idnits 2.17.1 draft-ietf-dime-overload-reqs-00.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 (September 21, 2012) is 4206 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) No issues found here. Summary: 0 errors (**), 0 flaws (~~), 1 warning (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group E. McMurry 3 Internet-Draft B. Campbell 4 Intended status: Standards Track Tekelec 5 Expires: March 25, 2013 September 21, 2012 7 Diameter Overload Control Requirements 8 draft-ietf-dime-overload-reqs-00 10 Abstract 12 When a Diameter server or agent becomes overloaded, it needs to be 13 able to gracefully reduce its load, typically by informing clients to 14 reduce sending traffic for some period of time. Otherwise, it must 15 continue to expend resources parsing and responding to Diameter 16 messages, possibly resulting in congestion collapse. The existing 17 mechanisms provided by Diameter are not sufficient for this purpose. 18 This document describes the limitations of the existing mechanisms, 19 and provides requirements for new overload management mechanisms. 21 Status of this Memo 23 This Internet-Draft is submitted in full conformance with the 24 provisions of BCP 78 and BCP 79. 26 Internet-Drafts are working documents of the Internet Engineering 27 Task Force (IETF). Note that other groups may also distribute 28 working documents as Internet-Drafts. The list of current Internet- 29 Drafts is at http://datatracker.ietf.org/drafts/current/. 31 Internet-Drafts are draft documents valid for a maximum of six months 32 and may be updated, replaced, or obsoleted by other documents at any 33 time. It is inappropriate to use Internet-Drafts as reference 34 material or to cite them other than as "work in progress." 36 This Internet-Draft will expire on March 25, 2013. 38 Copyright Notice 40 Copyright (c) 2012 IETF Trust and the persons identified as the 41 document authors. All rights reserved. 43 This document is subject to BCP 78 and the IETF Trust's Legal 44 Provisions Relating to IETF Documents 45 (http://trustee.ietf.org/license-info) in effect on the date of 46 publication of this document. Please review these documents 47 carefully, as they describe your rights and restrictions with respect 48 to this document. Code Components extracted from this document must 49 include Simplified BSD License text as described in Section 4.e of 50 the Trust Legal Provisions and are provided without warranty as 51 described in the Simplified BSD License. 53 Table of Contents 55 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 56 1.1. Causes of Overload . . . . . . . . . . . . . . . . . . . . 3 57 1.2. Effects of Overload . . . . . . . . . . . . . . . . . . . 5 58 1.3. Overload vs. Network Congestion . . . . . . . . . . . . . 5 59 1.4. Diameter Applications in a Broader Network . . . . . . . . 5 60 1.5. Documentation Conventions . . . . . . . . . . . . . . . . 6 61 2. Overload Scenarios . . . . . . . . . . . . . . . . . . . . . . 6 62 2.1. Peer to Peer Scenarios . . . . . . . . . . . . . . . . . . 7 63 2.2. Agent Scenarios . . . . . . . . . . . . . . . . . . . . . 9 64 2.3. Interconnect Scenario . . . . . . . . . . . . . . . . . . 12 65 3. Existing Mechanisms . . . . . . . . . . . . . . . . . . . . . 13 66 4. Issues with the Current Mechanisms . . . . . . . . . . . . . . 14 67 4.1. Problems with Implicit Mechanism . . . . . . . . . . . . . 15 68 4.2. Problems with Explicit Mechanisms . . . . . . . . . . . . 15 69 5. Diameter Overload Case Studies . . . . . . . . . . . . . . . . 16 70 5.1. Overload in Mobile Data Networks . . . . . . . . . . . . . 16 71 5.2. 3GPP Study on Core Network Overload . . . . . . . . . . . 17 72 6. Solution Requirements . . . . . . . . . . . . . . . . . . . . 17 73 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22 74 8. Security Considerations . . . . . . . . . . . . . . . . . . . 22 75 8.1. Access Control . . . . . . . . . . . . . . . . . . . . . . 23 76 8.2. Denial-of-Service Attacks . . . . . . . . . . . . . . . . 23 77 8.3. Replay Attacks . . . . . . . . . . . . . . . . . . . . . . 23 78 8.4. Man-in-the-Middle Attacks . . . . . . . . . . . . . . . . 24 79 8.5. Compromised Hosts . . . . . . . . . . . . . . . . . . . . 24 80 9. References . . . . . . . . . . . . . . . . . . . . . . . . . . 24 81 9.1. Normative References . . . . . . . . . . . . . . . . . . . 24 82 9.2. Informative References . . . . . . . . . . . . . . . . . . 25 83 Appendix A. Contributors . . . . . . . . . . . . . . . . . . . . 25 84 Appendix B. Acknowledgements . . . . . . . . . . . . . . . . . . 25 85 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 25 87 1. Introduction 89 When a Diameter [I-D.ietf-dime-rfc3588bis] server or agent becomes 90 overloaded, it needs to be able to gracefully reduce its load, 91 typically by informing clients to reduce sending traffic for some 92 period of time. Otherwise, it must continue to expend resources 93 parsing and responding to Diameter messages, possibly resulting in 94 congestion collapse. The existing mechanisms provided by Diameter 95 are not sufficient for this purpose. This document describes the 96 limitations of the existing mechanisms, and provides requirements for 97 new overload management mechanisms. 99 This document draws on [RFC5390] and the work done on SIP overload 100 control as well as on overload practices in SS7 networks and studies 101 done by 3GPP. 103 Diameter is not typically an end-user protocol; rather it is 104 generally used as one component in support of some end-user activity. 105 For example, a WiFi access point might use Diameter to authenticate 106 and authorize user access via 802.11. Overload in a network that 107 uses Diameter applications will likely spill over into the end-user 108 application network. The impact of Diameter overload on the client 109 application (a client application may use the Diameter protocol and 110 other protocols to do its job) is beyond the scope of this document. 112 This document presents non-normative descriptions of causes of 113 overload along with related scenarios and studies. Finally, it 114 offers a set of normative requirements for an improved overload 115 indication mechanism. 117 1.1. Causes of Overload 119 Overload occurs when an element, such as a Diameter server or agent, 120 has insufficient resources to successfully process all of the traffic 121 it is receiving. Resources include all of the capabilities of the 122 element used to process a request, including CPU processing, memory, 123 I/O, and disk resources. It can also include external resources such 124 as a database or DNS server, in which case the CPU, processing, 125 memory, I/O, and disk resources of those elements are effectively 126 part of the logical element processing the request. 128 Overload can occur for many reasons, including: 130 Inadequate capacity: When designing Diameter networks, that is, 131 application layer multi-node Diameter deployments, it can be very 132 difficult to predict all scenarios that may cause elevated 133 traffic. It may also be more costly to implement support for some 134 scenarios than a network operator may deem worthwhile. This 135 results in the likelihood that a Diameter network will not have 136 adequate capacity to handle all situations. 138 Dependency failures: A Diameter node can become overloaded because a 139 resource on which it is dependent has failed or become overloaded, 140 greatly reducing the logical capacity of the node. In these 141 cases, even minimal traffic might cause the node to go into 142 overload. Examples of such dependency overloads include DNS 143 servers, databases, disks, and network interfaces. 145 Component failures: A Diameter node can become overloaded when it is 146 a member of a cluster of servers that each share the load of 147 traffic, and one or more of the other members in the cluster fail. 148 In this case, the remaining nodes take over the work of the failed 149 nodes. Normally, capacity planning takes such failures into 150 account, and servers are typically run with enough spare capacity 151 to handle failure of another node. However, unusual failure 152 conditions can cause many nodes to fail at once. This is often 153 the case with software failures, where a bad packet or bad 154 database entry hits the same bug in a set of nodes in a cluster. 156 Network Initiated Traffic Flood: Issues with the radio access 157 network in a mobile network such as radio overlays with frequent 158 handovers, and operational changes are examples of network events 159 that can precipitate a flood of Diameter signaling traffic, such 160 as an avalanche restart. Failure of a Diameter proxy may also 161 result in a large amount of signaling as connections and sessions 162 are reestablished. 164 Subscriber Initiated Traffic Flood: Large gatherings of subscribers 165 or events that result in many subscribers interacting with the 166 network in close time proximity can result in Diameter signaling 167 traffic floods. For example, the finale of a large fireworks show 168 could be immediately followed by many subscribers posting 169 messages, pictures, and videos concentrated on one portion of a 170 network. Subscriber devices, such as smartphones, may use 171 aggressive registration strategies that generate unusually high 172 Diameter traffic loads. 174 DoS attacks: An attacker, wishing to disrupt service in the network, 175 can cause a large amount of traffic to be launched at a target 176 element. This can be done from a central source of traffic or 177 through a distributed DoS attack. In all cases, the volume of 178 traffic well exceeds the capacity of the element, sending the 179 system into overload. 181 1.2. Effects of Overload 183 Modern Diameter networks, comprised of application layer multi-node 184 deployments of Diameter elements, may operate at very large 185 transaction volumes. If a Diameter node becomes overloaded, or even 186 worse, fails completely, a large number of messages may be lost very 187 quickly. Even with redundant servers, many messages can be lost in 188 the time it takes for failover to complete. While a Diameter client 189 or agent should be able to retry such requests, an overloaded peer 190 may cause a sudden large increase in the number of transaction 191 transactions needing to be retried, rapidly filling local queues or 192 otherwise contributing to local overload. Therefore Diameter devices 193 need to be able to shed load before critical failures can occur. 195 Diameter depends heavily on The "Authentication, Authorization, 196 and Accounting (AAA) Transport Profile" [RFC3539], which states 197 assumptions about the scale of AAA services which may be incorrect 198 for current uses of Diameter. In particular, the document 199 suggests that AAA services will typically be low volume and that 200 traffic will typically be application-driven. Section 2.1 of that 201 document uses an example of a 48 port NAS. However, Diameter is 202 commonly used in large-scale mobile data environments, where a 203 typical client could be a packet gateway that serves millions of 204 users, and generates Diameter messages at network-driven rates. 206 1.3. Overload vs. Network Congestion 208 This document uses the term "overload" to refer to application-layer 209 overload at Diameter nodes. This is distinct from "network 210 congestion", that is, congestion that occurs at the lower networking 211 layers that may impact the delivery of Diameter messages between 212 nodes. The authors recognize that element overload and network 213 congestion are interrelated, and that overload can contribute to 214 network congestion and vice versa. 216 Network congestion issues are better handled by the transport 217 protocols. Diameter uses TCP and SCTP, both of which include 218 congestion management features. Analysis of whether those features 219 are sufficient for transport level congestion between Diameter nodes, 220 and any work to further mitigate network congestion is out of scope 221 both for this document, and for the work proposed by this document. 223 1.4. Diameter Applications in a Broader Network 225 Most elements using Diameter applications do not use Diameter 226 exclusively. It is important to realize that overload of an element 227 can be caused by a number of factors that may be unrelated to the 228 processing of Diameter or Diameter applications. 230 A element communicating via protocols other than Diameter that is 231 also using a Diameter application needs to be able to signal to 232 Diameter peers that it is experiencing overload regardless of the 233 cause of the overload, since the overload will affect that element's 234 ability to process Diameter transactions. The element may also need 235 to signal this on other protocols depending on its function and the 236 architecture of the network and application it is providing services 237 for. Whether that is necessary can only be decided within the 238 context of that architecture and application. A mechanism for 239 signaling overload with Diameter, which this specification details 240 the requirements for, provides applications the ability to signal 241 their Diameter peers of overload, mitigating that part of the issue. 242 Applications may need to use this, as well as other mechanisms, to 243 solve their broader overload issues. Indicating overload on 244 protocols other than Diameter is out of scope for this document, and 245 for the work proposed by this document. 247 1.5. Documentation Conventions 249 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 250 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 251 document are to be interpreted as described in [RFC2119]. 253 The terms "client", "server", "agent", "node", "peer", "upstream", 254 and "downstream" are used as defined in [I-D.ietf-dime-rfc3588bis]. 256 2. Overload Scenarios 258 Several Diameter deployment scenarios exist that may impact overload 259 management. The following scenarios help motivate the requirements 260 for an overload management mechanism. 262 These scenarios are by no means exhaustive, and are in general 263 simplified for the sake of clarity. In particular, the authors 264 assume for the sake of clarity that the client sends Diameter 265 requests to the server, and the server sends responses to client, 266 even though Diameter supports bidirectional applications. Each 267 direction in such an application can be modeled separately. 269 In a large scale deployment, many of the nodes represented in these 270 scenarios would be deployed as clusters of servers. The authors 271 assume that such a cluster is responsible for managing its own 272 internal load balancing and overload management so that it appears as 273 a single Diameter node. That is, other Diameter nodes can treat it 274 as single, monolithic node for the purposes of overload management. 276 These scenarios do not illustrate the client application. As 277 mentioned in Section 1, Diameter is not typically an end-user 278 protocol; rather it is generally used in support of some other client 279 application. These scenarios do not consider the impact of Diameter 280 overload on the client application. 282 2.1. Peer to Peer Scenarios 284 This section describes Diameter peer-to-peer scenarios. That is, 285 scenarios where a Diameter client talks directly with a Diameter 286 server, without the use of a Diameter agent. 288 Figure 1 illustrates the simplest possible Diameter relationship. 289 The client and server share a one-to-one peer-to-peer relationship. 290 If the server becomes overloaded, either because the client exceeds 291 the server's capacity, or because the server's capacity is reduced 292 due to some resource dependency, the client needs to reduce the 293 amount of Diameter traffic it sends to the server. Since the client 294 cannot forward requests to another server, it must either queue 295 requests until the server recovers, or itself become overloaded in 296 the context of the client application and other protocols it may also 297 use. 299 +------------------+ 300 | | 301 | | 302 | Server | 303 | | 304 +--------+---------+ 305 | 306 | 307 +--------+---------+ 308 | | 309 | | 310 | Client | 311 | | 312 +------------------+ 314 Figure 1: Basic Peer to Peer Scenario 316 Figure 2 shows a similar scenario, except in this case the client has 317 multiple servers that can handle work for a specific realm and 318 application. If server 1 becomes overloaded, the client can forward 319 traffic to server 2. Assuming server 2 has sufficient reserve 320 capacity to handle the forwarded traffic, the client should be able 321 to continue serving client application protocol users. If server 1 322 is approaching overload, but can still handle some number of new 323 request, it needs to be able to instruct the client to forward a 324 subset of its traffic to server 2. 326 +------------------+ +------------------+ 327 | | | | 328 | | | | 329 | Server 1 | | Server 2 | 330 | | | | 331 +--------+-`.------+ +------.'+---------+ 332 `. .' 333 `. .' 334 `. .' 335 `. .' 336 +-------`.'--------+ 337 | | 338 | | 339 | Client | 340 | | 341 +------------------+ 343 Figure 2: Multiple Server Peer to Peer Scenario 345 Figure 3 illustrates a peer-to-peer scenario with multiple Diameter 346 realm and application combinations. In this example, server 2 can 347 handle work for both applications. Each application might have 348 different resource dependencies. For example, a server might need to 349 access one database for application A, and another for application B. 350 This creates a possibility that Server 2 could become overloaded for 351 application A but not for application B, in which case the client 352 would need to divert some part of its application A requests to 353 server 1, but should not divert any application B requests. This 354 requires server 2 to be able to distinguish between applications when 355 it indicates an overload condition to the client. 357 On the other hand, it's possible that the servers host many 358 applications. If server 2 becomes overloaded for all applications, 359 it would be undesirable for it to have to notify the client 360 separately for each application. Therefore it also needs a way to 361 indicate that it is overloaded for all possible applications. 363 +----------------------------------------------+ 364 | Application A +------------------------+----------------------+ 365 |+------------------+ | +------------------+ | +------------------+| 366 || | | | | | | || 367 || | | | | | | || 368 || Server 1 | | | Server 2 | | | Server 3 || 369 || | | | | | | || 370 |+--------+---------+ | +--------+---------+ | +-+----------------+| 371 | | | | | | | 372 +---------+-----------+-----------+------------+ | | 373 | | | | | 374 | | | | Application B | 375 | +-----------+-----------------+-----------------+ 376 ``-.._ | | 377 `-..__ | _.-'' 378 `--._ | _.-'' 379 ``-.__ | _.-'' 380 +------`-.-''------+ 381 | | 382 | | 383 | Client | 384 | | 385 +------------------+ 387 Figure 3: Multiple Application Peer to Peer Scenario 389 2.2. Agent Scenarios 391 This section describes scenarios that include a Diameter agent, 392 either in the form of a Diameter relay or Diameter proxy. These 393 scenarios do not consider Diameter redirect agents, since they are 394 more readily modeled as end-servers. 396 Figure 4 illustrates a simple Diameter agent scenario with a single 397 client, agent, and server. In this case, overload can occur at the 398 server, at the agent, or both. But in most cases, client behavior is 399 the same whether overload occurs at the server or at the agent. From 400 the client's perspective, server overload and agent overload is the 401 same thing. 403 +------------------+ 404 | | 405 | | 406 | Server | 407 | | 408 +--------+---------+ 409 | 410 | 411 +--------+---------+ 412 | | 413 | | 414 | Agent | 415 | | 416 +--------+---------+ 417 | 418 | 419 +--------+---------+ 420 | | 421 | | 422 | Client | 423 | | 424 +------------------+ 426 Figure 4: Basic Agent Scenario 428 Figure 5 shows an agent scenario with multiple servers. If server 1 429 becomes overloaded, but server 2 has sufficient reserve capacity, the 430 agent may be able to transparently divert some or all Diameter 431 requests originally bound for server 1 to server 2. 433 In most cases, the client does not have detailed knowledge of the 434 Diameter topology upstream of the agent. If the agent uses dynamic 435 discovery to find eligible servers, the set of eligible servers may 436 not be enumerable from the perspective of the client. Therefore, in 437 most cases the agent needs to deal with any upstream overload issues 438 in a way that is transparent to the client. If one server notifies 439 the agent that it has become overloaded, the notification should not 440 be passed back to the client in a way where the client could 441 mistakenly perceive the agent itself as being overloaded. If the set 442 of all possible destinations upstream of the agent no longer has 443 sufficient capacity for incoming load, the agent itself becomes 444 effectively overloaded. 446 On the other hand, there are cases where the client needs to be able 447 to select a particular server from behind an agent. For example, if 448 a Diameter request is part of a multiple-round-trip authentication, 449 or is otherwise part of a Diameter "session", it may have a 450 DestinationHost AVP that requires the request to be served by server 451 1. Therefore the agent may need to inform a client that a particular 452 upstream server is overloaded or otherwise unavailable. Note that 453 there can be many ways a server can be specified, which may have 454 different implications (e.g. by IP address, by host name, etc). 456 +------------------+ +------------------+ 457 | | | | 458 | | | | 459 | Server 1 | | Server 2 | 460 | | | | 461 +--------+-`.------+ +------.'+---------+ 462 `. .' 463 `. .' 464 `. .' 465 `. .' 466 +-------`.'--------+ 467 | | 468 | | 469 | Agent | 470 | | 471 +--------+---------+ 472 | 473 | 474 | 475 +--------+---------+ 476 | | 477 | | 478 | Client | 479 | | 480 +------------------+ 482 Figure 5: Multiple Server Agent Scenario 484 Figure 6 shows a scenario where an agent routes requests to a set of 485 servers for more than one Diameter realm and application. In this 486 scenario, if server 1 becomes overloaded or unavailable, the agent 487 may effectively operate at reduced capacity for application A, but at 488 full capacity for application B. Therefore, the agent needs to be 489 able to report that it is overloaded for one application, but not for 490 another. 492 +----------------------------------------------+ 493 | Application A +------------------------+----------------------+ 494 |+------------------+ | +------------------+ | +------------------+| 495 || | | | | | | || 496 || | | | | | | || 497 || Server 1 | | | Server 2 | | | Server 3 || 498 || | | | | | | || 499 |+---------+--------+ | +--------+---------+ | +--+---------------+| 500 | | | | | | | 501 +----------+----------+-----------+------------+ | | 502 | | | | | 503 | | | | Application B | 504 | +-----------+------------------+----------------+ 505 | | | 506 ``--.__ | _. 507 ``-.__ | __.--'' 508 `--.._ | _..--' 509 +-----``-+.-''-----+ 510 | | 511 | | 512 | Agent | 513 | | 514 +--------+---------+ 515 | 516 | 517 +--------+---------+ 518 | | 519 | | 520 | Client | 521 | | 522 +------------------+ 524 Figure 6: Multiple Application Agent Scenario 526 2.3. Interconnect Scenario 528 Another scenario to consider when looking at Diameter overload is 529 that of multiple network operators using Diameter components 530 connected through an interconnect service, e.g. using IPX. Figure 7 531 shows two network operators with an interconnect network in-between. 532 There could be any number of these networks between any two network 533 operator's networks. 535 +-------------------------------------------+ 536 | Interconnect | 537 | | 538 | +--------------+ +--------------+ | 539 | | Edge Agent 1 |------| Edge Agent 2 | | 540 | +--------------+ +--------------+ | 541 | .' `. | 542 +------.-'--------------------------`.------+ 543 .' `. 544 .-' `. 545 ------------.'-----+ +----`.--------------- 546 +----------+ | | +----------+ 547 | Server 1 | | | | Server 2 | 548 +----------+ | | +----------+ 549 | | 550 Network Operator 1 | | Network Operator 2 551 -------------------+ +--------------------- 553 Figure 7: Two Network Interconnect Scenario 555 The characteristics of the information that an operator would want to 556 share over such a connection are different than the information 557 shared between components within a network operator's network. 558 Network operators may not want to convey topology or operational 559 information, which limits how overload and loading information can be 560 sent. For the interconnect scenario shown, Server 2 may want to 561 signal overload to Server 1, to affect traffic coming from Network 562 Operator 1. 564 This is different than internal to an network operator's network, 565 where there may be many more elements in a more complicated topology. 566 Also, the elements in the interconnect network may not support 567 diameter overload control, and the network operators may not want the 568 interconnect to use overload or loading information intended to pass 569 through the interconnect even if the elements in the interconnect 570 network do support diameter overload control. 572 3. Existing Mechanisms 574 Diameter offers both implicit and explicit mechanisms for a Diameter 575 node to learn that a peer is overloaded or unreachable. The implicit 576 mechanism is simply the lack of responses to requests. If a client 577 fails to receive a response in a certain time period, it assumes the 578 upstream peer is unavailable, or overloaded to the point of effective 579 unavailability. The watchdog mechanism [RFC3539] ensures that a 580 certain rate of transaction responses occur even when there is 581 otherwise little or no other Diameter traffic. 583 The explicit mechanism involves specific protocol error responses, 584 where an agent or server can tell a downstream peer that it is either 585 too busy to handle a request (DIAMETER_TOO_BUSY) or unable to route a 586 request to an upstream destination (DIAMETER_UNABLE_TO_DELIVER), 587 perhaps because that destination itself is overloaded to the point of 588 unavailability. 590 Once a Diameter node learns that an upstream peer has become 591 overloaded via one of these mechanisms, it can then attempt to take 592 action to reduce the load. This usually means forwarding traffic to 593 an alternate destination, if available. If no alternate destination 594 is available, the node must either reduce the number of messages it 595 originates (in the case of a client) or inform the client to reduce 596 traffic (in the case of an agent.) 598 Diameter requires the use of a congestion-managed transport layer, 599 currently TCP or SCTP, to mitigate network congestion. It is 600 expected that these transports manage network congestion and that 601 issues with transport (e.g. congestion propagation and window 602 management) are managed at that level. But even with a congestion- 603 managed transport, a Diameter node can become overloaded at the 604 Diameter protocol or application layers due to the causes described 605 in Section 1.1 and congestion managed transports do not provide 606 facilities (and are at the wrong level) to handle server overload. 607 Transport level congestion management is also not sufficient to 608 address overload in cases of multi-hop and multi-destination 609 signaling. 611 4. Issues with the Current Mechanisms 613 The currently available Diameter mechanisms for indicating an 614 overload condition are not adequate to avoid service outages due to 615 overload. This may, in turn, contribute to broader congestion 616 collapse due to unresponsive Diameter nodes causing application or 617 transport layer retransmissions. In particular, they do not allow a 618 Diameter agent or server to shed load as it approaches overload. At 619 best, a node can only indicate that it needs to entirely stop 620 receiving requests, i.e. that it has effectively failed. Even that 621 is problematic due to the inability to indicate durational validity 622 on the transient errors available in the base Diameter protocol. 623 Diameter offers no mechanism to allow a node to indicate different 624 overload states for different categories of messages, for example, if 625 it is overloaded for one Diameter application but not another. 627 4.1. Problems with Implicit Mechanism 629 The implicit mechanism doesn't allow an agent or server to inform the 630 client of a problem until it is effectively too late to do anything 631 about it. The client does not know to take action until the upstream 632 node has effectively failed. A Diameter node has no opportunity to 633 shed load early to avoid collapse in the first place. 635 Additionally, the implicit mechanism cannot distinguish between 636 overload of a Diameter node and network congestion. Diameter treats 637 the failure to receive an answer as a transport failure. 639 4.2. Problems with Explicit Mechanisms 641 The Diameter specification is ambiguous on how a client should handle 642 receipt of a DIAMETER_TOO_BUSY response. The base specification 643 [I-D.ietf-dime-rfc3588bis] indicates that the sending client should 644 attempt to send the request to a different peer. It makes no 645 suggestion that a the receipt of a DIAMETER_TOO_BUSY response should 646 affect future Diameter messages in any way. 648 The Authentication, Authorization, and Accounting (AAA) Transport 649 Profile [RFC3539] recommends that a AAA node that receives a "Busy" 650 response failover all remaining requests to a different agent or 651 server. But while the Diameter base specification explicitly depends 652 on RFC3539 to define transport behavior, it does not refer to RFC3539 653 in the description of behavior on receipt of DIAMETER_TOO_BUSY. 654 There's a strong likelihood that at least some implementations will 655 continue to send Diameter requests to an upstream peer even after 656 receiving a DIAMETER_TOO_BUSY error. 658 BCP 41 [RFC2914] describes, among other things, how end-to-end 659 application behavior can help avoid congestion collapse. In 660 particular, an application should avoid sending messages that will 661 never be delivered or processed. The DIAMETER_TOO_BUSY behavior as 662 described in the Diameter base specification fails at this, since if 663 an upstream node becomes overloaded, a client attempts each request, 664 and does not discover the need to failover the request until the 665 initial attempt fails. 667 The situation is improved if implementations follow the [RFC3539] 668 recommendation and keep state about upstream peer overload. But even 669 then, the Diameter specification offers no guidance on how long a 670 client should wait before retrying the overloaded destination. If an 671 agent or server supports multiple realms and/or applications, 672 DIAMETER_TOO_BUSY offers no way to indicate that it is overloaded for 673 one application but not another. A DIAMETER_TOO_BUSY error can only 674 indicate overload at a "whole server" scope. 676 Agent processing of a DIAMETER_TOO_BUSY response is also problematic 677 as described in the base specification. DIAMETER_TOO_BUSY is defined 678 as a protocol error. If an agent receives a protocol error, it may 679 either handle it locally or it may forward the response back towards 680 the downstream peer. (The Diameter specification is inconsistent 681 about whether a protocol error MAY or SHOULD be handled by an agent, 682 rather than forwarded downstream.) If a downstream peer receives the 683 DIAMETER_TOO_BUSY response, it may stop sending all requests to the 684 agent for some period of time, even though the agent may still be 685 able to deliver requests to other upstream peers. 687 DIAMETER_UNABLE_TO_DELIVER also has no mechanisms for specifying the 688 scope or cause of the failure, or the durational validity. 690 5. Diameter Overload Case Studies 692 5.1. Overload in Mobile Data Networks 694 As the number of Third Generation (3G) and Long Term Evolution (LTE) 695 enabled smartphone devices continue to expand in mobility networks, 696 there have been situations where high signaling traffic load led to 697 overload events at the Diameter-based Home Location Registries (HLR) 698 and/or Home Subscriber Servers (HSS). The root causes of the HLR 699 congestion events were manifold but included hardware failure and 700 procedural errors. The result was high signaling traffic load on the 701 HLR and HSS. 703 The 3GPP standards specification[need citation] for the end-to-end 704 signaling call flows in 3G and LTE, from the end user device 705 traversing through the radio and the core networks to the HLR/HSS, 706 did not have an equivalent load control mechanism which is provided 707 in the more traditional SS7 elements in GSM [need citation]. The 708 capabilities specified in the 3GPP standards do not adequately 709 address the abnormal condition where excessively high signaling 710 traffic load situations are experienced. 712 Smartphones contribute much more heavily to the continuation of a 713 registration surge due to their very aggressive registration 714 algorithms. The aggressive smartphone logic is designed to: 716 a. always have voice and data registration, and 718 b. constantly try to be on 3G or LTE data (and thus on 3G voice or 719 VoLTE) for their added benefits. 721 Non-smartphones typically have logic to wait for a time period after 722 registering successfully on voice and data. 724 The smartphone aggressive registration is problematic in two ways: 726 o first by generating excessive signaling load towards the HLR that 727 is ten times that from a non-smartphone, 729 o and second by causing continual registration attempts when a 730 network failure affects registrations through the 3G data network. 732 5.2. 3GPP Study on Core Network Overload 734 A study in 3GPP SA2 on core network overload has produced the 735 technical report [TR23.843]. This enumerates several causes of 736 overload in mobile core networks including portions that are signaled 737 using Diameter. This document is a work in progress and is not 738 complete. However, it is useful for pointing out scenarios and the 739 general need for an overload control mechanism for Diameter. 741 It is common for mobile networks to employ more than one radio 742 technology and to do so in an overlay fashion with multiple 743 technologies present in the same location (such as GSM, UMTS or CDMA 744 along with LTE). This presents opportunities for traffic storms when 745 issues occur on one overlay and not another as all devices that had 746 been on the overlay with issues switch. This causes a large amount 747 of Diameter traffic as locations and policies are updated. 749 Another scenario called out by this study is a flood of registration 750 and mobility management events caused by some element in the core 751 network failing. This flood of traffic from end nodes falls under 752 the network initiated traffic flood category. There is likely to 753 also be traffic resulting directly from the component failure in this 754 case. 756 Subscriber initiated traffic floods are also indicated in this study 757 as an overload mechanism where a large number of mobile devices 758 attempting to access services at the same time, such as in response 759 to an entertainment event or a catastrophic event. 761 While this study is concerned with the broader effects of these 762 scenarios on wireless networks and their elements, they have 763 implications specifically for Diameter signaling. One of the goals 764 of this document is to provide guidance for a core mechanism that can 765 be used to mitigate the scenarios called out by this study. 767 6. Solution Requirements 769 This section proposes requirements for an improved mechanism to 770 control Diameter overload, with the goals of improving the issues 771 described in Section 4 and supporting the scenarios described in 772 Section 2 774 REQ 1: The overload mechanism MUST provide a communication method 775 for Diameter nodes to exchange overload information. 777 REQ 2: The overload mechanism MUST be useable with any existing or 778 future Diameter application. It MUST NOT require 779 specification changes for existing Diameter applications. 781 REQ 3: The overload mechanism MUST limit the impact of overload on 782 the overall useful throughput of a Diameter server, even 783 when the incoming load on the network is far in excess of 784 its capacity. The overall useful throughput under load is 785 the ultimate measure of the value of an overload control 786 mechanism. 788 REQ 4: Diameter allows requests to be sent from either side of a 789 connection and either side of a connection may have need to 790 provide its overload status. The mechanism MUST allow each 791 side of a connection to independently inform the other of 792 its overload status. 794 REQ 5: Diameter allows nodes to determine their peers via dynamic 795 discovery or manual configuration. The mechanism MUST work 796 consistently without regard to how peers are determined. 798 REQ 6: The mechanism designers SHOULD seek to minimize the amount 799 of new configuration required in order to work. For 800 example, it is better to allow peers to advertise or 801 negotiate support for the mechanism, rather than to require 802 this knowledge to be configured at each node. 804 REQ 7: The overload mechanism MUST ensure that the system remains 805 stable. When the offered load drops from above the overall 806 capacity of the network to below the overall capacity, the 807 throughput MUST stabilize and become equal to the offered 808 load. 810 REQ 8: The mechanism MUST allow nodes to shed load without 811 introducing oscillations. Note that this requirement 812 implies a need for supporting nodes to be able to 813 distinguish current overload information from stale 814 information, and to make decisions using the most currently 815 available information. 817 REQ 9: The mechanism MUST function across fully loaded as well as 818 quiescent transport connections. This is partially derived 819 from the requirements for stability and hysteresis control 820 above. 822 REQ 10: Consumers of overload state indications MUST be able to 823 determine when the overload condition improves or ends. 825 REQ 11: The overload mechanism MUST be scalable. That is, it MUST 826 be able to operate in different sized networks. 828 REQ 12: When a single network node fails, goes into overload, or 829 suffers from reduced processing capacity, the mechanism MUST 830 make it possible to limit the impact of this on other nodes 831 in the network. This helps to prevent a small-scale failure 832 from becoming a widespread outage. 834 REQ 13: The mechanism MUST NOT introduce substantial additional work 835 for node in an overloaded state. For example, a requirement 836 for an overloaded node to send overload information every 837 time it received a new request would introduce substantial 838 work. Existing messaging is likely to have the 839 characteristic of increasing as an overload condition 840 approaches, allowing for the possibility of increased 841 feedback for information piggybacked on it. 843 REQ 14: Some scenarios that result in overload involve a rapid 844 increase of traffic with little time between normal levels 845 and overload inducing levels. The mechanism SHOULD provide 846 for increased feedback when traffic levels increase. The 847 mechanism MUST NOT do this in such a way that it increases 848 the number of messages while at high loads. 850 REQ 15: The mechanism MUST NOT interfere with the congestion control 851 mechanisms of underlying transport protocols. For example, 852 a mechanism that opened additional TCP connections when the 853 network is congested would reduce the effectiveness of the 854 underlying congestion control mechanisms. 856 REQ 16: The mechanism MUST operate without malfunction in an 857 environment with a mix of nodes that do, and nodes that do 858 not, support the mechanism. 860 REQ 17: In a mixed environment with nodes that support the overload 861 control mechanism and that do not, the mechanism MUST NOT 862 result in less useful throughput than would have resulted if 863 it were not present. It SHOULD result in less severe 864 congestion in this environment. 866 REQ 18: In a mixed environment of nodes that support the overload 867 control mechanism and that do not, users and operators of 868 nodes that do not support the mechanism MUST NOT benefit 869 from the mechanism more than users and operators of nodes 870 that support the mechanism. 872 REQ 19: It MUST be possible to use the mechanism between nodes in 873 different realms and in different administrative domains. 875 REQ 20: Any explicit overload indication MUST distinguish between 876 actual overload, as opposed to other, non-overload related 877 failures. 879 REQ 21: In cases where a network node fails, is so overloaded that 880 it cannot process messages, or cannot communicate due to a 881 network failure, it may not be able to provide explicit 882 indications of the nature of the failure or its levels of 883 congestion. The mechanism MUST properly function in these 884 cases. 886 REQ 22: The mechanism MUST provide a way for an node to throttle the 887 amount of traffic it receives from an peer node. This 888 throttling SHOULD be graded so that it can be applied 889 gradually as offered load increases. Overload is not a 890 binary state; there may be degrees of overload. 892 REQ 23: The mechanism MUST enable a supporting node to minimize the 893 chance that retries due to an overloaded or failed node 894 result in additional traffic to other overloaded nodes, or 895 cause additional nodes to become overloaded. Moreover, the 896 mechanism SHOULD provide unambiguous directions to clients 897 on when they should retry a request and when they should not 898 considering the various causes of overload such as avalanche 899 restart. 901 REQ 24: The mechanism MUST provide sufficient information to enable 902 a load balancing node to divert messages that are rejected 903 or otherwise throttled by an overloaded upstream node to 904 other upstream nodes that are the most likely to have 905 sufficient capacity to process them. 907 REQ 25: The mechanism MUST provide a mechanism for indicating load 908 levels even when not in an overloaded condition, to assist 909 nodes making decisions to prevent overload conditions from 910 occurring. 912 REQ 26: The specification for the overload mechanism SHOULD offer 913 guidance on which message types might be desirable to send 914 or process over others during times of overload, based on 915 Diameter-specific considerations. For example, it may be 916 more beneficial to process messages for existing sessions 917 ahead of new sessions. 919 REQ 27: The mechanism MUST NOT prevent a node from prioritizing 920 requests based on any local policy, so that certain requests 921 are given preferential treatment, given additional 922 retransmission, or processed ahead of others. 924 REQ 28: The overload mechanism MUST NOT provide new vulnerabilities 925 to malicious attack, or increase the severity of any 926 existing vulnerabilities. This includes vulnerabilities to 927 DoS and DDoS attacks as well as replay and man-in-the middle 928 attacks. 930 REQ 29: The mechanism MUST provide a means to match an overload 931 indication with the node that originated it. In particular, 932 the mechanism MUST allow a node to distinguish between 933 overload at a next-hop peer from overload at a node upstream 934 of the peer. For example, in Figure 5, the client must not 935 mistake overload at server 1 for overload at the agent, 936 whether or not the agent supports the mechanism.( see REQ 937 4). 939 REQ 30: The mechanism MUST NOT depend on being deployed in 940 environments where all Diameter nodes are completely 941 trusted. It SHOULD operate as effectively as possible in 942 environments where other nodes are malicious; this includes 943 preventing malicious nodes from obtaining more than a fair 944 share of service. Note that this does not imply any 945 responsibility on the mechanism to detect, or take 946 countermeasures against, malicious nodes. 948 REQ 31: It MUST be possible for a supporting node to make 949 authorization decisions about what information will be sent 950 to peer nodes based on the identity of those nodes. This 951 allows a domain administrator who considers the load of 952 their nodes to be sensitive information to restrict access 953 to that information. Of course, in such cases, there is no 954 expectation that the overload mechanism itself will help 955 prevent overload from that peer node. 957 REQ 32: The mechanism MUST NOT interfere with any Diameter compliant 958 method that a node may use to protect itself from overload 959 from non-supporting nodes, or from denial of service 960 attacks. 962 REQ 33: There are multiple situations where a Diameter node may be 963 overloaded for some purposes but not others. For example, 964 this can happen to an agent or server that supports multiple 965 applications, or when a server depends on multiple external 966 resources, some of which may become overloaded while others 967 are fully available. The mechanism MUST allow Diameter 968 nodes to indicate overload with sufficient granularity to 969 allow clients to take action based on the overloaded 970 resources without forcing available capacity to go unused. 971 The mechanism MUST support specification of overload 972 information with granularities of at least "Diameter node", 973 "realm", "Diameter application", and "Diameter session", and 974 SHOULD allow extensibility for others to be added in the 975 future. 977 REQ 34: The mechanism MUST provide a method for extending the 978 information communicated and the algorithms used for 979 overload control. 981 REQ 35: The mechanism SHOULD provide a method for exchanging 982 overload and load information between elements that are 983 connected by intermediaries that do not support the 984 mechanism. A separate mechanism or extension of the 985 mechanism to support this may be warranted for this. 987 7. IANA Considerations 989 This document makes no requests of IANA. 991 8. Security Considerations 993 A Diameter overload control mechanism is primarily concerned with the 994 load and overload related behavior of nodes in a Diameter network, 995 and the information used to affect that behavior. Load and overload 996 information is shared between nodes and directly affects the behavior 997 and thus is potentially vulnerable to a number of methods of attack. 999 Load and overload information may also be sensitive from both 1000 business and network protection viewpoints. Operators of Diameter 1001 equipment want to control visibility to load and overload information 1002 to keep it from being used for competitive intelligence or for 1003 targeting attacks. It is also important that the Diameter overload 1004 control mechanism not introduce any way in which any other 1005 information carried by Diameter is sent inappropriately. 1007 This document includes requirements intended to mitigate the effects 1008 of attacks and to protect the information used by the mechanism. 1010 8.1. Access Control 1012 To control the visibility of load and overload information, sending 1013 should be subject to some form of authentication and authorization of 1014 the receiver. It is also important to the receivers that they are 1015 confident the load and overload information they receive is from a 1016 legitimate source. Note that this implies a certain amount of 1017 configurability on the nodes supporting the Diameter overload control 1018 mechanism. 1020 8.2. Denial-of-Service Attacks 1022 An overload control mechanism provides a very attractive target for 1023 denial-of-service attacks. A small number of messages may affect a 1024 large service disruption by falsely reporting overload conditions. 1025 Alternately, attacking servers nearing, or in, overload may also be 1026 facilitated by disrupting their overload indications, potentially 1027 preventing them from mitigating their overload condition. 1029 A design goal for the Diameter overload control mechanism is to 1030 minimize or eliminate the possibility of using the mechanism for this 1031 type of attack. 1033 As the intent of some denial-of-service attacks is to induce overload 1034 conditions, an effective overload control mechanism should help to 1035 mitigate the effects of an such an attack. 1037 8.3. Replay Attacks 1039 An attacker that has managed to obtain some messages from the 1040 overload control mechanism may attempt to affect the behavior of 1041 nodes supporting the mechanism by sending those messages at 1042 potentially inopportune times. In addition to time shifting, replay 1043 attacks may send messages to other nodes as well (target shifting). 1045 A design goal for the Diameter overload control mechanism is to 1046 minimize or eliminate the possibility of causing disruption by using 1047 a replay attack on the Diameter overload control mechanism. 1049 8.4. Man-in-the-Middle Attacks 1051 By inserting themselves in between two nodes supporting the Diameter 1052 overload control mechanism, an attacker may potentially both access 1053 and alter the information sent between those nodes. This can be used 1054 for information gathering for business intelligence and attack 1055 targeting, as well as direct attacks. 1057 A design goal for the Diameter overload control mechanism is to 1058 minimize or eliminate the possibility of causing disruption man-in- 1059 the-middle attacks on the Diameter overload control mechanism. A 1060 transport using TLS and/or IPSEC may be desirable for this. 1062 8.5. Compromised Hosts 1064 A compromised host that supports the Diameter overload control 1065 mechanism could be used for information gathering as well as for 1066 sending malicious information to any Diameter node that would 1067 normally accept information from it. While is is beyond the scope of 1068 the Diameter overload control mechanism to mitigate any operational 1069 interruption to the compromised host, a reasonable design goal is to 1070 minimize the impact that a compromised host can have on other nodes 1071 through the use of the Diameter overload control mechanism. Of 1072 course, a compromised host could be used to cause damage in a number 1073 of other ways. This is out of scope for a Diameter overload control 1074 mechanism. 1076 9. References 1078 9.1. Normative References 1080 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1081 Requirement Levels", BCP 14, RFC 2119, March 1997. 1083 [I-D.ietf-dime-rfc3588bis] 1084 Fajardo, V., Arkko, J., Loughney, J., and G. Zorn, 1085 "Diameter Base Protocol", draft-ietf-dime-rfc3588bis-34 1086 (work in progress), June 2012. 1088 [RFC2914] Floyd, S., "Congestion Control Principles", BCP 41, 1089 RFC 2914, September 2000. 1091 [RFC3539] Aboba, B. and J. Wood, "Authentication, Authorization and 1092 Accounting (AAA) Transport Profile", RFC 3539, June 2003. 1094 9.2. Informative References 1096 [RFC5390] Rosenberg, J., "Requirements for Management of Overload in 1097 the Session Initiation Protocol", RFC 5390, December 2008. 1099 [TR23.843] 1100 3GPP, "Study on Core Network Overload Solutions", 1101 TR 23.843 0.4.0, April 2011. 1103 Appendix A. Contributors 1105 Significant contributions to this document were made by Adam Roach 1106 and Eric Noel. 1108 Appendix B. Acknowledgements 1110 Review of, and contributions to, this specification by Martin Dolly, 1111 Carolyn Johnson, Jianrong Wang, Imtiaz Shaikh, Jouni Korhonen, and 1112 Robert Sparks were most appreciated. We would like to thank them for 1113 their time and expertise. 1115 Authors' Addresses 1117 Eric McMurry 1118 Tekelec 1119 17210 Campbell Rd. 1120 Suite 250 1121 Dallas, TX 75252 1122 US 1124 Email: eric.mcmurry@tekelec.com 1126 Ben Campbell 1127 Tekelec 1128 17210 Campbell Rd. 1129 Suite 250 1130 Dallas, TX 75252 1131 US 1133 Email: ben@nostrum.com