congestion Problem 791 Minimizing congestion is a significant aspect of Internet traffic 792 engineering. This subsection gives an overview of the general 793 approaches that have been used or proposed to combat congestion 794 problems. 796 Congestion management policies can be categorized based upon the 797 following criteria (see e.g., [YARE95] for a more detailed taxonomy 798 of congestion control schemes): 800 o Response time scale which can be characterized as long, medium, or 801 short 803 o reactive versus preventive which relates to congestion control and 804 congestion avoidance 806 o supply side versus demand side congestion management schemes. 808 These aspects are discussed in the following paragraphs. 810 1. Congestion Management based on Response Time Scales 811 * Long (weeks to months): Expanding network capacity by adding 812 new equipement, routers and links, takes time and is 813 comparatively costly. Capacity planning needs to take this 814 into consideration. Network capacity is expanded based on 815 estimates or forcasts of future traffic development and 816 traffic distribution. These upgrades are typically carried 817 out over weeks or months, or maybe even years. 819 * Medium (minutes to days): Several control policies fall within 820 the medium timescale category. Examples include: 822 a. Adjusting IGP and/or BGP parameters to route traffic away 823 or towards certain segments of the network 825 b. Setting up and/or adjusting some explicitly routed LSPs in 826 MPLS networks to route some traffic trunks away from 827 possibly congested resources or toward possibly more 828 favorable routes 830 c. Re-configuring the logical topology of the network to make 831 it correlate more closely with the spatial traffic 832 distribution using for example some underlying path- 833 oriented technology such as MPLS LSPs or optical channel 834 trails. 836 Many of these adaptive medium time scale response schemes rely 837 on a measurement system. The measurement system monitors 838 changes in traffic distribution, traffic shifts, and network 839 resource utilization. The measurement system then provides 840 feedback to the online and/or offline traffic engineering 841 mechanisms and tools which employ this feedback information to 842 trigger certain control actions to occur within the network. 843 The traffic engineering mechanisms and tools can be 844 implemented in a distributed or centralized fashion, and may 845 have a hierarchical or flat structure. The comparative merits 846 of distributed and centralized control structures for networks 847 are well known. A centralized scheme may have global 848 visibility into the network state and may produce potentially 849 more optimal solutions. However, centralized schemes are 850 prone to single points of failure and may not scale as well as 851 distributed schemes. Moreover, the information utilized by a 852 centralized scheme may be stale and may not reflect the actual 853 state of the network. It is not an objective of this memo to 854 make a recommendation between distributed and centralized 855 schemes. This is a choice that network administrators must 856 make based on their specific needs. 858 * Short (picoseconds to minutes): This category includes packet 859 level processing functions and events on the order of several 860 round trip times. It includes router mechanisms such as 861 passive and active buffer management. These mechanisms are 862 used to control congestion and/or signal congestion to end 863 systems so that they can adaptively regulate the rate at which 864 traffic is injected into the network. One of the most popular 865 active queue management schemes, especially for TCP traffic, 866 is Random Early Detection (RED) [FLJA93]. RED supports 867 congestion avoidance by controlling the average queue size. 868 During congestion (but before the queue is filled), the RED 869 scheme chooses arriving packets to "mark" according to a 870 probabilistic algorithm which takes into account the average 871 queue size. For a router that does not utilize explicit 872 congestion notification (ECN) see e.g., [FLOY94], the marked 873 packets can simply be dropped to signal the inception of 874 congestion to end systems. On the other hand, if the router 875 supports ECN, then it can set the ECN field in the packet 876 header. Several variations of RED have been proposed to 877 support different drop precedence levels in multi-class 878 environments [RFC2597], e.g., RED with In and Out (RIO) and 879 Weighted RED. There is general consensus that RED provides 880 congestion avoidance performance which is not worse than 881 traditional Tail-Drop (TD) queue management (drop arriving 882 packets only when the queue is full). Importantly, however, 883 RED reduces the possibility of global synchronization and 884 improves fairness among different TCP sessions. However, RED 885 by itself can not prevent congestion and unfairness caused by 886 sources unresponsive to RED, e.g., UDP traffic and some 887 misbehaved greedy connections. Other schemes have been 888 proposed to improve the performance and fairness in the 889 presence of unresponsive traffic. Some of these schemes were 890 proposed as theoretical frameworks and are typically not 891 available in existing commercial products. Two such schemes 892 are Longest Queue Drop (LQD) and Dynamic Soft Partitioning 893 with Random Drop (RND) [SLDC98]. 895 2. Congestion Management: Reactive versus Preventive Schemes 897 * Reactive: Reactive (recovery) congestion management policies 898 react to existing congestion problems to improve it. All the 899 policies described in the long and medium time scales above 900 can be categorized as being reactive especially if the 901 policies are based on monitoring and identifying existing 902 congestion problems, and on the initiation of relevant actions 903 to ease a situation. 905 * Preventive: Preventive (predictive/avoidance) policies take 906 proactive action to prevent congestion based on estimates and 907 predictions of future potential congestion problems. Some of 908 the policies described in the long and medium time scales fall 909 into this category. They do not necessarily respond 910 immediately to existing congestion problems. Instead 911 forecasts of traffic demand and workload distribution are 912 considered and action may be taken to prevent potential 913 congestion problems in the future. The schemes described in 914 the short time scale (e.g., RED and its variations, ECN, LQD, 915 and RND) are also used for congestion avoidance since dropping 916 or marking packets before queues actually overflow would 917 trigger corresponding TCP sources to slow down. 919 3. Congestion Management: Supply-Side versus Demand-Side Schemes 921 * Supply-side: Supply-side congestion management policies 922 increase the effective capacity available to traffic in order 923 to control or reduce congestion. This can be accomplished by 924 increasing capacity. Another way to accomplish this is to 925 minimize congestion by having a relatively balanced 926 distribution of traffic over the network. For example, 927 capacity planning should aim to provide a physical topology 928 and associated link bandwidths that match estimated traffic 929 workload and traffic distribution. This may be based on 930 forecasting and subject to budgetary or other constraints. If 931 actual traffic distribution does not match the topology 932 derived from capacity panning, then the traffic can be mapped 933 onto the existing topology using routing control mechanisms, 934 using path oriented technologies (e.g., MPLS LSPs and optical 935 channel trails) to modify the logical topology, or by using 936 some other load redistribution mechanisms. 938 * Demand-side: Demand-side congestion management policies 939 control or regulate the offered traffic to alleviate 940 congestion problems. For example, some of the short time 941 scale mechanisms described earlier (such as RED and its 942 variations, ECN, LQD, and RND) as well as policing and rate- 943 shaping mechanisms attempt to regulate the offered load in 944 various ways. Tariffs may also be applied as a demand side 945 instrument. To date, however, tariffs have not been used as a 946 means of demand-side congestion management within the 947 Internet. 949 In summary, a variety of mechanisms can be used to address congestion 950 problems in IP networks. These mechanisms may operate at multiple 951 time-scales and at multiple traffic aggregation levels. 953 2.5. Implementation and Operational Context 955 The operational context of Internet traffic engineering is 956 characterized by constant changes which occur at multiple levels of 957 abstraction. The implementation context demands effective planning, 958 organization, and execution. The planning aspects may involve 959 determining prior sets of actions to achieve desired objectives. 960 Organizing involves arranging and assigning responsibility to the 961 various components of the traffic engineering system and coordinating 962 the activities to accomplish the desired TE objectives. Execution 963 involves measuring and applying corrective or perfective actions to 964 attain and maintain desired TE goals. 966 3. Traffic Engineering Process Models 968 This section describes a generic process model that captures the 969 high-level practical aspects of Internet traffic engineering in an 970 operational context. The process model is described as a sequence of 971 actions that a traffic engineer, or more generally a traffic 972 engineering system, must perform to optimize the performance of an 973 operational network (see also [RFC2702], AWD2]). This process model 974 may be enacted explicitly or implicitly, by an automaton and/or by a 975 human. 977 The traffic engineering process model is iterative [AWD2]. The four 978 phases of the process model described below are repeated continually. 980 o Define the relevant control policies that govern the operation of 981 the network. 983 o A feedback mechanism involving the acquisition of measurement data 984 from the operational network. 986 o Analyze the network state and to characterize traffic workload. 987 Performance analysis may be proactive and/or reactive. Proactive 988 performance analysis identifies potential problems that do not 989 exist, but could manifest in the future. Reactive performance 990 analysis identifies existing problems, determines their cause 991 through diagnosis, and valuates alternative approaches to remedy 992 the problem, if necessary. 994 o Performance optimization of the network. It involves a decision 995 process which selects and implements a set of actions from a set 996 of alternatives. Optimization actions may include the use of 997 appropriate techniques to either control the offered traffic or to 998 control the distribution of traffic across the network. 1000 3.1. Components of the Traffic Engineering Process Model 1002 The key components of the traffic engineering process model are: 1004 1. Measurement is crucial to the traffic engineering function. The 1005 operational state of a network can be conclusively determined 1006 only through measurement. Measurement is also critical to the 1007 optimization function because it provides feedback data which is 1008 used by traffic engineering control subsystems. This data is 1009 used to adaptively optimize network performance in response to 1010 events and stimuli originating within and outside the network. 1011 Measurement in support of the TE function can occur at different 1012 levels of abstraction. For example, measurement can be used to 1013 derive packet level characteristics, flow level characteristics, 1014 user or customer level characteristics, traffic aggregate 1015 characteristics, component level characteristics, and network 1016 wide characteristics. 1018 2. Modeling, analysis, and simulation are important aspects of 1019 Internet traffic engineering. Modeling involves constructing an 1020 abstract or physical representation which depicts relevant 1021 traffic characteristics and network attributes. A network model 1022 is an abstract representation of the network which captures 1023 relevant network features, attributes, and characteristic. 1024 Network simulation tools are extremely useful for traffic 1025 engineering. Because of the complexity of realistic quantitative 1026 analysis of network behavior, certain aspects of network 1027 performance studies can only be conducted effectively using 1028 simulation. 1030 3. Network performance optimization involves resolving network 1031 issues by transforming such issues into concepts that enable a 1032 solution, identification of a solution, and implementation of the 1033 solution. Network performance optimization can be corrective or 1034 perfective. In corrective optimization, the goal is to remedy a 1035 problem that has occurred or that is incipient. In perfective 1036 optimization, the goal is to improve network performance even 1037 when explicit problems do not exist and are not anticipated. 1039 4. Review of TE Techniques 1041 This section briefly reviews different traffic engineering approaches 1042 proposed and implemented in telecommunications and computer networks. 1043 The discussion is not intended to be comprehensive. It is primarily 1044 intended to illuminate pre-existing perspectives and prior art 1045 concerning traffic engineering in the Internet and in legacy 1046 telecommunications networks. A historic overview is provided in 1047 Appendix A. 1049 4.1. Overview of IETF Projects Related to Traffic Engineering 1051 This subsection reviews a number of IETF activities pertinent to 1052 Internet traffic engineering. These activities are primarily 1053 intended to evolve the IP architecture to support new service 1054 definitions which allow preferential or differentiated treatment to 1055 be accorded to certain types of traffic. 1057 4.1.1. Constraint-Based Routing 1059 Constraint-based routing refers to a class of routing systems that 1060 compute routes through a network subject to the satisfaction of a set 1061 of constraints and requirements. In the most general setting, 1062 constraint-based routing may also seek to optimize overall network 1063 performance while minimizing costs. 1065 The constraints and requirements may be imposed by the network itself 1066 or by administrative policies. Constraints may include bandwidth, 1067 hop count, delay, and policy instruments such as resource class 1068 attributes. Constraints may also include domain specific attributes 1069 of certain network technologies and contexts which impose 1070 restrictions on the solution space of the routing function. Path 1071 oriented technologies such as MPLS have made constraint-based routing 1072 feasible and attractive in public IP networks. 1074 The concept of constraint-based routing within the context of MPLS 1075 traffic engineering requirements in IP networks was first described 1076 in [RFC2702] and led to developments such as MPLS-TE [RFC3209] as 1077 described in Section 4.1.5. 1079 Unlike QoS routing (for example, see [RFC2386] and [MA]) which 1080 generally addresses the issue of routing individual traffic flows to 1081 satisfy prescribed flow based QoS requirements subject to network 1082 resource availability, constraint-based routing is applicable to 1083 traffic aggregates as well as flows and may be subject to a wide 1084 variety of constraints which may include policy restrictions. 1086 4.1.2. Integrated Services 1088 The IETF Integrated Services working group developed the integrated 1089 services (Intserv) model. This model requires resources, such as 1090 bandwidth and buffers, to be reserved a priori for a given traffic 1091 flow to ensure that the quality of service requested by the traffic 1092 flow is satisfied. The integrated services model includes additional 1093 components beyond those used in the best-effort model such as packet 1094 classifiers, packet schedulers, and admission control. A packet 1095 classifier is used to identify flows that are to receive a certain 1096 level of service. A packet scheduler handles the scheduling of 1097 service to different packet flows to ensure that QoS commitments are 1098 met. Admission control is used to determine whether a router has the 1099 necessary resources to accept a new flow. 1101 The main issue with the Integrated Services model has been 1102 scalability [RFC2998], especially in large public IP networks which 1103 may potentially have millions of active micro-flows in transit 1104 concurrently. 1106 A notable feature of the Integrated Services model is that it 1107 requires explicit signaling of QoS requirements from end systems to 1108 routers [RFC2753]. The Resource Reservation Protocol (RSVP) performs 1109 this signaling function and is a critical component of the Integrated 1110 Services model. RSVP is described next. 1112 4.1.3. RSVP 1114 RSVP is a soft state signaling protocol [RFC2205]. It supports 1115 receiver initiated establishment of resource reservations for both 1116 multicast and unicast flows. RSVP was originally developed as a 1117 signaling protocol within the integrated services framework for 1118 applications to communicate QoS requirements to the network and for 1119 the network to reserve relevant resources to satisfy the QoS 1120 requirements [RFC2205]. 1122 Under RSVP, the sender or source node sends a PATH message to the 1123 receiver with the same source and destination addresses as the 1124 traffic which the sender will generate. The PATH message contains: 1125 (1) a sender Tspec specifying the characteristics of the traffic, (2) 1126 a sender Template specifying the format of the traffic, and (3) an 1127 optional Adspec which is used to support the concept of One Pass With 1128 Advertising (OPWA) [RFC2205]. Every intermediate router along the 1129 path forwards the PATH Message to the next hop determined by the 1130 routing protocol. Upon receiving a PATH Message, the receiver 1131 responds with a RESV message which includes a flow descriptor used to 1132 request resource reservations. The RESV message travels to the 1133 sender or source node in the opposite direction along the path that 1134 the PATH message traversed. Every intermediate router along the path 1135 can reject or accept the reservation request of the RESV message. If 1136 the request is rejected, the rejecting router will send an error 1137 message to the receiver and the signaling process will terminate. If 1138 the request is accepted, link bandwidth and buffer space are 1139 allocated for the flow and the related flow state information is 1140 installed in the router. 1142 One of the issues with the original RSVP specification was 1143 Scalability. This is because reservations were required for micro- 1144 flows, so that the amount of state maintained by network elements 1145 tends to increase linearly with the number of micro-flows. These 1146 issues are described in [RFC2961]. 1148 Recently, RSVP has been modified and extended in several ways to 1149 mitigate the scaling problems. As a result, it is becoming a 1150 versatile signaling protocol for the Internet. For example, RSVP has 1151 been extended to reserve resources for aggregation of flows, to set 1152 up MPLS explicit label switched paths, and to perform other signaling 1153 functions within the Internet. There are also a number of proposals 1154 to reduce the amount of refresh messages required to maintain 1155 established RSVP sessions [RFC2961]. 1157 A number of IETF working groups have been engaged in activities 1158 related to the RSVP protocol. These include the original RSVP 1159 working group, the MPLS working group, the Resource Allocation 1160 Protocol working group, and the Policy Framework working group. 1162 4.1.4. Differentiated Services 1164 The goal of the Differentiated Services (Diffserv) effort within the 1165 IETF is to devise scalable mechanisms for categorization of traffic 1166 into behavior aggregates, which ultimately allows each behavior 1167 aggregate to be treated differently, especially when there is a 1168 shortage of resources such as link bandwidth and buffer space 1169 [RFC2475]. One of the primary motivations for the Diffserv effort 1170 was to devise alternative mechanisms for service differentiation in 1171 the Internet that mitigate the scalability issues encountered with 1172 the Intserv model. 1174 The IETF Diffserv working group has defined a Differentiated Services 1175 field in the IP header (DS field). The DS field consists of six bits 1176 of the part of the IP header formerly known as the TOS octet. The DS 1177 field is used to indicate the forwarding treatment that a packet 1178 should receive at a node [RFC2474]. The Diffserv working group has 1179 also standardized a number of Per-Hop Behavior (PHB) groups. Using 1180 the PHBs, several classes of services can be defined using different 1181 classification, policing, shaping, and scheduling rules. 1183 For an end-user of network services to receive Differentiated 1184 Services from its Internet Service Provider (ISP), it may be 1185 necessary for the user to have a Service Level Agreement (SLA) with 1186 the ISP. An SLA may explicitly or implicitly specify a Traffic 1187 Conditioning Agreement (TCA) which defines classifier rules as well 1188 as metering, marking, discarding, and shaping rules. 1190 Packets are classified, and possibly policed and shaped at the 1191 ingress to a Diffserv network. When a packet traverses the boundary 1192 between different Diffserv domains, the DS field of the packet may be 1193 re-marked according to existing agreements between the domains. 1195 Differentiated Services allows only a finite number of service 1196 classes to be specified by the DS field. The main advantage of the 1197 Diffserv approach relative to the Intserv model is scalability. 1198 Resources are allocated on a per-class basis and the amount of state 1199 information is proportional to the number of classes rather than to 1200 the number of application flows. 1202 It should be obvious from the previous discussion that the Diffserv 1203 model essentially deals with traffic management issues on a per hop 1204 basis. The Diffserv control model consists of a collection of micro- 1205 TE control mechanisms. Other traffic engineering capabilities, such 1206 as capacity management (including routing control), are also required 1207 in order to deliver acceptable service quality in Diffserv networks. 1208 The concept of Per Domain Behaviors has been introduced to better 1209 capture the notion of differentiated services across a complete 1210 domain [RFC3086]. 1212 4.1.5. MPLS 1214 MPLS is an advanced forwarding scheme which also includes extensions 1215 to conventional IP control plane protocols. MPLS extends the 1216 Internet routing model and enhances packet forwarding and path 1217 control [RFC3031]. 1219 At the ingress to an MPLS domain, Label Switching Routers (LSRs) 1220 classify IP packets into Forwarding Equivalence Classes (FECs) based 1221 on a variety of factors, including, e.g., a combination of the 1222 information carried in the IP header of the packets and the local 1223 routing information maintained by the LSRs. An MPLS label stack 1224 entry is then prepended to each packet according to their forwarding 1225 equivalence classes. The MPLS label stack entry is 32 bits long and 1226 contains a 20-bit label field. 1228 An LSR makes forwarding decisions by using the label prepended to 1229 packets as the index into a local next hop label forwarding entry 1230 (NHLFE). The packet is then processed as specified in the NHLFE. 1231 The incoming label may be replaced by an outgoing label (label swap), 1232 and the packet may be forwarded to the next LSR. Before a packet 1233 leaves an MPLS domain, its MPLS label may be removed (label pop). A 1234 Label Switched Path (LSP) is the path between an ingress LSRs and an 1235 egress LSRs through which a labeled packet traverses. The path of an 1236 explicit LSP is defined at the originating (ingress) node of the LSP. 1237 MPLS can use a signaling protocol such as RSVP or LDP to set up LSPs. 1239 MPLS is a very powerful technology for Internet traffic engineering 1240 because it supports explicit LSPs which allow constraint-based 1241 routing to be implemented efficiently in IP networks [AWD2]. The 1242 requirements for traffic engineering over MPLS are described in 1243 [RFC2702]. Extensions to RSVP to support instantiation of explicit 1244 LSP are discussed in [RFC3209]. 1246 4.1.6. Generalized MPLS 1248 GMPLS extends MPLS control protocols to encompass time-division 1249 (e.g., SONET/SDH, PDH, G.709), wavelength (lambdas), and spatial 1250 switching (e.g., incoming port or fiber to outgoing port or fiber) as 1251 well as continuing to support packet switching. GMPLS provides a 1252 common set of control protocols for all of these layers (including 1253 some technology-specific extensions) each of which has a diverse data 1254 or forwarding plane. GMPLS covers both the signaling and the routing 1255 part of that control plane and is based on the Traffic Engineering 1256 extensions to MPLS (see Section 4.1.5). 1258 In GMPLS, the original MPLS architecture is extended to include LSRs 1259 whose forwarding planes rely on circuit switching, and therefore 1260 cannot forward data based on the information carried in either packet 1261 or cell headers. Specifically, such LSRs include devices where the 1262 switching is based on time slots, wavelengths, or physical ports. 1263 These additions impact basic LSP properties: how labels are requested 1264 and communicated, the unidirectional nature of MPLS LSPs, how errors 1265 are propagated, and information provided for synchronizing the 1266 ingress and egress LSRs. 1268 4.1.7. IP Performance Metrics 1270 The IETF IP Performance Metrics (IPPM) working group has been 1271 developing a set of standard metrics that can be used to monitor the 1272 quality, performance, and reliability of Internet services. These 1273 metrics can be applied by network operators, end-users, and 1274 independent testing groups to provide users and service providers 1275 with a common understanding of the performance and reliability of the 1276 Internet component 'clouds' they use/provide [RFC2330]. The criteria 1277 for performance metrics developed by the IPPM WG are described in 1278 [RFC2330]. Examples of performance metrics include one-way packet 1279 loss [RFC7680], one-way delay [RFC7679], and connectivity measures 1280 between two nodes [RFC2678]. Other metrics include second-order 1281 measures of packet loss and delay. 1283 Some of the performance metrics specified by the IPPM WG are useful 1284 for specifying Service Level Agreements (SLAs). SLAs are sets of 1285 service level objectives negotiated between users and service 1286 providers, wherein each objective is a combination of one or more 1287 performance metrics, possibly subject to certain constraints. 1289 4.1.8. Flow Measurement 1291 The IETF Real Time Flow Measurement (RTFM) working group has produced 1292 an architecture document defining a method to specify traffic flows 1293 as well as a number of components for flow measurement (meters, meter 1294 readers, manager) [RFC2722]. A flow measurement system enables 1295 network traffic flows to be measured and analyzed at the flow level 1296 for a variety of purposes. As noted in RFC 2722, a flow measurement 1297 system can be very useful in the following contexts: 1299 o understanding the behavior of existing networks 1301 o planning for network development and expansion 1303 o quantification of network performance 1305 o verifying the quality of network service 1307 o attribution of network usage to users. 1309 A flow measurement system consists of meters, meter readers, and 1310 managers. A meter observes packets passing through a measurement 1311 point, classifies them into certain groups, accumulates certain usage 1312 data (such as the number of packets and bytes for each group), and 1313 stores the usage data in a flow table. A group may represent a user 1314 application, a host, a network, a group of networks, etc. A meter 1315 reader gathers usage data from various meters so it can be made 1316 available for analysis. A manager is responsible for configuring and 1317 controlling meters and meter readers. The instructions received by a 1318 meter from a manager include flow specification, meter control 1319 parameters, and sampling techniques. The instructions received by a 1320 meter reader from a manager include the address of the meter whose 1321 date is to be collected, the frequency of data collection, and the 1322 types of flows to be collected. 1324 4.1.9. Endpoint Congestion Management 1326 [RFC3124] is intended to provide a set of congestion control 1327 mechanisms that transport protocols can use. It is also intended to 1328 develop mechanisms for unifying congestion control across a subset of 1329 an endpoint's active unicast connections (called a congestion group). 1330 A congestion manager continuously monitors the state of the path for 1331 each congestion group under its control. The manager uses that 1332 information to instruct a scheduler on how to partition bandwidth 1333 among the connections of that congestion group. 1335 4.1.10. TE Extensions to the IGPs 1337 TBD 1339 4.1.11. Link-State BGP 1341 In a number of environments, a component external to a network is 1342 called upon to perform computations based on the network topology and 1343 current state of the connections within the network, including 1344 traffic engineering information. This is information typically 1345 distributed by IGP routing protocols within the network (see 1346 Section 4.1.10. 1348 The Border Gateway Protocol (BGP) Section 7 is one of the essential 1349 routing protocols that glue the Internet together. BGP Link State 1350 (BGP-LS) [RFC7752] is a mechanism by which link-state and traffic 1351 engineering information can be collected from networks and shared 1352 with external components using the BGP routing protocol. The 1353 mechanism is applicable to physical and virtual IGP links, and is 1354 subject to policy control. 1356 Information collected by BGP-LS can be used to construct the Traffic 1357 Engineering Database (TED, see Section 4.1.17) for use by the Path 1358 Computation Element (PCE, see Section 4.1.12), or may be used by 1359 Application-Layer Traffic Optimization (ALTO) servers (see 1360 Section 4.1.13). 1362 4.1.12. Path Computation Element 1364 Constraint-based path computation is a fundamental building block for 1365 traffic engineering in MPLS and GMPLS networks. Path computation in 1366 large, multi-domain networks is complex and may require special 1367 computational components and cooperation between the elements in 1368 different domains. The Path Computation Element (PCE) [RFC4655] is 1369 an entity (component, application, or network node) that is capable 1370 of computing a network path or route based on a network graph and 1371 applying computational constraints. 1373 Thus, a PCE can provide a central component in a traffic engineering 1374 system operating on the Traffic Engineering Database (TED, see 1375 Section 4.1.17) with delegated responsibility for determining paths 1376 in MPLS, GMPLS, or Segment Routing networks. The PCE uses the Path 1377 Computation Element Communication Protocol (PCEP) [RFC5440] to 1378 communicate with Path Computation Clients (PCCs), such as MPLS LSRs, 1379 to answer their requests for computed paths or to instruct them to 1380 initiate new paths [RFC8281] and maintain state about paths already 1381 installed in the network [RFC8231]. 1383 PCEs form key components of a number of traffic engineering systems, 1384 such as the Application of the Path Computation Element Architecture 1385 [RFC6805], the Applicability of a Stateful Path Computation Element 1386 [RFC8051], Abstraction and Control of TE Networks (ACTN) 1387 Section 4.1.15, Centralized Network Control [RFC8283], and Software 1388 Defined Networking (SDN) Section 5.3.2. 1390 4.1.13. Application-Layer Traffic Optimization 1392 TBD 1394 4.1.14. Segment Routing with MPLS encapsuation (SR-MPLS) 1396 Segment Routing (SR) leverages the source routing and tunneling 1397 paradigms: The path packet takes is defined at the ingress and 1398 tunneled to the egress. 1400 A node steers a packet through a controlled set of instructions, 1401 called segments, by prepending the packet with an SR header, label 1402 stack in MPLS case. 1404 A segment can represent any instruction, topological or service- 1405 based, thanks to the MPLS architecture [RFC3031]. Labels cand be 1406 looked up in a global context (platform wide) as well as in some 1407 other context (see "context labels" in section 3 of [RFC5331]). 1409 4.1.14.1. Base Segment Routing Identifier Types 1411 Segments are identified by Segment Identifiers (SIDs). There are 1412 four types of SID that are relevant for traffic engineering. 1414 Prefix SID: Uses SR Global Block (SRGB), must be unique within the 1415 routing domain SRGB, and is advertised by an IGP. The Prefix-SID 1416 can be configured as an absolute value or an index. 1418 Node SID: A Node SID is a prefix SID with the 'N' (node) bit set, it 1419 is associated with a host prefix (/32 or /128) that identifies the 1420 node. More than 1 Node SID can be configured per node. 1422 Adjacency SID: An Adjacency SID is locally significant (by default). 1423 It can be made globally significant through use of the 'L' flag. 1424 It identifies unidirectional adjacency. In most implementations 1425 Adjacency SIDs are automatically allocated for each adjacency. 1426 They are always encoded as an absolute (not indexed) value. 1428 Binding SID: A Binding SID has two purposes 1430 1. Mapping Server in ISIS 1431 ISIS:The SID/Label Binding TLV is used to advertise 1432 prefixes to SID/Label mappings. This functionality is 1433 called the Segment Routing Mapping Server (SRMS). The 1434 behavior of the SRMS is defined in [RFC8661] 1436 2. Cross-connect (label to FEC mapping) 1438 This is fundamental for multi-domain/multi-layer operation. 1439 The Binding SID identifies a new (could be SR or 1440 hierarchical, at another OSI Layer) path available at the 1441 anchor point. Is always local to the originator (must not 1442 be at the top of the stack), must be looked up in the 1443 context of the nodal SID. It could be provisioned through 1444 Netconf/Restconf, PCEP, BGP, or the CLI. 1446 4.1.15. Network Virtualization and Abstraction 1448 One of the main drivers for Software-Defined Networking (SDN) 1449 [RFC7149] is a decoupling of the network control plane from the data 1450 plane. This separation has been achieved for TE networks with the 1451 development of MPLS/GMPLS [RFC3945] and the Path Computation Element 1452 (PCE) [RFC4655]. One of the advantages of SDN is its logically 1453 centralized control regime that allows a global view of the 1454 underlying networks. Centralized control in SDN helps improve 1455 network resource utilization compared with distributed network 1456 control. 1458 Abstraction and Control of TE networks (ACTN) [RFC8453] defines an 1459 hierarchical SDN architecture which describes the functional entities 1460 and methods for the coordination of resources across multiple 1461 domains, to provide end-to-end traffic engineered services. ACTN 1462 facilitates end-to-end connections and provides them to the user. 1463 ACTN is focused on aspects like abstraction, virtualization and 1464 presentation. In particular it deals with: 1466 o Abstraction of the underlying network resources and how they are 1467 provided to higher-layer applications and customers. 1469 o Virtualization of underlying resources, whose selection criterion 1470 is the allocation of those resources for the customer, 1471 application, or service. The creation of a virtualized 1472 environment allowis operators to view and control multi-domain 1473 networks as a single virtualized network. 1475 o Presentation to customers of networks as a virtual network via 1476 open and programmable interfaces. 1478 The ACTN managed infrastructure are traffic engineered network 1479 resources, which may include statistical packet bandwidth, physical 1480 forwarding plane sources (such as wavelengths and time slots), 1481 forwarding and cross connect capabilities. The ACTN type of network 1482 virtualization provides customers and applications (tenants) to 1483 utilise and independently control allocated virtual network resources 1484 as if resources as if they were physically their own resource. The 1485 ACTN network is "sliced", with tenants being given a different 1486 partial and abstracted topology view of the physical underlying 1487 network. 1489 4.1.16. Deterministic Networking 1491 TBD 1493 4.1.17. Network TE State Definition and Presentation 1495 The network states that are relevant to the traffic engineering need 1496 to be stored in the system and presented to the user. The Traffic 1497 Engineering Database (TED) is a collection of all TE information 1498 about all TE nodes and TE links in the network, which is an essential 1499 component of a TE system, such as MPLS-TE [RFC2702] and GMPLS 1500 [RFC3945]. In order to formally define the data in the TED and to 1501 present the data to the user with high usability, the data modeling 1502 language YANG [RFC7950] can be used as described in 1503 [I-D.ietf-teas-yang-te-topo]. 1505 4.1.18. System Management and Control Interfaces 1507 The traffic engineering control system needs to have a management 1508 interface that is human-friendly and a control interfaces that is 1509 programable for automation. The Network Configuration Protocol 1510 (NETCONF) [RFC6241] or the RESTCONF Protocol [RFC8040] provide 1511 programmable interfaces that are also human-friendly. These 1512 protocols use XML or JSON encoded messages. When message compactness 1513 or protocol bandwidth consumption needs to be optimized for the 1514 control interface, other protocols, such as Group Communication for 1515 the Constrained Application Protocol (CoAP) [RFC7390] or gRPC, are 1516 available, especially when the protocol messages are encoded in a 1517 binary format. Along with any of these protocols, the data modeling 1518 language YANG [RFC7950] can be used to formally and precisely define 1519 the interface data. 1521 The Path Computation Element Communication Protocol (PCEP) [RFC5440] 1522 is another protocol that has evolved to be an option for the TE 1523 system control interface. The messages of PCEP are TLV-based, not 1524 defined by a data modeling language such as YANG. 1526 4.2. Content Distribution 1528 The Internet is dominated by client-server interactions, especially 1529 Web traffic (in the future, more sophisticated media servers may 1530 become dominant). The location and performance of major information 1531 servers has a significant impact on the traffic patterns within the 1532 Internet as well as on the perception of service quality by end 1533 users. 1535 A number of dynamic load balancing techniques have been devised to 1536 improve the performance of replicated information servers. These 1537 techniques can cause spatial traffic characteristics to become more 1538 dynamic in the Internet because information servers can be 1539 dynamically picked based upon the location of the clients, the 1540 location of the servers, the relative utilization of the servers, the 1541 relative performance of different networks, and the relative 1542 performance of different parts of a network. This process of 1543 assignment of distributed servers to clients is called Traffic 1544 Directing. It is an application layer function. 1546 Traffic Directing schemes that allocate servers in multiple 1547 geographically dispersed locations to clients may require empirical 1548 network performance statistics to make more effective decisions. In 1549 the future, network measurement systems may need to provide this type 1550 of information. The exact parameters needed are not yet defined. 1552 When congestion exists in the network, Traffic Directing and Traffic 1553 Engineering systems should act in a coordinated manner. This topic 1554 is for further study. 1556 The issues related to location and replication of information 1557 servers, particularly web servers, are important for Internet traffic 1558 engineering because these servers contribute a substantial proportion 1559 of Internet traffic. 1561 5. Taxonomy of Traffic Engineering Systems 1563 This section presents a short taxonomy of traffic engineering 1564 systems. A taxonomy of traffic engineering systems can be 1565 constructed based on traffic engineering styles and views as listed 1566 below: 1568 o Time-dependent vs State-dependent vs Event-dependent 1570 o Offline vs Online 1572 o Centralized vs Distributed 1573 o Local vs Global Information 1575 o Prescriptive vs Descriptive 1577 o Open Loop vs Closed Loop 1579 o Tactical vs Strategic 1581 These classification systems are described in greater detail in the 1582 following subsections of this document. 1584 5.1. Time-Dependent Versus State-Dependent Versus Event Dependent 1586 Traffic engineering methodologies can be classified as time- 1587 dependent, or state-dependent, or event-dependent. All TE schemes 1588 are considered to be dynamic in this document. Static TE implies 1589 that no traffic engineering methodology or algorithm is being 1590 applied. 1592 In the time-dependent TE, historical information based on periodic 1593 variations in traffic, (such as time of day), is used to pre-program 1594 routing plans and other TE control mechanisms. Additionally, 1595 customer subscription or traffic projection may be used. Pre- 1596 programmed routing plans typically change on a relatively long time 1597 scale (e.g., diurnal). Time-dependent algorithms do not attempt to 1598 adapt to random variations in traffic or changing network conditions. 1599 An example of a time-dependent algorithm is a global centralized 1600 optimizer where the input to the system is a traffic matrix and 1601 multi-class QoS requirements as described [MR99]. Another example of 1602 such a methodology is the application of data mining to Internet 1603 traffic [AJ19]. Data mining enables the use of various machine 1604 learning algorithms to identify patterns within historically 1605 collected datasets about Internet traffic, and to extract information 1606 in order to guide decision-making, and to improve efficiency and 1607 productivity of operational processes. 1609 State-dependent TE adapts the routing plans for packets based on the 1610 current state of the network. The current state of the network 1611 provides additional information on variations in actual traffic 1612 (i.e., perturbations from regular variations) that could not be 1613 predicted using historical information. Constraint-based routing is 1614 an example of state-dependent TE operating in a relatively long time 1615 scale. An example operating in a relatively short timescale is a 1616 load-balancing algorithm described in [MATE]. 1618 The state of the network can be based on parameters such as 1619 utilization, packet delay, packet loss, etc. These parameters can be 1620 obtained in several ways. For example, each router may flood these 1621 parameters periodically or by means of some kind of trigger to other 1622 routers. Another approach is for a particular router performing 1623 adaptive TE to send probe packets along a path to gather the state of 1624 that path. [RFC6374] defines protocol extensions to collect 1625 performance measurements from MPLS networks. Another approach is for 1626 a management system to gather the relevant information directly from 1627 network elements using telemetry data collection "publication/ 1628 subscription" techniques [RFC7923]. 1630 Expeditious and accurate gathering and distribution of state 1631 information is critical for adaptive TE due to the dynamic nature of 1632 network conditions. State-dependent algorithms may be applied to 1633 increase network efficiency and resilience. Time-dependent 1634 algorithms are more suitable for predictable traffic variations. On 1635 the other hand, state-dependent algorithms are more suitable for 1636 adapting to the prevailing network state. 1638 Event-dependent TE methods can also be used for TE path selection. 1639 Event-dependent TE methods are distinct from time-dependent and 1640 state-dependent TE methods in the manner in which paths are selected. 1641 These algorithms are adaptive and distributed in nature and typically 1642 use learning models to find good paths for TE in a network. While 1643 state-dependent TE models typically use available-link-bandwidth 1644 (ALB) flooding for TE path selection, event-dependent TE methods do 1645 not require ALB flooding. Rather, event-dependent TE methods 1646 typically search out capacity by learning models, as in the success- 1647 to-the-top (STT) method. ALB flooding can be resource intensive, 1648 since it requires link bandwidth to carry LSAs, processor capacity to 1649 process LSAs, and the overhead can limit area/Autonomous System (AS) 1650 size. Modeling results suggest that event-dependent TE methods could 1651 lead to a reduction in ALB flooding overhead without loss of network 1652 throughput performance [I-D.ietf-tewg-qos-routing]. 1654 5.2. Offline Versus Online 1656 Traffic engineering requires the computation of routing plans. The 1657 computation may be performed offline or online. The computation can 1658 be done offline for scenarios where routing plans need not be 1659 executed in real-time. For example, routing plans computed from 1660 forecast information may be computed offline. Typically, offline 1661 computation is also used to perform extensive searches on multi- 1662 dimensional solution spaces. 1664 Online computation is required when the routing plans must adapt to 1665 changing network conditions as in state-dependent algorithms. Unlike 1666 offline computation (which can be computationally demanding), online 1667 computation is geared toward relative simple and fast calculations to 1668 select routes, fine-tune the allocations of resources, and perform 1669 load balancing. 1671 5.3. Centralized Versus Distributed 1673 Centralized control has a central authority which determines routing 1674 plans and perhaps other TE control parameters on behalf of each 1675 router. The central authority collects the network-state information 1676 from all routers periodically and returns the routing information to 1677 the routers. The routing update cycle is a critical parameter 1678 directly impacting the performance of the network being controlled. 1679 Centralized control may need high processing power and high bandwidth 1680 control channels. 1682 Distributed control determines route selection by each router 1683 autonomously based on the routers view of the state of the network. 1684 The network state information may be obtained by the router using a 1685 probing method or distributed by other routers on a periodic basis 1686 using link state advertisements. Network state information may also 1687 be disseminated under exceptional conditions. Examples of protocol 1688 extensions used to advertise network link state information are 1689 defined in [RFC5305], [RFC6119], [RFC7471], [RFC7810], and [RFC8571]. 1691 5.3.1. Hybrid Systems 1693 TBD 1695 5.3.2. Considerations for Software Defined Networking 1697 TBD 1699 5.4. Local Versus Global 1701 Traffic engineering algorithms may require local or global network- 1702 state information. 1704 Local information pertains to the state of a portion of the domain. 1705 Examples include the bandwidth and packet loss rate of a particular 1706 path. Local state information may be sufficient for certain 1707 instances of distributed-controlled TEs. 1709 Global information pertains to the state of the entire domain 1710 undergoing traffic engineering. Examples include a global traffic 1711 matrix and loading information on each link throughout the domain of 1712 interest. Global state information is typically required with 1713 centralized control. Distributed TE systems may also need global 1714 information in some cases. 1716 5.5. Prescriptive Versus Descriptive 1718 TE systems may also be classified as prescriptive or descriptive. 1720 Prescriptive traffic engineering evaluates alternatives and 1721 recommends a course of action. Prescriptive traffic engineering can 1722 be further categorized as either corrective or perfective. 1723 Corrective TE prescribes a course of action to address an existing or 1724 predicted anomaly. Perfective TE prescribes a course of action to 1725 evolve and improve network performance even when no anomalies are 1726 evident. 1728 Descriptive traffic engineering, on the other hand, characterizes the 1729 state of the network and assesses the impact of various policies 1730 without recommending any particular course of action. 1732 5.5.1. Intent-Based Networking 1734 TBD 1736 5.6. Open-Loop Versus Closed-Loop 1738 Open-loop traffic engineering control is where control action does 1739 not use feedback information from the current network state. The 1740 control action may use its own local information for accounting 1741 purposes, however. 1743 Closed-loop traffic engineering control is where control action 1744 utilizes feedback information from the network state. The feedback 1745 information may be in the form of historical information or current 1746 measurement. 1748 5.7. Tactical vs Strategic 1750 Tactical traffic engineering aims to address specific performance 1751 problems (such as hot-spots) that occur in the network from a 1752 tactical perspective, without consideration of overall strategic 1753 imperatives. Without proper planning and insights, tactical TE tends 1754 to be ad hoc in nature. 1756 Strategic traffic engineering approaches the TE problem from a more 1757 organized and systematic perspective, taking into consideration the 1758 immediate and longer term consequences of specific policies and 1759 actions. 1761 6. Recommendations for Internet Traffic Engineering 1763 This section describes high-level recommendations for traffic 1764 engineering in the Internet. These recommendations are presented in 1765 general terms. 1767 The recommendations describe the capabilities needed to solve a 1768 traffic engineering problem or to achieve a traffic engineering 1769 objective. Broadly speaking, these recommendations can be 1770 categorized as either functional or non-functional recommendations. 1772 Functional recommendations for Internet traffic engineering describe 1773 the functions that a traffic engineering system should perform. 1774 These functions are needed to realize traffic engineering objectives 1775 by addressing traffic engineering problems. 1777 Non-functional recommendations for Internet traffic engineering 1778 relate to the quality attributes or state characteristics of a 1779 traffic engineering system. These recommendations may contain 1780 conflicting assertions and may sometimes be difficult to quantify 1781 precisely. 1783 6.1. Generic Non-functional Recommendations 1785 The generic non-functional recommendations for Internet traffic 1786 engineering include: usability, automation, scalability, stability, 1787 visibility, simplicity, efficiency, reliability, correctness, 1788 maintainability, extensibility, interoperability, and security. In a 1789 given context, some of these recommendations may be critical while 1790 others may be optional. Therefore, prioritization may be required 1791 during the development phase of a traffic engineering system (or 1792 components thereof) to tailor it to a specific operational context. 1794 In the following paragraphs, some of the aspects of the non- 1795 functional recommendations for Internet traffic engineering are 1796 summarized. 1798 Usability: Usability is a human factor aspect of traffic engineering 1799 systems. Usability refers to the ease with which a traffic 1800 engineering system can be deployed and operated. In general, it is 1801 desirable to have a TE system that can be readily deployed in an 1802 existing network. It is also desirable to have a TE system that is 1803 easy to operate and maintain. 1805 Automation: Whenever feasible, a traffic engineering system should 1806 automate as many traffic engineering functions as possible to 1807 minimize the amount of human effort needed to control and analyze 1808 operational networks. Automation is particularly imperative in large 1809 scale public networks because of the high cost of the human aspects 1810 of network operations and the high risk of network problems caused by 1811 human errors. Automation may entail the incorporation of automatic 1812 feedback and intelligence into some components of the traffic 1813 engineering system. 1815 Scalability: Contemporary public networks are growing very fast with 1816 respect to network size and traffic volume. Therefore, a TE system 1817 should be scalable to remain applicable as the network evolves. In 1818 particular, a TE system should remain functional as the network 1819 expands with regard to the number of routers and links, and with 1820 respect to the traffic volume. A TE system should have a scalable 1821 architecture, should not adversely impair other functions and 1822 processes in a network element, and should not consume too much 1823 network resources when collecting and distributing state information 1824 or when exerting control. 1826 Stability: Stability is a very important consideration in traffic 1827 engineering systems that respond to changes in the state of the 1828 network. State-dependent traffic engineering methodologies typically 1829 mandate a tradeoff between responsiveness and stability. It is 1830 strongly recommended that when tradeoffs are warranted between 1831 responsiveness and stability, that the tradeoff should be made in 1832 favor of stability (especially in public IP backbone networks). 1834 Flexibility: A TE system should be flexible to allow for changes in 1835 optimization policy. In particular, a TE system should provide 1836 sufficient configuration options so that a network administrator can 1837 tailor the TE system to a particular environment. It may also be 1838 desirable to have both online and offline TE subsystems which can be 1839 independently enabled and disabled. TE systems that are used in 1840 multi-class networks should also have options to support class based 1841 performance evaluation and optimization. 1843 Visibility: As part of the TE system, mechanisms should exist to 1844 collect statistics from the network and to analyze these statistics 1845 to determine how well the network is functioning. Derived statistics 1846 such as traffic matrices, link utilization, latency, packet loss, and 1847 other performance measures of interest which are determined from 1848 network measurements can be used as indicators of prevailing network 1849 conditions. Other examples of status information which should be 1850 observed include existing functional routing information 1851 (additionally, in the context of MPLS existing LSP routes), etc. 1853 Simplicity: Generally, a TE system should be as simple as possible. 1854 More importantly, the TE system should be relatively easy to use 1855 (i.e., clean, convenient, and intuitive user interfaces). Simplicity 1856 in user interface does not necessarily imply that the TE system will 1857 use naive algorithms. When complex algorithms and internal 1858 structures are used, such complexities should be hidden as much as 1859 possible from the network administrator through the user interface. 1861 Interoperability: Whenever feasible, traffic engineering systems and 1862 their components should be developed with open standards based 1863 interfaces to allow interoperation with other systems and components. 1865 Security: Security is a critical consideration in traffic engineering 1866 systems. Such traffic engineering systems typically exert control 1867 over certain functional aspects of the network to achieve the desired 1868 performance objectives. Therefore, adequate measures must be taken 1869 to safeguard the integrity of the traffic engineering system. 1870 Adequate measures must also be taken to protect the network from 1871 vulnerabilities that originate from security breaches and other 1872 impairments within the traffic engineering system. 1874 The remainder of this section will focus on some of the high-level 1875 functional recommendations for traffic engineering. 1877 6.2. Routing Recommendations 1879 Routing control is a significant aspect of Internet traffic 1880 engineering. Routing impacts many of the key performance measures 1881 associated with networks, such as throughput, delay, and utilization. 1882 Generally, it is very difficult to provide good service quality in a 1883 wide area network without effective routing control. A desirable 1884 routing system is one that takes traffic characteristics and network 1885 constraints into account during route selection while maintaining 1886 stability. 1888 Traditional shortest path first (SPF) interior gateway protocols are 1889 based on shortest path algorithms and have limited control 1890 capabilities for traffic engineering [RFC2702], [AWD2]. These 1891 limitations include: 1893 1. The well known issues with pure SPF protocols, which do not take 1894 network constraints and traffic characteristics into account 1895 during route selection. For example, since IGPs always use the 1896 shortest paths (based on administratively assigned link metrics) 1897 to forward traffic, load sharing cannot be accomplished among 1898 paths of different costs. Using shortest paths to forward 1899 traffic conserves network resources, but may cause the following 1900 problems: 1) If traffic from a source to a destination exceeds 1901 the capacity of a link along the shortest path, the link (hence 1902 the shortest path) becomes congested while a longer path between 1903 these two nodes may be under-utilized; 2) the shortest paths from 1904 different sources can overlap at some links. If the total 1905 traffic from the sources exceeds the capacity of any of these 1906 links, congestion will occur. Problems can also occur because 1907 traffic demand changes over time but network topology and routing 1908 configuration cannot be changed as rapidly. This causes the 1909 network topology and routing configuration to become sub-optimal 1910 over time, which may result in persistent congestion problems. 1912 2. The Equal-Cost Multi-Path (ECMP) capability of SPF IGPs supports 1913 sharing of traffic among equal cost paths between two nodes. 1914 However, ECMP attempts to divide the traffic as equally as 1915 possible among the equal cost shortest paths. Generally, ECMP 1916 does not support configurable load sharing ratios among equal 1917 cost paths. The result is that one of the paths may carry 1918 significantly more traffic than other paths because it may also 1919 carry traffic from other sources. This situation can result in 1920 congestion along the path that carries more traffic. 1922 3. Modifying IGP metrics to control traffic routing tends to have 1923 network-wide effect. Consequently, undesirable and unanticipated 1924 traffic shifts can be triggered as a result. Recent work 1925 described in Section 8 may be capable of better control [FT00], 1926 [FT01]. 1928 Because of these limitations, new capabilities are needed to enhance 1929 the routing function in IP networks. Some of these capabilities have 1930 been described elsewhere and are summarized below. 1932 Constraint-based routing is desirable to evolve the routing 1933 architecture of IP networks, especially public IP backbones with 1934 complex topologies [RFC2702]. Constraint-based routing computes 1935 routes to fulfill requirements subject to constraints. Constraints 1936 may include bandwidth, hop count, delay, and administrative policy 1937 instruments such as resource class attributes [RFC2702], [RFC2386]. 1938 This makes it possible to select routes that satisfy a given set of 1939 requirements subject to network and administrative policy 1940 constraints. Routes computed through constraint-based routing are 1941 not necessarily the shortest paths. Constraint-based routing works 1942 best with path oriented technologies that support explicit routing, 1943 such as MPLS. 1945 Constraint-based routing can also be used as a way to redistribute 1946 traffic onto the infrastructure (even for best effort traffic). For 1947 example, if the bandwidth requirements for path selection and 1948 reservable bandwidth attributes of network links are appropriately 1949 defined and configured, then congestion problems caused by uneven 1950 traffic distribution may be avoided or reduced. In this way, the 1951 performance and efficiency of the network can be improved. 1953 A number of enhancements are needed to conventional link state IGPs, 1954 such as OSPF and IS-IS, to allow them to distribute additional state 1955 information required for constraint-based routing. These extensions 1956 to OSPF were described in [RFC3630] and to IS-IS in [RFC5305]. 1957 Essentially, these enhancements require the propagation of additional 1958 information in link state advertisements. Specifically, in addition 1959 to normal link-state information, an enhanced IGP is required to 1960 propagate topology state information needed for constraint-based 1961 routing. Some of the additional topology state information include 1962 link attributes such as reservable bandwidth and link resource class 1963 attribute (an administratively specified property of the link). The 1964 resource class attribute concept was defined in [RFC2702]. The 1965 additional topology state information is carried in new TLVs and sub- 1966 TLVs in IS-IS, or in the Opaque LSA in OSPF [RFC5305], [RFC3630]. 1968 An enhanced link-state IGP may flood information more frequently than 1969 a normal IGP. This is because even without changes in topology, 1970 changes in reservable bandwidth or link affinity can trigger the 1971 enhanced IGP to initiate flooding. A tradeoff is typically required 1972 between the timeliness of the information flooded and the flooding 1973 frequency to avoid excessive consumption of link bandwidth and 1974 computational resources, and more importantly, to avoid instability. 1976 In a TE system, it is also desirable for the routing subsystem to 1977 make the load splitting ratio among multiple paths (with equal cost 1978 or different cost) configurable. This capability gives network 1979 administrators more flexibility in the control of traffic 1980 distribution across the network. It can be very useful for avoiding/ 1981 relieving congestion in certain situations. Examples can be found in 1982 [XIAO]. 1984 The routing system should also have the capability to control the 1985 routes of subsets of traffic without affecting the routes of other 1986 traffic if sufficient resources exist for this purpose. This 1987 capability allows a more refined control over the distribution of 1988 traffic across the network. For example, the ability to move traffic 1989 from a source to a destination away from its original path to another 1990 path (without affecting other traffic paths) allows traffic to be 1991 moved from resource-poor network segments to resource-rich segments. 1992 Path oriented technologies such as MPLS inherently support this 1993 capability as discussed in [AWD2]. 1995 Additionally, the routing subsystem should be able to select 1996 different paths for different classes of traffic (or for different 1997 traffic behavior aggregates) if the network supports multiple classes 1998 of service (different behavior aggregates). 2000 6.3. Traffic Mapping Recommendations 2002 Traffic mapping pertains to the assignment of traffic workload onto 2003 pre-established paths to meet certain requirements. Thus, while 2004 constraint-based routing deals with path selection, traffic mapping 2005 deals with the assignment of traffic to established paths which may 2006 have been selected by constraint-based routing or by some other 2007 means. Traffic mapping can be performed by time-dependent or state- 2008 dependent mechanisms, as described in Section 5.1. 2010 An important aspect of the traffic mapping function is the ability to 2011 establish multiple paths between an originating node and a 2012 destination node, and the capability to distribute the traffic 2013 between the two nodes across the paths according to some policies. A 2014 pre-condition for this scheme is the existence of flexible mechanisms 2015 to partition traffic and then assign the traffic partitions onto the 2016 parallel paths. This requirement was noted in [RFC2702]. When 2017 traffic is assigned to multiple parallel paths, it is recommended 2018 that special care should be taken to ensure proper ordering of 2019 packets belonging to the same application (or micro-flow) at the 2020 destination node of the parallel paths. 2022 As a general rule, mechanisms that perform the traffic mapping 2023 functions should aim to map the traffic onto the network 2024 infrastructure to minimize congestion. If the total traffic load 2025 cannot be accommodated, or if the routing and mapping functions 2026 cannot react fast enough to changing traffic conditions, then a 2027 traffic mapping system may rely on short time scale congestion 2028 control mechanisms (such as queue management, scheduling, etc.) to 2029 mitigate congestion. Thus, mechanisms that perform the traffic 2030 mapping functions should complement existing congestion control 2031 mechanisms. In an operational network, it is generally desirable to 2032 map the traffic onto the infrastructure such that intra-class and 2033 inter-class resource contention are minimized. 2035 When traffic mapping techniques that depend on dynamic state feedback 2036 (e.g., MATE and such like) are used, special care must be taken to 2037 guarantee network stability. 2039 6.4. Measurement Recommendations 2041 The importance of measurement in traffic engineering has been 2042 discussed throughout this document. Mechanisms should be provided to 2043 measure and collect statistics from the network to support the 2044 traffic engineering function. Additional capabilities may be needed 2045 to help in the analysis of the statistics. The actions of these 2046 mechanisms should not adversely affect the accuracy and integrity of 2047 the statistics collected. The mechanisms for statistical data 2048 acquisition should also be able to scale as the network evolves. 2050 Traffic statistics may be classified according to long-term or short- 2051 term timescales. Long-term timescale traffic statistics are very 2052 useful for traffic engineering. Long-term time scale traffic 2053 statistics may capture or reflect periodicity in network workload 2054 (such as hourly, daily, and weekly variations in traffic profiles) as 2055 well as traffic trends. Aspects of the monitored traffic statistics 2056 may also depict class of service characteristics for a network 2057 supporting multiple classes of service. Analysis of the long-term 2058 traffic statistics may yield secondary statistics such as busy hour 2059 characteristics, traffic growth patterns, persistent congestion 2060 problems, hot-spot, and imbalances in link utilization caused by 2061 routing anomalies. 2063 A mechanism for constructing traffic matrices for both long-term and 2064 short-term traffic statistics should be in place. In multi-service 2065 IP networks, the traffic matrices may be constructed for different 2066 service classes. Each element of a traffic matrix represents a 2067 statistic of traffic flow between a pair of abstract nodes. An 2068 abstract node may represent a router, a collection of routers, or a 2069 site in a VPN. 2071 Measured traffic statistics should provide reasonable and reliable 2072 indicators of the current state of the network on the short-term 2073 scale. Some short term traffic statistics may reflect link 2074 utilization and link congestion status. Examples of congestion 2075 indicators include excessive packet delay, packet loss, and high 2076 resource utilization. Examples of mechanisms for distributing this 2077 kind of information include SNMP, probing techniques, FTP, IGP link 2078 state advertisements, etc. 2080 6.5. Network Survivability 2082 Network survivability refers to the capability of a network to 2083 maintain service continuity in the presence of faults. This can be 2084 accomplished by promptly recovering from network impairments and 2085 maintaining the required QoS for existing services after recovery. 2086 Survivability has become an issue of great concern within the 2087 Internet community due to the increasing demands to carry mission 2088 critical traffic, real-time traffic, and other high priority traffic 2089 over the Internet. Survivability can be addressed at the device 2090 level by developing network elements that are more reliable; and at 2091 the network level by incorporating redundancy into the architecture, 2092 design, and operation of networks. It is recommended that a 2093 philosophy of robustness and survivability should be adopted in the 2094 architecture, design, and operation of traffic engineering that 2095 control IP networks (especially public IP networks). Because 2096 different contexts may demand different levels of survivability, the 2097 mechanisms developed to support network survivability should be 2098 flexible so that they can be tailored to different needs. A number 2099 of tools and techniques have been developed to enable network 2100 survivability including MPLS Fast Reroute [RFC4090], RSVP-TE 2101 Extensions in Support of End-to-End Generalized Multi-Protocol Label 2102 Switching (GMPLS) Recovery [RFC4872], and GMPLS Segment Recovery 2103 [RFC4873]. 2105 Failure protection and restoration capabilities have become available 2106 from multiple layers as network technologies have continued to 2107 improve. At the bottom of the layered stack, optical networks are 2108 now capable of providing dynamic ring and mesh restoration 2109 functionality at the wavelength level as well as traditional 2110 protection functionality. At the SONET/SDH layer survivability 2111 capability is provided with Automatic Protection Switching (APS) as 2112 well as self-healing ring and mesh architectures. Similar 2113 functionality is provided by layer 2 technologies such as ATM 2114 (generally with slower mean restoration times). Rerouting is 2115 traditionally used at the IP layer to restore service following link 2116 and node outages. Rerouting at the IP layer occurs after a period of 2117 routing convergence which may require seconds to minutes to complete. 2118 Some new developments in the MPLS context make it possible to achieve 2119 recovery at the IP layer prior to convergence [RFC3469]. 2121 To support advanced survivability requirements, path-oriented 2122 technologies such a MPLS can be used to enhance the survivability of 2123 IP networks in a potentially cost effective manner. The advantages 2124 of path oriented technologies such as MPLS for IP restoration becomes 2125 even more evident when class based protection and restoration 2126 capabilities are required. 2128 Recently, a common suite of control plane protocols has been proposed 2129 for both MPLS and optical transport networks under the acronym Multi- 2130 protocol Lambda Switching [AWD1]. This new paradigm of Multi- 2131 protocol Lambda Switching will support even more sophisticated mesh 2132 restoration capabilities at the optical layer for the emerging IP 2133 over WDM network architectures. 2135 Another important aspect regarding multi-layer survivability is that 2136 technologies at different layers provide protection and restoration 2137 capabilities at different temporal granularities (in terms of time 2138 scales) and at different bandwidth granularity (from packet-level to 2139 wavelength level). Protection and restoration capabilities can also 2140 be sensitive to different service classes and different network 2141 utility models. 2143 The impact of service outages varies significantly for different 2144 service classes depending upon the effective duration of the outage. 2145 The duration of an outage can vary from milliseconds (with minor 2146 service impact) to seconds (with possible call drops for IP telephony 2147 and session time-outs for connection oriented transactions) to 2148 minutes and hours (with potentially considerable social and business 2149 impact). 2151 Coordinating different protection and restoration capabilities across 2152 multiple layers in a cohesive manner to ensure network survivability 2153 is maintained at reasonable cost is a challenging task. Protection 2154 and restoration coordination across layers may not always be 2155 feasible, because networks at different layers may belong to 2156 different administrative domains. 2158 The following paragraphs present some of the general recommendations 2159 for protection and restoration coordination. 2161 o Protection and restoration capabilities from different layers 2162 should be coordinated whenever feasible and appropriate to provide 2163 network survivability in a flexible and cost effective manner. 2164 Minimization of function duplication across layers is one way to 2165 achieve the coordination. Escalation of alarms and other fault 2166 indicators from lower to higher layers may also be performed in a 2167 coordinated manner. A temporal order of restoration trigger 2168 timing at different layers is another way to coordinate multi- 2169 layer protection/restoration. 2171 o Spare capacity at higher layers is often regarded as working 2172 traffic at lower layers. Placing protection/restoration functions 2173 in many layers may increase redundancy and robustness, but it 2174 should not result in significant and avoidable inefficiencies in 2175 network resource utilization. 2177 o It is generally desirable to have protection and restoration 2178 schemes that are bandwidth efficient. 2180 o Failure notification throughout the network should be timely and 2181 reliable. 2183 o Alarms and other fault monitoring and reporting capabilities 2184 should be provided at appropriate layers. 2186 6.5.1. Survivability in MPLS Based Networks 2188 MPLS is an important emerging technology that enhances IP networks in 2189 terms of features, capabilities, and services. Because MPLS is path- 2190 oriented, it can potentially provide faster and more predictable 2191 protection and restoration capabilities than conventional hop by hop 2192 routed IP systems. This subsection describes some of the basic 2193 aspects and recommendations for MPLS networks regarding protection 2194 and restoration. See [RFC3469] for a more comprehensive discussion 2195 on MPLS based recovery. 2197 Protection types for MPLS networks can be categorized as link 2198 protection, node protection, path protection, and segment protection. 2200 o Link Protection: The objective for link protection is to protect 2201 an LSP from a given link failure. Under link protection, the path 2202 of the protection or backup LSP (the secondary LSP) is disjoint 2203 from the path of the working or operational LSP at the particular 2204 link over which protection is required. When the protected link 2205 fails, traffic on the working LSP is switched over to the 2206 protection LSP at the head-end of the failed link. This is a 2207 local repair method which can be fast. It might be more 2208 appropriate in situations where some network elements along a 2209 given path are less reliable than others. 2211 o Node Protection: The objective of LSP node protection is to 2212 protect an LSP from a given node failure. Under node protection, 2213 the path of the protection LSP is disjoint from the path of the 2214 working LSP at the particular node to be protected. The secondary 2215 path is also disjoint from the primary path at all links 2216 associated with the node to be protected. When the node fails, 2217 traffic on the working LSP is switched over to the protection LSP 2218 at the upstream LSR directly connected to the failed node. 2220 o Path Protection: The goal of LSP path protection is to protect an 2221 LSP from failure at any point along its routed path. Under path 2222 protection, the path of the protection LSP is completely disjoint 2223 from the path of the working LSP. The advantage of path 2224 protection is that the backup LSP protects the working LSP from 2225 all possible link and node failures along the path, except for 2226 failures that might occur at the ingress and egress LSRs, or for 2227 correlated failures that might impact both working and backup 2228 paths simultaneously. Additionally, since the path selection is 2229 end-to-end, path protection might be more efficient in terms of 2230 resource usage than link or node protection. However, path 2231 protection may be slower than link and node protection in general. 2233 o Segment Protection: An MPLS domain may be partitioned into 2234 multiple protection domains whereby a failure in a protection 2235 domain is rectified within that domain. In cases where an LSP 2236 traverses multiple protection domains, a protection mechanism 2237 within a domain only needs to protect the segment of the LSP that 2238 lies within the domain. Segment protection will generally be 2239 faster than path protection because recovery generally occurs 2240 closer to the fault. 2242 6.5.2. Protection Option 2244 Another issue to consider is the concept of protection options. The 2245 protection option uses the notation m:n protection, where m is the 2246 number of protection LSPs used to protect n working LSPs. Feasible 2247 protection options follow. 2249 o 1:1: one working LSP is protected/restored by one protection LSP. 2251 o 1:n: one protection LSP is used to protect/restore n working LSPs. 2253 o n:1: one working LSP is protected/restored by n protection LSPs, 2254 possibly with configurable load splitting ratio. When more than 2255 one protection LSP is used, it may be desirable to share the 2256 traffic across the protection LSPs when the working LSP fails to 2257 satisfy the bandwidth requirement of the traffic trunk associated 2258 with the working LSP. This may be especially useful when it is 2259 not feasible to find one path that can satisfy the bandwidth 2260 requirement of the primary LSP. 2262 o 1+1: traffic is sent concurrently on both the working LSP and the 2263 protection LSP. In this case, the egress LSR selects one of the 2264 two LSPs based on a local traffic integrity decision process, 2265 which compares the traffic received from both the working and the 2266 protection LSP and identifies discrepancies. It is unlikely that 2267 this option would be used extensively in IP networks due to its 2268 resource utilization inefficiency. However, if bandwidth becomes 2269 plentiful and cheap, then this option might become quite viable 2270 and attractive in IP networks. 2272 6.6. Traffic Engineering in Diffserv Environments 2274 This section provides an overview of the traffic engineering features 2275 and recommendations that are specifically pertinent to Differentiated 2276 Services (Diffserv) [RFC2475] capable IP networks. 2278 Increasing requirements to support multiple classes of traffic, such 2279 as best effort and mission critical data, in the Internet calls for 2280 IP networks to differentiate traffic according to some criteria, and 2281 to accord preferential treatment to certain types of traffic. Large 2282 numbers of flows can be aggregated into a few behavior aggregates 2283 based on some criteria in terms of common performance requirements in 2284 terms of packet loss ratio, delay, and jitter; or in terms of common 2285 fields within the IP packet headers. 2287 As Diffserv evolves and becomes deployed in operational networks, 2288 traffic engineering will be critical to ensuring that SLAs defined 2289 within a given Diffserv service model are met. Classes of service 2290 (CoS) can be supported in a Diffserv environment by concatenating 2291 per-hop behaviors (PHBs) along the routing path, using service 2292 provisioning mechanisms, and by appropriately configuring edge 2293 functionality such as traffic classification, marking, policing, and 2294 shaping. PHB is the forwarding behavior that a packet receives at a 2295 DS node (a Diffserv-compliant node). This is accomplished by means 2296 of buffer management and packet scheduling mechanisms. In this 2297 context, packets belonging to a class are those that are members of a 2298 corresponding ordering aggregate. 2300 Traffic engineering can be used as a compliment to Diffserv 2301 mechanisms to improve utilization of network resources, but not as a 2302 necessary element in general. When traffic engineering is used, it 2303 can be operated on an aggregated basis across all service classes 2304 [RFC3270] or on a per service class basis. The former is used to 2305 provide better distribution of the aggregate traffic load over the 2306 network resources. (See [RFC3270] for detailed mechanisms to support 2307 aggregate traffic engineering.) The latter case is discussed below 2308 since it is specific to the Diffserv environment, with so called 2309 Diffserv-aware traffic engineering [RFC4124]. 2311 For some Diffserv networks, it may be desirable to control the 2312 performance of some service classes by enforcing certain 2313 relationships between the traffic workload contributed by each 2314 service class and the amount of network resources allocated or 2315 provisioned for that service class. Such relationships between 2316 demand and resource allocation can be enforced using a combination 2317 of, for example: 2319 o traffic engineering mechanisms on a per service class basis that 2320 enforce the desired relationship between the amount of traffic 2321 contributed by a given service class and the resources allocated 2322 to that class 2324 o mechanisms that dynamically adjust the resources allocated to a 2325 given service class to relate to the amount of traffic contributed 2326 by that service class. 2328 It may also be desirable to limit the performance impact of high 2329 priority traffic on relatively low priority traffic. This can be 2330 achieved by, for example, controlling the percentage of high priority 2331 traffic that is routed through a given link. Another way to 2332 accomplish this is to increase link capacities appropriately so that 2333 lower priority traffic can still enjoy adequate service quality. 2334 When the ratio of traffic workload contributed by different service 2335 classes vary significantly from router to router, it may not suffice 2336 to rely exclusively on conventional IGP routing protocols or on 2337 traffic engineering mechanisms that are insensitive to different 2338 service classes. Instead, it may be desirable to perform traffic 2339 engineering, especially routing control and mapping functions, on a 2340 per service class basis. One way to accomplish this in a domain that 2341 supports both MPLS and Diffserv is to define class specific LSPs and 2342 to map traffic from each class onto one or more LSPs that correspond 2343 to that service class. An LSP corresponding to a given service class 2344 can then be routed and protected/restored in a class dependent 2345 manner, according to specific policies. 2347 Performing traffic engineering on a per class basis may require 2348 certain per-class parameters to be distributed. Note that it is 2349 common to have some classes share some aggregate constraint (e.g., 2350 maximum bandwidth requirement) without enforcing the constraint on 2351 each individual class. These classes then can be grouped into a 2352 class-type and per-class-type parameters can be distributed instead 2353 to improve scalability. It also allows better bandwidth sharing 2354 between classes in the same class-type. A class-type is a set of 2355 classes that satisfy the following two conditions: 2357 o Classes in the same class-type have common aggregate requirements 2358 to satisfy required performance levels. 2360 o There is no requirement to be enforced at the level of individual 2361 class in the class-type. Note that it is still possible, 2362 nevertheless, to implement some priority policies for classes in 2363 the same class-type to permit preferential access to the class- 2364 type bandwidth through the use of preemption priorities. 2366 An example of the class-type can be a low-loss class-type that 2367 includes both AF1-based and AF2-based Ordering Aggregates. With such 2368 a class-type, one may implement some priority policy which assigns 2369 higher preemption priority to AF1-based traffic trunks over AF2-based 2370 ones, vice versa, or the same priority. 2372 See [RFC4124] for detailed requirements on Diffserv-aware traffic 2373 engineering. 2375 6.7. Network Controllability 2377 Off-line (and on-line) traffic engineering considerations would be of 2378 limited utility if the network could not be controlled effectively to 2379 implement the results of TE decisions and to achieve desired network 2380 performance objectives. Capacity augmentation is a coarse grained 2381 solution to traffic engineering issues. However, it is simple and 2382 may be advantageous if bandwidth is abundant and cheap or if the 2383 current or expected network workload demands it. However, bandwidth 2384 is not always abundant and cheap, and the workload may not always 2385 demand additional capacity. Adjustments of administrative weights 2386 and other parameters associated with routing protocols provide finer 2387 grained control, but is difficult to use and imprecise because of the 2388 routing interactions that occur across the network. In certain 2389 network contexts, more flexible, finer grained approaches which 2390 provide more precise control over the mapping of traffic to routes 2391 and over the selection and placement of routes may be appropriate and 2392 useful. 2394 Control mechanisms can be manual (e.g., administrative 2395 configuration), partially-automated (e.g., scripts) or fully- 2396 automated (e.g., policy based management systems). Automated 2397 mechanisms are particularly required in large scale networks. Multi- 2398 vendor interoperability can be facilitated by developing and 2399 deploying standardized management systems (e.g., standard MIBs) and 2400 policies (PIBs) to support the control functions required to address 2401 traffic engineering objectives such as load distribution and 2402 protection/restoration. 2404 Network control functions should be secure, reliable, and stable as 2405 these are often needed to operate correctly in times of network 2406 impairments (e.g., during network congestion or security attacks). 2408 7. Inter-Domain Considerations 2410 Inter-domain traffic engineering is concerned with the performance 2411 optimization for traffic that originates in one administrative domain 2412 and terminates in a different one. 2414 Traffic exchange between autonomous systems in the Internet occurs 2415 through exterior gateway protocols. Currently, BGP [RFC4271] is the 2416 standard exterior gateway protocol for the Internet. BGP provides a 2417 number of attributes and capabilities (e.g., route filtering) that 2418 can be used for inter-domain traffic engineering. More specifically, 2419 BGP permits the control of routing information and traffic exchange 2420 between Autonomous Systems (ASes) in the Internet. BGP incorporates 2421 a sequential decision process which calculates the degree of 2422 preference for various routes to a given destination network. There 2423 are two fundamental aspects to inter-domain traffic engineering using 2424 BGP: 2426 o Route Redistribution: controlling the import and export of routes 2427 between AS's, and controlling the redistribution of routes between 2428 BGP and other protocols within an AS. 2430 o Best path selection: selecting the best path when there are 2431 multiple candidate paths to a given destination network. Best 2432 path selection is performed by the BGP decision process based on a 2433 sequential procedure, taking a number of different considerations 2434 into account. Ultimately, best path selection under BGP boils 2435 down to selecting preferred exit points out of an AS towards 2436 specific destination networks. The BGP path selection process can 2437 be influenced by manipulating the attributes associated with the 2438 BGP decision process. These attributes include: NEXT-HOP, WEIGHT 2439 (Cisco proprietary which is also implemented by some other 2440 vendors), LOCAL-PREFERENCE, AS-PATH, ROUTE-ORIGIN, MULTI-EXIT- 2441 DESCRIMINATOR (MED), IGP METRIC, etc. 2443 Route-maps provide the flexibility to implement complex BGP policies 2444 based on pre-configured logical conditions. In particular, Route- 2445 maps can be used to control import and export policies for incoming 2446 and outgoing routes, control the redistribution of routes between BGP 2447 and other protocols, and influence the selection of best paths by 2448 manipulating the attributes associated with the BGP decision process. 2449 Very complex logical expressions that implement various types of 2450 policies can be implemented using a combination of Route-maps, BGP- 2451 attributes, Access-lists, and Community attributes. 2453 When looking at possible strategies for inter-domain TE with BGP, it 2454 must be noted that the outbound traffic exit point is controllable, 2455 whereas the interconnection point where inbound traffic is received 2456 from an EBGP peer typically is not, unless a special arrangement is 2457 made with the peer sending the traffic. Therefore, it is up to each 2458 individual network to implement sound TE strategies that deal with 2459 the efficient delivery of outbound traffic from one's customers to 2460 one's peering points. The vast majority of TE policy is based upon a 2461 "closest exit" strategy, which offloads interdomain traffic at the 2462 nearest outbound peer point towards the destination autonomous 2463 system. Most methods of manipulating the point at which inbound 2464 traffic enters a network from an EBGP peer (inconsistent route 2465 announcements between peering points, AS pre-pending, and sending 2466 MEDs) are either ineffective, or not accepted in the peering 2467 community. 2469 Inter-domain TE with BGP is generally effective, but it is usually 2470 applied in a trial-and-error fashion. A systematic approach for 2471 inter-domain traffic engineering is yet to be devised. 2473 Inter-domain TE is inherently more difficult than intra-domain TE 2474 under the current Internet architecture. The reasons for this are 2475 both technical and administrative. Technically, while topology and 2476 link state information are helpful for mapping traffic more 2477 effectively, BGP does not propagate such information across domain 2478 boundaries for stability and scalability reasons. Administratively, 2479 there are differences in operating costs and network capacities 2480 between domains. Generally, what may be considered a good solution 2481 in one domain may not necessarily be a good solution in another 2482 domain. Moreover, it would generally be considered inadvisable for 2483 one domain to permit another domain to influence the routing and 2484 management of traffic in its network. 2486 MPLS TE-tunnels (explicit LSPs) can potentially add a degree of 2487 flexibility in the selection of exit points for inter-domain routing. 2488 The concept of relative and absolute metrics can be applied to this 2489 purpose. The idea is that if BGP attributes are defined such that 2490 the BGP decision process depends on IGP metrics to select exit points 2491 for inter-domain traffic, then some inter-domain traffic destined to 2492 a given peer network can be made to prefer a specific exit point by 2493 establishing a TE-tunnel between the router making the selection to 2494 the peering point via a TE-tunnel and assigning the TE-tunnel a 2495 metric which is smaller than the IGP cost to all other peering 2496 points. If a peer accepts and processes MEDs, then a similar MPLS 2497 TE-tunnel based scheme can be applied to cause certain entrance 2498 points to be preferred by setting MED to be an IGP cost, which has 2499 been modified by the tunnel metric. 2501 Similar to intra-domain TE, inter-domain TE is best accomplished when 2502 a traffic matrix can be derived to depict the volume of traffic from 2503 one autonomous system to another. 2505 Generally, redistribution of inter-domain traffic requires 2506 coordination between peering partners. An export policy in one 2507 domain that results in load redistribution across peer points with 2508 another domain can significantly affect the local traffic matrix 2509 inside the domain of the peering partner. This, in turn, will affect 2510 the intra-domain TE due to changes in the spatial distribution of 2511 traffic. Therefore, it is mutually beneficial for peering partners 2512 to coordinate with each other before attempting any policy changes 2513 that may result in significant shifts in inter-domain traffic. In 2514 certain contexts, this coordination can be quite challenging due to 2515 technical and non- technical reasons. 2517 It is a matter of speculation as to whether MPLS, or similar 2518 technologies, can be extended to allow selection of constrained paths 2519 across domain boundaries. 2521 8. Overview of Contemporary TE Practices in Operational IP Networks 2523 This section provides an overview of some contemporary traffic 2524 engineering practices in IP networks. The focus is primarily on the 2525 aspects that pertain to the control of the routing function in 2526 operational contexts. The intent here is to provide an overview of 2527 the commonly used practices. The discussion is not intended to be 2528 exhaustive. 2530 Currently, service providers apply many of the traffic engineering 2531 mechanisms discussed in this document to optimize the performance of 2532 their IP networks. These techniques include capacity planning for 2533 long timescales, routing control using IGP metrics and MPLS for 2534 medium timescales, the overlay model also for medium timescales, and 2535 traffic management mechanisms for short timescale. 2537 When a service provider plans to build an IP network, or expand the 2538 capacity of an existing network, effective capacity planning should 2539 be an important component of the process. Such plans may take the 2540 following aspects into account: location of new nodes if any, 2541 existing and predicted traffic patterns, costs, link capacity, 2542 topology, routing design, and survivability. 2544 Performance optimization of operational networks is usually an 2545 ongoing process in which traffic statistics, performance parameters, 2546 and fault indicators are continually collected from the network. 2547 This empirical data is then analyzed and used to trigger various 2548 traffic engineering mechanisms. Tools that perform what-if analysis 2549 can also be used to assist the TE process by allowing various 2550 scenarios to be reviewed before a new set of configurations are 2551 implemented in the operational network. 2553 Traditionally, intra-domain real-time TE with IGP is done by 2554 increasing the OSPF or IS-IS metric of a congested link until enough 2555 traffic has been diverted from that link. This approach has some 2556 limitations as discussed in Section 6.2. Recently, some new intra- 2557 domain TE approaches/tools have been proposed [RR94] [FT00] [FT01] 2558 [WANG]. Such approaches/tools take traffic matrix, network topology, 2559 and network performance objectives as input, and produce some link 2560 metrics and possibly some unequal load-sharing ratios to be set at 2561 the head-end routers of some ECMPs as output. These new progresses 2562 open new possibility for intra-domain TE with IGP to be done in a 2563 more systematic way. 2565 The overlay model (IP over ATM, or IP over Frame Relay) is another 2566 approach which was commonly used [AWD2], but has been replaced by 2567 MPLS and router hardware technology. 2569 Deployment of MPLS for traffic engineering applications has commenced 2570 in some service provider networks. One operational scenario is to 2571 deploy MPLS in conjunction with an IGP (IS-IS-TE or OSPF-TE) that 2572 supports the traffic engineering extensions, in conjunction with 2573 constraint-based routing for explicit route computations, and a 2574 signaling protocol (e.g., RSVP-TE) for LSP instantiation. 2576 In contemporary MPLS traffic engineering contexts, network 2577 administrators specify and configure link attributes and resource 2578 constraints such as maximum reservable bandwidth and resource class 2579 attributes for links (interfaces) within the MPLS domain. A link 2580 state protocol that supports TE extensions (IS-IS-TE or OSPF-TE) is 2581 used to propagate information about network topology and link 2582 attribute to all routers in the routing area. Network administrators 2583 also specify all the LSPs that are to originate each router. For 2584 each LSP, the network administrator specifies the destination node 2585 and the attributes of the LSP which indicate the requirements that to 2586 be satisfied during the path selection process. Each router then 2587 uses a local constraint-based routing process to compute explicit 2588 paths for all LSPs originating from it. Subsequently, a signaling 2589 protocol is used to instantiate the LSPs. By assigning proper 2590 bandwidth values to links and LSPs, congestion caused by uneven 2591 traffic distribution can generally be avoided or mitigated. 2593 The bandwidth attributes of LSPs used for traffic engineering can be 2594 updated periodically. The basic concept is that the bandwidth 2595 assigned to an LSP should relate in some manner to the bandwidth 2596 requirements of traffic that actually flows through the LSP. The 2597 traffic attribute of an LSP can be modified to accommodate traffic 2598 growth and persistent traffic shifts. If network congestion occurs 2599 due to some unexpected events, existing LSPs can be rerouted to 2600 alleviate the situation or network administrator can configure new 2601 LSPs to divert some traffic to alternative paths. The reservable 2602 bandwidth of the congested links can also be reduced to force some 2603 LSPs to be rerouted to other paths. 2605 In an MPLS domain, a traffic matrix can also be estimated by 2606 monitoring the traffic on LSPs. Such traffic statistics can be used 2607 for a variety of purposes including network planning and network 2608 optimization. Current practice suggests that deploying an MPLS 2609 network consisting of hundreds of routers and thousands of LSPs is 2610 feasible. In summary, recent deployment experience suggests that 2611 MPLS approach is very effective for traffic engineering in IP 2612 networks [XIAO]. 2614 As mentioned previously in Section 7, one usually has no direct 2615 control over the distribution of inbound traffic. Therefore, the 2616 main goal of contemporary inter-domain TE is to optimize the 2617 distribution of outbound traffic between multiple inter-domain links. 2618 When operating a global network, maintaining the ability to operate 2619 the network in a regional fashion where desired, while continuing to 2620 take advantage of the benefits of a global network, also becomes an 2621 important objective. 2623 Inter-domain TE with BGP usually begins with the placement of 2624 multiple peering interconnection points in locations that have high 2625 peer density, are in close proximity to originating/terminating 2626 traffic locations on one's own network, and are lowest in cost. 2627 There are generally several locations in each region of the world 2628 where the vast majority of major networks congregate and 2629 interconnect. Some location-decision problems that arise in 2630 association with inter-domain routing are discussed in [AWD5]. 2632 Once the locations of the interconnects are determined, and circuits 2633 are implemented, one decides how best to handle the routes heard from 2634 the peer, as well as how to propagate the peers' routes within one's 2635 own network. One way to engineer outbound traffic flows on a network 2636 with many EBGP peers is to create a hierarchy of peers. Generally, 2637 the Local Preferences of all peers are set to the same value so that 2638 the shortest AS paths will be chosen to forward traffic. Then, by 2639 over-writing the inbound MED metric (Multi-exit-discriminator metric, 2640 also referred to as "BGP metric". Both terms are used 2641 interchangeably in this document) with BGP metrics to routes received 2642 at different peers, the hierarchy can be formed. For example, all 2643 Local Preferences can be set to 200, preferred private peers can be 2644 assigned a BGP metric of 50, the rest of the private peers can be 2645 assigned a BGP metric of 100, and public peers can be assigned a BGP 2646 metric of 600. "Preferred" peers might be defined as those peers 2647 with whom the most available capacity exists, whose customer base is 2648 larger in comparison to other peers, whose interconnection costs are 2649 the lowest, and with whom upgrading existing capacity is the easiest. 2650 In a network with low utilization at the edge, this works well. The 2651 same concept could be applied to a network with higher edge 2652 utilization by creating more levels of BGP metrics between peers, 2653 allowing for more granularity in selecting the exit points for 2654 traffic bound for a dual homed customer on a peer's network. 2656 By only replacing inbound MED metrics with BGP metrics, only equal 2657 AS-Path length routes' exit points are being changed. (The BGP 2658 decision considers Local Preference first, then AS-Path length, and 2659 then BGP metric). For example, assume a network has two possible 2660 egress points, peer A and peer B. Each peer has 40% of the 2661 Internet's routes exclusively on its network, while the remaining 20% 2662 of the Internet's routes are from customers who dual home between A 2663 and B. Assume that both peers have a Local Preference of 200 and a 2664 BGP metric of 100. If the link to peer A is congested, increasing 2665 its BGP metric while leaving the Local Preference at 200 will ensure 2666 that the 20% of total routes belonging to dual homed customers will 2667 prefer peer B as the exit point. The previous example would be used 2668 in a situation where all exit points to a given peer were close to 2669 congestion levels, and traffic needed to be shifted away from that 2670 peer entirely. 2672 When there are multiple exit points to a given peer, and only one of 2673 them is congested, it is not necessary to shift traffic away from the 2674 peer entirely, but only from the one congested circuit. This can be 2675 achieved by using passive IGP-metrics, AS-path filtering, or prefix 2676 filtering. 2678 Occasionally, more drastic changes are needed, for example, in 2679 dealing with a "problem peer" who is difficult to work with on 2680 upgrades or is charging high prices for connectivity to their 2681 network. In that case, the Local Preference to that peer can be 2682 reduced below the level of other peers. This effectively reduces the 2683 amount of traffic sent to that peer to only originating traffic 2684 (assuming no transit providers are involved). This type of change 2685 can affect a large amount of traffic, and is only used after other 2686 methods have failed to provide the desired results. 2688 Although it is not much of an issue in regional networks, the 2689 propagation of a peer's routes back through the network must be 2690 considered when a network is peering on a global scale. Sometimes, 2691 business considerations can influence the choice of BGP policies in a 2692 given context. For example, it may be imprudent, from a business 2693 perspective, to operate a global network and provide full access to 2694 the global customer base to a small network in a particular country. 2695 However, for the purpose of providing one's own customers with 2696 quality service in a particular region, good connectivity to that in- 2697 country network may still be necessary. This can be achieved by 2698 assigning a set of communities at the edge of the network, which have 2699 a known behavior when routes tagged with those communities are 2700 propagating back through the core. Routes heard from local peers 2701 will be prevented from propagating back to the global network, 2702 whereas routes learned from larger peers may be allowed to propagate 2703 freely throughout the entire global network. By implementing a 2704 flexible community strategy, the benefits of using a single global AS 2705 Number (ASN) can be realized, while the benefits of operating 2706 regional networks can also be taken advantage of. An alternative to 2707 doing this is to use different ASNs in different regions, with the 2708 consequence that the AS path length for routes announced by that 2709 service provider will increase. 2711 9. Conclusion 2713 This document described principles for traffic engineering in the 2714 Internet. It presented an overview of some of the basic issues 2715 surrounding traffic engineering in IP networks. The context of TE 2716 was described, a TE process models and a taxonomy of TE styles were 2717 presented. A brief historical review of pertinent developments 2718 related to traffic engineering was provided. A survey of 2719 contemporary TE techniques in operational networks was presented. 2720 Additionally, the document specified a set of generic requirements, 2721 recommendations, and options for Internet traffic engineering. 2723 10. Security Considerations 2725 This document does not introduce new security issues. 2727 11. IANA Considerations 2729 This draft makes no requests for IANA action. 2731 12. Acknowledgments 2733 Much of the text in this document is derived from RFC 3272. The 2734 authors of this document would like to express their gratitude to all 2735 involved in that work. Although the source text has been edited in 2736 the production of this document, the orginal authors should be 2737 considered as Contributors to this work. They were: 2739 Daniel O. Awduche 2740 Movaz Networks 2741 7926 Jones Branch Drive, Suite 615 2742 McLean, VA 22102 2744 Phone: 703-298-5291 2745 EMail: awduche@movaz.com 2747 Angela Chiu 2748 Celion Networks 2749 1 Sheila Dr., Suite 2 2750 Tinton Falls, NJ 07724 2752 Phone: 732-747-9987 2753 EMail: angela.chiu@celion.com 2755 Anwar Elwalid 2756 Lucent Technologies 2757 Murray Hill, NJ 07974 2759 Phone: 908 582-7589 2760 EMail: anwar@lucent.com 2762 Indra Widjaja 2763 Bell Labs, Lucent Technologies 2764 600 Mountain Avenue 2765 Murray Hill, NJ 07974 2767 Phone: 908 582-0435 2768 EMail: iwidjaja@research.bell-labs.com 2770 XiPeng Xiao 2771 Redback Networks 2772 300 Holger Way 2773 San Jose, CA 95134 2775 Phone: 408-750-5217 2776 EMail: xipeng@redback.com 2778 The acknowledgements in RFC3272 were as below. All people who helped 2779 in the production of that document also need to be thanked for the 2780 carry-over into this new document. 2782 The authors would like to thank Jim Boyle for inputs on the 2783 recommendations section, Francois Le Faucheur for inputs on 2784 Diffserv aspects, Blaine Christian for inputs on measurement, 2785 Gerald Ash for inputs on routing in telephone networks and for 2786 text on event-dependent TE methods, Steven Wright for inputs 2787 on network controllability, and Jonathan Aufderheide for 2788 inputs on inter-domain TE with BGP. Special thanks to 2789 Randy Bush for proposing the TE taxonomy based on "tactical vs 2790 strategic" methods. The subsection describing an "Overview of 2791 ITU Activities Related to Traffic Engineering" was adapted from 2792 a contribution by Waisum Lai. Useful feedback and pointers to 2793 relevant materials were provided by J. Noel Chiappa. 2794 Additional comments were provided by Glenn Grotefeld during 2795 the working last call process. Finally, the authors would like 2796 to thank Ed Kern, the TEWG co-chair, for his comments and 2797 support. 2799 The early versions of this document were produced by the TEAS Working 2800 Group's RFC3272bis Design Team. The full list of members of this 2801 team is: 2803 Acee Lindem 2804 Adrian Farrel 2805 Aijun Wang 2806 Daniele Ceccarelli 2807 Dieter Beller 2808 Jeff Tantsura 2809 Julien Meuric 2810 Liu Hua 2811 Loa Andersson 2812 Luis Miguel Contreras 2813 Martin Horneffer 2814 Tarek Saad 2815 Xufeng Liu 2817 The production of this document includes a fix to the original text 2818 resulting from an Errata Report by Jean-Michel Grimaldi. 2820 The authors of this document would also like to thank TBD. 2822 13. Contributors 2824 The following people contributed substantive text to this document: 2826 Gert Grammel 2827 EMail: ggrammel@juniper.net 2829 Loa Andersson 2830 EMail: loa@pi.nu 2832 Xufeng Liu 2833 EMail: xufeng.liu.ietf@gmail.com 2835 14. Informative References 2837 [AJ19] Adekitan, A., Abolade, J., and O. Shobayo, "Data mining 2838 approach for predicting the daily Internet data traffic of 2839 a smart university", Article Journal of Big Data, 2019, 2840 Volume 6, Number 1, Page 1, 1998. 2842 [ASH2] Ash, J., "Dynamic Routing in Telecommunications Networks", 2843 Book McGraw Hill, 1998. 2845 [AWD1] Awduche, D. and Y. Rekhter, "Multiprocotol Lambda 2846 Switching - Combining MPLS Traffic Engineering Control 2847 with Optical Crossconnects", Article IEEE Communications 2848 Magazine, March 2001. 2850 [AWD2] Awduche, D., "MPLS and Traffic Engineering in IP 2851 Networks", Article IEEE Communications Magazine, December 2852 1999. 2854 [AWD5] Awduche, D., "An Approach to Optimal Peering Between 2855 Autonomous Systems in the Internet", Paper International 2856 Conference on Computer Communications and Networks 2857 (ICCCN'98), October 1998. 2859 [FLJA93] Floyd, S. and V. Jacobson, "Random Early Detection 2860 Gateways for Congestion Avoidance", Article IEEE/ACM 2861 Transactions on Networking, Vol. 1, p. 387-413, November 2862 1993. 2864 [FLOY94] Floyd, S., "TCP and Explicit Congestion Notification", 2865 Article ACM Computer Communication Review, V. 24, No. 5, 2866 p. 10-23, October 1994. 2868 [FT00] Fortz, B. and M. Thorup, "Internet Traffic Engineering by 2869 Optimizing OSPF Weights", Article IEEE INFOCOM 2000, March 2870 2000. 2872 [FT01] Fortz, B. and M. Thorup, "Optimizing OSPF/IS-IS Weights in 2873 a Changing World", n.d., 2874 . 2876 [HUSS87] Hurley, B., Seidl, C., and W. Sewel, "A Survey of Dynamic 2877 Routing Methods for Circuit-Switched Traffic", 2878 Article IEEE Communication Magazine, September 1987. 2880 [I-D.ietf-teas-yang-te-topo] 2881 Liu, X., Bryskin, I., Beeram, V., Saad, T., Shah, H., and 2882 O. Dios, "YANG Data Model for Traffic Engineering (TE) 2883 Topologies", draft-ietf-teas-yang-te-topo-22 (work in 2884 progress), June 2019. 2886 [I-D.ietf-tewg-qos-routing] 2887 Ash, G., "Traffic Engineering & QoS Methods for IP-, ATM-, 2888 & Based Multiservice Networks", draft-ietf-tewg-qos- 2889 routing-04 (work in progress), October 2001. 2891 [ITU-E600] 2892 "Terms and Definitions of Traffic Engineering", 2893 Recommendation ITU-T Recommendation E.600, March 1993. 2895 [ITU-E701] 2896 "Reference Connections for Traffic Engineering", 2897 Recommendation ITU-T Recommendation E.701, October 1993. 2899 [ITU-E801] 2900 "Framework for Service Quality Agreement", 2901 Recommendation ITU-T Recommendation E.801, October 1996. 2903 [MA] Ma, Q., "Quality of Service Routing in Integrated Services 2904 Networks", Ph.D. PhD Dissertation, CMU-CS-98-138, CMU, 2905 1998. 2907 [MATE] Elwalid, A., Jin, C., Low, S., and I. Widjaja, "MATE - 2908 MPLS Adaptive Traffic Engineering", 2909 Proceedings INFOCOM'01, April 2001. 2911 [MCQ80] McQuillan, J., Richer, I., and E. Rosen, "The New Routing 2912 Algorithm for the ARPANET", Transaction IEEE Transactions 2913 on Communications, vol. 28, no. 5, p. 711-719, May 1980. 2915 [MR99] Mitra, D. and K. Ramakrishnan, "A Case Study of 2916 Multiservice, Multipriority Traffic Engineering Design for 2917 Data Networks", Proceedings Globecom'99, December 1999. 2919 [RFC1992] Castineyra, I., Chiappa, N., and M. Steenstrup, "The 2920 Nimrod Routing Architecture", RFC 1992, 2921 DOI 10.17487/RFC1992, August 1996, 2922 . 2924 [RFC2205] Braden, R., Ed., Zhang, L., Berson, S., Herzog, S., and S. 2925 Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1 2926 Functional Specification", RFC 2205, DOI 10.17487/RFC2205, 2927 September 1997, . 2929 [RFC2328] Moy, J., "OSPF Version 2", STD 54, RFC 2328, 2930 DOI 10.17487/RFC2328, April 1998, 2931 . 2933 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 2934 "Framework for IP Performance Metrics", RFC 2330, 2935 DOI 10.17487/RFC2330, May 1998, 2936 . 2938 [RFC2386] Crawley, E., Nair, R., Rajagopalan, B., and H. Sandick, "A 2939 Framework for QoS-based Routing in the Internet", 2940 RFC 2386, DOI 10.17487/RFC2386, August 1998, 2941 . 2943 [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, 2944 "Definition of the Differentiated Services Field (DS 2945 Field) in the IPv4 and IPv6 Headers", RFC 2474, 2946 DOI 10.17487/RFC2474, December 1998, 2947 . 2949 [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., 2950 and W. Weiss, "An Architecture for Differentiated 2951 Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, 2952 . 2954 [RFC2597] Heinanen, J., Baker, F., Weiss, W., and J. Wroclawski, 2955 "Assured Forwarding PHB Group", RFC 2597, 2956 DOI 10.17487/RFC2597, June 1999, 2957 . 2959 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring 2960 Connectivity", RFC 2678, DOI 10.17487/RFC2678, September 2961 1999, . 2963 [RFC2702] Awduche, D., Malcolm, J., Agogbua, J., O'Dell, M., and J. 2964 McManus, "Requirements for Traffic Engineering Over MPLS", 2965 RFC 2702, DOI 10.17487/RFC2702, September 1999, 2966 . 2968 [RFC2722] Brownlee, N., Mills, C., and G. Ruth, "Traffic Flow 2969 Measurement: Architecture", RFC 2722, 2970 DOI 10.17487/RFC2722, October 1999, 2971 . 2973 [RFC2753] Yavatkar, R., Pendarakis, D., and R. Guerin, "A Framework 2974 for Policy-based Admission Control", RFC 2753, 2975 DOI 10.17487/RFC2753, January 2000, 2976 . 2978 [RFC2961] Berger, L., Gan, D., Swallow, G., Pan, P., Tommasi, F., 2979 and S. Molendini, "RSVP Refresh Overhead Reduction 2980 Extensions", RFC 2961, DOI 10.17487/RFC2961, April 2001, 2981 . 2983 [RFC2998] Bernet, Y., Ford, P., Yavatkar, R., Baker, F., Zhang, L., 2984 Speer, M., Braden, R., Davie, B., Wroclawski, J., and E. 2985 Felstaine, "A Framework for Integrated Services Operation 2986 over Diffserv Networks", RFC 2998, DOI 10.17487/RFC2998, 2987 November 2000, . 2989 [RFC3031] Rosen, E., Viswanathan, A., and R. Callon, "Multiprotocol 2990 Label Switching Architecture", RFC 3031, 2991 DOI 10.17487/RFC3031, January 2001, 2992 . 2994 [RFC3086] Nichols, K. and B. Carpenter, "Definition of 2995 Differentiated Services Per Domain Behaviors and Rules for 2996 their Specification", RFC 3086, DOI 10.17487/RFC3086, 2997 April 2001, . 2999 [RFC3124] Balakrishnan, H. and S. Seshan, "The Congestion Manager", 3000 RFC 3124, DOI 10.17487/RFC3124, June 2001, 3001 . 3003 [RFC3209] Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan, V., 3004 and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP 3005 Tunnels", RFC 3209, DOI 10.17487/RFC3209, December 2001, 3006 . 3008 [RFC3270] Le Faucheur, F., Wu, L., Davie, B., Davari, S., Vaananen, 3009 P., Krishnan, R., Cheval, P., and J. Heinanen, "Multi- 3010 Protocol Label Switching (MPLS) Support of Differentiated 3011 Services", RFC 3270, DOI 10.17487/RFC3270, May 2002, 3012 . 3014 [RFC3272] Awduche, D., Chiu, A., Elwalid, A., Widjaja, I., and X. 3015 Xiao, "Overview and Principles of Internet Traffic 3016 Engineering", RFC 3272, DOI 10.17487/RFC3272, May 2002, 3017 . 3019 [RFC3469] Sharma, V., Ed. and F. Hellstrand, Ed., "Framework for 3020 Multi-Protocol Label Switching (MPLS)-based Recovery", 3021 RFC 3469, DOI 10.17487/RFC3469, February 2003, 3022 . 3024 [RFC3630] Katz, D., Kompella, K., and D. Yeung, "Traffic Engineering 3025 (TE) Extensions to OSPF Version 2", RFC 3630, 3026 DOI 10.17487/RFC3630, September 2003, 3027 . 3029 [RFC3945] Mannie, E., Ed., "Generalized Multi-Protocol Label 3030 Switching (GMPLS) Architecture", RFC 3945, 3031 DOI 10.17487/RFC3945, October 2004, 3032 . 3034 [RFC4090] Pan, P., Ed., Swallow, G., Ed., and A. Atlas, Ed., "Fast 3035 Reroute Extensions to RSVP-TE for LSP Tunnels", RFC 4090, 3036 DOI 10.17487/RFC4090, May 2005, 3037 . 3039 [RFC4124] Le Faucheur, F., Ed., "Protocol Extensions for Support of 3040 Diffserv-aware MPLS Traffic Engineering", RFC 4124, 3041 DOI 10.17487/RFC4124, June 2005, 3042 . 3044 [RFC4271] Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A 3045 Border Gateway Protocol 4 (BGP-4)", RFC 4271, 3046 DOI 10.17487/RFC4271, January 2006, 3047 . 3049 [RFC4655] Farrel, A., Vasseur, J., and J. Ash, "A Path Computation 3050 Element (PCE)-Based Architecture", RFC 4655, 3051 DOI 10.17487/RFC4655, August 2006, 3052 . 3054 [RFC4872] Lang, J., Ed., Rekhter, Y., Ed., and D. Papadimitriou, 3055 Ed., "RSVP-TE Extensions in Support of End-to-End 3056 Generalized Multi-Protocol Label Switching (GMPLS) 3057 Recovery", RFC 4872, DOI 10.17487/RFC4872, May 2007, 3058 . 3060 [RFC4873] Berger, L., Bryskin, I., Papadimitriou, D., and A. Farrel, 3061 "GMPLS Segment Recovery", RFC 4873, DOI 10.17487/RFC4873, 3062 May 2007, . 3064 [RFC5305] Li, T. and H. Smit, "IS-IS Extensions for Traffic 3065 Engineering", RFC 5305, DOI 10.17487/RFC5305, October 3066 2008, . 3068 [RFC5331] Aggarwal, R., Rekhter, Y., and E. Rosen, "MPLS Upstream 3069 Label Assignment and Context-Specific Label Space", 3070 RFC 5331, DOI 10.17487/RFC5331, August 2008, 3071 . 3073 [RFC5440] Vasseur, JP., Ed. and JL. Le Roux, Ed., "Path Computation 3074 Element (PCE) Communication Protocol (PCEP)", RFC 5440, 3075 DOI 10.17487/RFC5440, March 2009, 3076 . 3078 [RFC6119] Harrison, J., Berger, J., and M. Bartlett, "IPv6 Traffic 3079 Engineering in IS-IS", RFC 6119, DOI 10.17487/RFC6119, 3080 February 2011, . 3082 [RFC6241] Enns, R., Ed., Bjorklund, M., Ed., Schoenwaelder, J., Ed., 3083 and A. Bierman, Ed., "Network Configuration Protocol 3084 (NETCONF)", RFC 6241, DOI 10.17487/RFC6241, June 2011, 3085 . 3087 [RFC6374] Frost, D. and S. Bryant, "Packet Loss and Delay 3088 Measurement for MPLS Networks", RFC 6374, 3089 DOI 10.17487/RFC6374, September 2011, 3090 . 3092 [RFC6805] King, D., Ed. and A. Farrel, Ed., "The Application of the 3093 Path Computation Element Architecture to the Determination 3094 of a Sequence of Domains in MPLS and GMPLS", RFC 6805, 3095 DOI 10.17487/RFC6805, November 2012, 3096 . 3098 [RFC7149] Boucadair, M. and C. Jacquenet, "Software-Defined 3099 Networking: A Perspective from within a Service Provider 3100 Environment", RFC 7149, DOI 10.17487/RFC7149, March 2014, 3101 . 3103 [RFC7390] Rahman, A., Ed. and E. Dijk, Ed., "Group Communication for 3104 the Constrained Application Protocol (CoAP)", RFC 7390, 3105 DOI 10.17487/RFC7390, October 2014, 3106 . 3108 [RFC7471] Giacalone, S., Ward, D., Drake, J., Atlas, A., and S. 3109 Previdi, "OSPF Traffic Engineering (TE) Metric 3110 Extensions", RFC 7471, DOI 10.17487/RFC7471, March 2015, 3111 . 3113 [RFC7679] Almes, G., Kalidindi, S., Zekauskas, M., and A. Morton, 3114 Ed., "A One-Way Delay Metric for IP Performance Metrics 3115 (IPPM)", STD 81, RFC 7679, DOI 10.17487/RFC7679, January 3116 2016, . 3118 [RFC7680] Almes, G., Kalidindi, S., Zekauskas, M., and A. Morton, 3119 Ed., "A One-Way Loss Metric for IP Performance Metrics 3120 (IPPM)", STD 82, RFC 7680, DOI 10.17487/RFC7680, January 3121 2016, . 3123 [RFC7752] Gredler, H., Ed., Medved, J., Previdi, S., Farrel, A., and 3124 S. Ray, "North-Bound Distribution of Link-State and 3125 Traffic Engineering (TE) Information Using BGP", RFC 7752, 3126 DOI 10.17487/RFC7752, March 2016, 3127 . 3129 [RFC7810] Previdi, S., Ed., Giacalone, S., Ward, D., Drake, J., and 3130 Q. Wu, "IS-IS Traffic Engineering (TE) Metric Extensions", 3131 RFC 7810, DOI 10.17487/RFC7810, May 2016, 3132 . 3134 [RFC7923] Voit, E., Clemm, A., and A. Gonzalez Prieto, "Requirements 3135 for Subscription to YANG Datastores", RFC 7923, 3136 DOI 10.17487/RFC7923, June 2016, 3137 . 3139 [RFC7950] Bjorklund, M., Ed., "The YANG 1.1 Data Modeling Language", 3140 RFC 7950, DOI 10.17487/RFC7950, August 2016, 3141 . 3143 [RFC8040] Bierman, A., Bjorklund, M., and K. Watsen, "RESTCONF 3144 Protocol", RFC 8040, DOI 10.17487/RFC8040, January 2017, 3145 . 3147 [RFC8051] Zhang, X., Ed. and I. Minei, Ed., "Applicability of a 3148 Stateful Path Computation Element (PCE)", RFC 8051, 3149 DOI 10.17487/RFC8051, January 2017, 3150 . 3152 [RFC8231] Crabbe, E., Minei, I., Medved, J., and R. Varga, "Path 3153 Computation Element Communication Protocol (PCEP) 3154 Extensions for Stateful PCE", RFC 8231, 3155 DOI 10.17487/RFC8231, September 2017, 3156 . 3158 [RFC8281] Crabbe, E., Minei, I., Sivabalan, S., and R. Varga, "Path 3159 Computation Element Communication Protocol (PCEP) 3160 Extensions for PCE-Initiated LSP Setup in a Stateful PCE 3161 Model", RFC 8281, DOI 10.17487/RFC8281, December 2017, 3162 . 3164 [RFC8283] Farrel, A., Ed., Zhao, Q., Ed., Li, Z., and C. Zhou, "An 3165 Architecture for Use of PCE and the PCE Communication 3166 Protocol (PCEP) in a Network with Central Control", 3167 RFC 8283, DOI 10.17487/RFC8283, December 2017, 3168 . 3170 [RFC8453] Ceccarelli, D., Ed. and Y. Lee, Ed., "Framework for 3171 Abstraction and Control of TE Networks (ACTN)", RFC 8453, 3172 DOI 10.17487/RFC8453, August 2018, 3173 . 3175 [RFC8571] Ginsberg, L., Ed., Previdi, S., Wu, Q., Tantsura, J., and 3176 C. Filsfils, "BGP - Link State (BGP-LS) Advertisement of 3177 IGP Traffic Engineering Performance Metric Extensions", 3178 RFC 8571, DOI 10.17487/RFC8571, March 2019, 3179 . 3181 [RFC8661] Bashandy, A., Ed., Filsfils, C., Ed., Previdi, S., 3182 Decraene, B., and S. Litkowski, "Segment Routing MPLS 3183 Interworking with LDP", RFC 8661, DOI 10.17487/RFC8661, 3184 December 2019, . 3186 [RR94] Rodrigues, M. and K. Ramakrishnan, "Optimal Routing in 3187 Shortest Path Networks", Proceedings ITS'94, Rio de 3188 Janeiro, Brazil, 1994. 3190 [SLDC98] Suter, B., Lakshman, T., Stiliadis, D., and A. Choudhury, 3191 "Design Considerations for Supporting TCP with Per-flow 3192 Queueing", Proceedings INFOCOM'98, p. 299-306, 1998. 3194 [WANG] Wang, Y., Wang, Z., and L. Zhang, "Internet traffic 3195 engineering without full mesh overlaying", 3196 Proceedings INFOCOM'2001, April 2001. 3198 [XIAO] Xiao, X., Hannan, A., Bailey, B., and L. Ni, "Traffic 3199 Engineering with MPLS in the Internet", Article IEEE 3200 Network Magazine, March 2000. 3202 [YARE95] Yang, C. and A. Reddy, "A Taxonomy for Congestion Control 3203 Algorithms in Packet Switching Networks", Article IEEE 3204 Network Magazine, p. 34-45, 1995. 3206 Appendix A. Historic Overview 3208 A.1. Traffic Engineering in Classical Telephone Networks 3210 This subsection presents a brief overview of traffic engineering in 3211 telephone networks which often relates to the way user traffic is 3212 steered from an originating node to the terminating node. This 3213 subsection presents a brief overview of this topic. A detailed 3214 description of the various routing strategies applied in telephone 3215 networks is included in the book by G. Ash [ASH2]. 3217 The early telephone network relied on static hierarchical routing, 3218 whereby routing patterns remained fixed independent of the state of 3219 the network or time of day. The hierarchy was intended to 3220 accommodate overflow traffic, improve network reliability via 3221 alternate routes, and prevent call looping by employing strict 3222 hierarchical rules. The network was typically over-provisioned since 3223 a given fixed route had to be dimensioned so that it could carry user 3224 traffic during a busy hour of any busy day. Hierarchical routing in 3225 the telephony network was found to be too rigid upon the advent of 3226 digital switches and stored program control which were able to manage 3227 more complicated traffic engineering rules. 3229 Dynamic routing was introduced to alleviate the routing inflexibility 3230 in the static hierarchical routing so that the network would operate 3231 more efficiently. This resulted in significant economic gains 3232 [HUSS87]. Dynamic routing typically reduces the overall loss 3233 probability by 10 to 20 percent (compared to static hierarchical 3234 routing). Dynamic routing can also improve network resilience by 3235 recalculating routes on a per-call basis and periodically updating 3236 routes. 3238 There are three main types of dynamic routing in the telephone 3239 network. They are time-dependent routing, state-dependent routing 3240 (SDR), and event dependent routing (EDR). 3242 In time-dependent routing, regular variations in traffic loads (such 3243 as time of day or day of week) are exploited in pre-planned routing 3244 tables. In state-dependent routing, routing tables are updated 3245 online according to the current state of the network (e.g., traffic 3246 demand, utilization, etc.). In event dependent routing, routing 3247 changes are incepted by events (such as call setups encountering 3248 congested or blocked links) whereupon new paths are searched out 3249 using learning models. EDR methods are real-time adaptive, but they 3250 do not require global state information as does SDR. Examples of EDR 3251 schemes include the dynamic alternate routing (DAR) from BT, the 3252 state-and-time dependent routing (STR) from NTT, and the success-to- 3253 the-top (STT) routing from AT&T. 3255 Dynamic non-hierarchical routing (DNHR) is an example of dynamic 3256 routing that was introduced in the AT&T toll network in the 1980's to 3257 respond to time-dependent information such as regular load variations 3258 as a function of time. Time-dependent information in terms of load 3259 may be divided into three timescales: hourly, weekly, and yearly. 3260 Correspondingly, three algorithms are defined to pre-plan the routing 3261 tables. The network design algorithm operates over a year-long 3262 interval while the demand servicing algorithm operates on a weekly 3263 basis to fine tune link sizes and routing tables to correct forecast 3264 errors on the yearly basis. At the smallest timescale, the routing 3265 algorithm is used to make limited adjustments based on daily traffic 3266 variations. Network design and demand servicing are computed using 3267 offline calculations. Typically, the calculations require extensive 3268 searches on possible routes. On the other hand, routing may need 3269 online calculations to handle crankback. DNHR adopts a "two-link" 3270 approach whereby a path can consist of two links at most. The 3271 routing algorithm presents an ordered list of route choices between 3272 an originating switch and a terminating switch. If a call overflows, 3273 a via switch (a tandem exchange between the originating switch and 3274 the terminating switch) would send a crankback signal to the 3275 originating switch. This switch would then select the next route, 3276 and so on, until there are no alternative routes available in which 3277 the call is blocked. 3279 A.2. Evolution of Traffic Engineering in Packet Networks 3281 This subsection reviews related prior work that was intended to 3282 improve the performance of data networks. Indeed, optimization of 3283 the performance of data networks started in the early days of the 3284 ARPANET. Other early commercial networks such as SNA also recognized 3285 the importance of performance optimization and service 3286 differentiation. 3288 In terms of traffic management, the Internet has been a best effort 3289 service environment until recently. In particular, very limited 3290 traffic management capabilities existed in IP networks to provide 3291 differentiated queue management and scheduling services to packets 3292 belonging to different classes. 3294 In terms of routing control, the Internet has employed distributed 3295 protocols for intra-domain routing. These protocols are highly 3296 scalable and resilient. However, they are based on simple algorithms 3297 for path selection which have very limited functionality to allow 3298 flexible control of the path selection process. 3300 In the following subsections, the evolution of practical traffic 3301 engineering mechanisms in IP networks and its predecessors are 3302 reviewed. 3304 A.2.1. Adaptive Routing in the ARPANET 3306 The early ARPANET recognized the importance of adaptive routing where 3307 routing decisions were based on the current state of the network 3308 [MCQ80]. Early minimum delay routing approaches forwarded each 3309 packet to its destination along a path for which the total estimated 3310 transit time was the smallest. Each node maintained a table of 3311 network delays, representing the estimated delay that a packet would 3312 experience along a given path toward its destination. The minimum 3313 delay table was periodically transmitted by a node to its neighbors. 3314 The shortest path, in terms of hop count, was also propagated to give 3315 the connectivity information. 3317 One drawback to this approach is that dynamic link metrics tend to 3318 create "traffic magnets" causing congestion to be shifted from one 3319 location of a network to another location, resulting in oscillation 3320 and network instability. 3322 A.2.2. Dynamic Routing in the Internet 3324 The Internet evolved from the ARPANET and adopted dynamic routing 3325 algorithms with distributed control to determine the paths that 3326 packets should take en-route to their destinations. The routing 3327 algorithms are adaptations of shortest path algorithms where costs 3328 are based on link metrics. The link metric can be based on static or 3329 dynamic quantities. The link metric based on static quantities may 3330 be assigned administratively according to local criteria. The link 3331 metric based on dynamic quantities may be a function of a network 3332 congestion measure such as delay or packet loss. 3334 It was apparent early that static link metric assignment was 3335 inadequate because it can easily lead to unfavorable scenarios in 3336 which some links become congested while others remain lightly loaded. 3337 One of the many reasons for the inadequacy of static link metrics is 3338 that link metric assignment was often done without considering the 3339 traffic matrix in the network. Also, the routing protocols did not 3340 take traffic attributes and capacity constraints into account when 3341 making routing decisions. This results in traffic concentration 3342 being localized in subsets of the network infrastructure and 3343 potentially causing congestion. Even if link metrics are assigned in 3344 accordance with the traffic matrix, unbalanced loads in the network 3345 can still occur due to a number factors including: 3347 o Resources may not be deployed in the most optimal locations from a 3348 routing perspective. 3350 o Forecasting errors in traffic volume and/or traffic distribution. 3352 o Dynamics in traffic matrix due to the temporal nature of traffic 3353 patterns, BGP policy change from peers, etc. 3355 The inadequacy of the legacy Internet interior gateway routing system 3356 is one of the factors motivating the interest in path oriented 3357 technology with explicit routing and constraint-based routing 3358 capability such as MPLS. 3360 A.2.3. ToS Routing 3362 Type-of-Service (ToS) routing involves different routes going to the 3363 same destination with selection dependent upon the ToS field of an IP 3364 packet [RFC2474]. The ToS classes may be classified as low delay and 3365 high throughput. Each link is associated with multiple link costs 3366 and each link cost is used to compute routes for a particular ToS. A 3367 separate shortest path tree is computed for each ToS. The shortest 3368 path algorithm must be run for each ToS resulting in very expensive 3369 computation. Classical ToS-based routing is now outdated as the IP 3370 header field has been replaced by a Diffserv field. Effective 3371 traffic engineering is difficult to perform in classical ToS-based 3372 routing because each class still relies exclusively on shortest path 3373 routing which results in localization of traffic concentration within 3374 the network. 3376 A.2.4. Equal Cost Multi-Path 3378 Equal Cost Multi-Path (ECMP) is another technique that attempts to 3379 address the deficiency in the Shortest Path First (SPF) interior 3380 gateway routing systems [RFC2328]. In the classical SPF algorithm, 3381 if two or more shortest paths exist to a given destination, the 3382 algorithm will choose one of them. The algorithm is modified 3383 slightly in ECMP so that if two or more equal cost shortest paths 3384 exist between two nodes, the traffic between the nodes is distributed 3385 among the multiple equal-cost paths. Traffic distribution across the 3386 equal-cost paths is usually performed in one of two ways: (1) packet- 3387 based in a round-robin fashion, or (2) flow-based using hashing on 3388 source and destination IP addresses and possibly other fields of the 3389 IP header. The first approach can easily cause out- of-order packets 3390 while the second approach is dependent upon the number and 3391 distribution of flows. Flow-based load sharing may be unpredictable 3392 in an enterprise network where the number of flows is relatively 3393 small and less heterogeneous (for example, hashing may not be 3394 uniform), but it is generally effective in core public networks where 3395 the number of flows is large and heterogeneous. 3397 In ECMP, link costs are static and bandwidth constraints are not 3398 considered, so ECMP attempts to distribute the traffic as equally as 3399 possible among the equal-cost paths independent of the congestion 3400 status of each path. As a result, given two equal-cost paths, it is 3401 possible that one of the paths will be more congested than the other. 3402 Another drawback of ECMP is that load sharing cannot be achieved on 3403 multiple paths which have non-identical costs. 3405 A.2.5. Nimrod 3407 Nimrod was a routing system developed to provide heterogeneous 3408 service specific routing in the Internet, while taking multiple 3409 constraints into account [RFC1992]. Essentially, Nimrod was a link 3410 state routing protocol to support path oriented packet forwarding. 3411 It used the concept of maps to represent network connectivity and 3412 services at multiple levels of abstraction. Mechanisms allowed 3413 restriction of the distribution of routing information. 3415 Even though Nimrod did not enjoy deployment in the public Internet, a 3416 number of key concepts incorporated into the Nimrod architecture, 3417 such as explicit routing which allows selection of paths at 3418 originating nodes, are beginning to find applications in some recent 3419 constraint-based routing initiatives. 3421 A.3. Development of Internet Traffic Engineering 3423 A.3.1. Overlay Model 3425 In the overlay model, a virtual-circuit network, such as Sonet/SDH, 3426 OTN, or WDM, provides virtual-circuit connectivity between routers 3427 that are located at the edges of a virtual-circuit cloud. In this 3428 mode, two routers that are connected through a virtual circuit see a 3429 direct adjacency between themselves independent of the physical route 3430 taken by the virtual circuit through the ATM, frame relay, or WDM 3431 network. Thus, the overlay model essentially decouples the logical 3432 topology that routers see from the physical topology that the ATM, 3433 frame relay, or WDM network manages. The overlay model based on ATM 3434 or frame relay enables a network administrator or an automaton to 3435 employ traffic engineering concepts to perform path optimization by 3436 re-configuring or rearranging the virtual circuits so that a virtual 3437 circuit on a congested or sub-optimal physical link can be re-routed 3438 to a less congested or more optimal one. In the overlay model, 3439 traffic engineering is also employed to establish relationships 3440 between the traffic management parameters (e.g., PCR, SCR, and MBS 3441 for ATM) of the virtual-circuit technology and the actual traffic 3442 that traverses each circuit. These relationships can be established 3443 based upon known or projected traffic profiles, and some other 3444 factors. 3446 Appendix B. Overview of Traffic Engineering Related Work in Other SDOs 3448 B.1. Overview of ITU Activities Related to Traffic Engineering 3450 This section provides an overview of prior work within the ITU-T 3451 pertaining to traffic engineering in traditional telecommunications 3452 networks. 3454 ITU-T Recommendations E.600 [ITU-E600], E.701 [ITU-E701], and E.801 3455 [ITU-E801] address traffic engineering issues in traditional 3456 telecommunications networks. Recommendation E.600 provides a 3457 vocabulary for describing traffic engineering concepts, while E.701 3458 defines reference connections, Grade of Service (GOS), and traffic 3459 parameters for ISDN. Recommendation E.701 uses the concept of a 3460 reference connection to identify representative cases of different 3461 types of connections without describing the specifics of their actual 3462 realizations by different physical means. As defined in 3463 Recommendation E.600, "a connection is an association of resources 3464 providing means for communication between two or more devices in, or 3465 attached to, a telecommunication network." Also, E.600 defines "a 3466 resource as any set of physically or conceptually identifiable 3467 entities within a telecommunication network, the use of which can be 3468 unambiguously determined" [ITU-E600]. There can be different types 3469 of connections as the number and types of resources in a connection 3470 may vary. 3472 Typically, different network segments are involved in the path of a 3473 connection. For example, a connection may be local, national, or 3474 international. The purposes of reference connections are to clarify 3475 and specify traffic performance issues at various interfaces between 3476 different network domains. Each domain may consist of one or more 3477 service provider networks. 3479 Reference connections provide a basis to define grade of service 3480 (GoS) parameters related to traffic engineering within the ITU-T 3481 framework. As defined in E.600, "GoS refers to a number of traffic 3482 engineering variables which are used to provide a measure of the 3483 adequacy of a group of resources under specified conditions." These 3484 GoS variables may be probability of loss, dial tone, delay, etc. 3486 They are essential for network internal design and operation as well 3487 as for component performance specification. 3489 GoS is different from quality of service (QoS) in the ITU framework. 3490 QoS is the performance perceivable by a telecommunication service 3491 user and expresses the user's degree of satisfaction of the service. 3492 QoS parameters focus on performance aspects observable at the service 3493 access points and network interfaces, rather than their causes within 3494 the network. GoS, on the other hand, is a set of network oriented 3495 measures which characterize the adequacy of a group of resources 3496 under specified conditions. For a network to be effective in serving 3497 its users, the values of both GoS and QoS parameters must be related, 3498 with GoS parameters typically making a major contribution to the QoS. 3500 Recommendation E.600 stipulates that a set of GoS parameters must be 3501 selected and defined on an end-to-end basis for each major service 3502 category provided by a network to assist the network provider with 3503 improving efficiency and effectiveness of the network. Based on a 3504 selected set of reference connections, suitable target values are 3505 assigned to the selected GoS parameters under normal and high load 3506 conditions. These end-to-end GoS target values are then apportioned 3507 to individual resource components of the reference connections for 3508 dimensioning purposes. 3510 Author's Address 3512 Adrian Farrel (editor) 3513 Old Dog Consulting 3515 Email: adrian@olddog.co.uk