idnits 2.17.1 draft-ietf-tsvwg-sctp-failover-07.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 : ---------------------------------------------------------------------------- ** The abstract seems to contain references ([RFC4960]), which it shouldn't. Please replace those with straight textual mentions of the documents in question. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (October 23, 2014) is 3472 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) ** Obsolete normative reference: RFC 4960 (Obsoleted by RFC 9260) Summary: 2 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 Y. Nishida 3 Internet-Draft GE Global Research 4 Intended status: Standards Track P. Natarajan 5 Expires: April 26, 2015 Cisco Systems 6 A. Caro 7 BBN Technologies 8 P. Amer 9 University of Delaware 10 K. Nielsen 11 Ericsson 12 October 23, 2014 14 Quick Failover Algorithm in SCTP 15 draft-ietf-tsvwg-sctp-failover-07.txt 17 Abstract 19 One of the major advantages of SCTP is that it supports multi-homed 20 communication. A multi-homed SCTP end-point has the ability to 21 withstand network failures by migrating the traffic from an inactive 22 network to an active one. However, if the [RFC4960] specified 23 failover operation is followed there can be a significant delay in 24 the migration to the active destination addresses, thus severely 25 reducing the effectiveness of SCTP multi-homed operation. 27 The memo complements RFC4960 by the introduction of the Potentially 28 Failed state and associated new Quick Failover operation to apply 29 during network failure and specifies for SCTP senders to support this 30 more performance optimal failover procedure as an add-on to the 31 [RFC4960] failover operation. The memo in addition complements 32 [RFC4960] by introduction of alternative switchover operation modes 33 for the data transfer path management after a failover. These 34 operation modes offer for more performance optimal operation in some 35 network environments. From the perspective of this memo the 36 implementation of the additional switchover operation modes is 37 considered optional. 39 The procedures defined require only minimal modifications to the 40 current specification. The procedures are sender-side only and do 41 not impact the SCTP receiver. 43 Status of This Memo 45 This Internet-Draft is submitted in full conformance with the 46 provisions of BCP 78 and BCP 79. 48 Internet-Drafts are working documents of the Internet Engineering 49 Task Force (IETF). Note that other groups may also distribute 50 working documents as Internet-Drafts. The list of current Internet- 51 Drafts is at http://datatracker.ietf.org/drafts/current/. 53 Internet-Drafts are draft documents valid for a maximum of six months 54 and may be updated, replaced, or obsoleted by other documents at any 55 time. It is inappropriate to use Internet-Drafts as reference 56 material or to cite them other than as "work in progress." 58 This Internet-Draft will expire on April 26, 2015. 60 Copyright Notice 62 Copyright (c) 2014 IETF Trust and the persons identified as the 63 document authors. All rights reserved. 65 This document is subject to BCP 78 and the IETF Trust's Legal 66 Provisions Relating to IETF Documents 67 (http://trustee.ietf.org/license-info) in effect on the date of 68 publication of this document. Please review these documents 69 carefully, as they describe your rights and restrictions with respect 70 to this document. Code Components extracted from this document must 71 include Simplified BSD License text as described in Section 4.e of 72 the Trust Legal Provisions and are provided without warranty as 73 described in the Simplified BSD License. 75 Table of Contents 77 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 78 2. Conventions and Terminology . . . . . . . . . . . . . . . . . 4 79 3. Issues with the SCTP Path Management . . . . . . . . . . . . 4 80 4. SCTP with Potentially-Failed Destination State (SCTP-PF) . . 5 81 4.1. SCTP-PF Concept . . . . . . . . . . . . . . . . . . . . . 5 82 4.2. SCTP-PF Algorithm Detail . . . . . . . . . . . . . . . . 6 83 4.3. Optional Feature: Permanent Failover . . . . . . . . . . 9 84 5. Socket API Considerations . . . . . . . . . . . . . . . . . . 10 85 5.1. Support for the Potentially Failed Path State . . . . . . 11 86 5.2. Peer Address Thresholds (SCTP_PEER_ADDR_THLDS) Socket 87 Option . . . . . . . . . . . . . . . . . . . . . . . . . 12 88 5.3. Exposing the Potentially Failed Path State 89 (SCTP_EXPOSE_POTENTIALLY_FAILED_STATE) Socket Option . . 13 90 6. Security Considerations . . . . . . . . . . . . . . . . . . . 13 91 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 14 92 8. Proposed Change of Status (to be Deleted before Publication) 14 93 9. References . . . . . . . . . . . . . . . . . . . . . . . . . 14 94 9.1. Normative References . . . . . . . . . . . . . . . . . . 14 95 9.2. Informative References . . . . . . . . . . . . . . . . . 14 97 Appendix A. Discussions of Alternative Approaches . . . . . . . 15 98 A.1. Reduce Path.Max.Retrans (PMR) . . . . . . . . . . . . . . 15 99 A.2. Adjust RTO related parameters . . . . . . . . . . . . . . 16 100 Appendix B. Discussions for Path Bouncing Effect . . . . . . . . 16 101 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 17 103 1. Introduction 105 The Stream Control Transmission Protocol (SCTP) as specified in 106 [RFC4960] supports multihoming at the transport layer -- an SCTP 107 association can bind to multiple IP addresses at each endpoint. 108 SCTP's multihoming features include failure detection and failover 109 procedures to provide network interface redundancy and improved end- 110 to-end fault tolerance. 112 In SCTP's current failure detection procedure, the sender must 113 experience Path.Max.Retrans (PMR) number of consecutive failed 114 retransmissions on a destination before detecting a path failure. 115 The sender fails over to an alternate active destination only after 116 failure detection. Until detecting the failover, the sender 117 continues to transmit data on the failed path, which degrades the 118 SCTP performance. Concurrent Multipath Transfer (CMT) [IYENGAR06] is 119 an extension to SCTP and allows the sender to transmit data on 120 multiple paths simultaneously. Research [NATARAJAN09] shows that the 121 current failure detection procedure worsens CMT performance during 122 failover and can be significantly improved by employing a better 123 failover algorithm. 125 This document specifies an alternative failure detection procedure 126 for SCTP that improves the SCTP performance during a failover. 128 Also the operation after a failover impacts the performance of the 129 protocol. With [RFC4960] procedures, SCTP will, after a failover 130 from the primary path, switch back to use the primary path for data 131 transfer as soon as this path becomes available. From a performance 132 perspective, as confirmed in research [CARO02], such a switchback of 133 the data transmission path is not optimal in general. As an optional 134 alternative to the switchback operation of [RFC4960], this document 135 specifies for SCTP to support the Permanent Failover switchover 136 procedures proposed by [CARO02]. Additional discussions for 137 alternative approach that does not require modifications to [RFC4960] 138 and path bouncing effects that might be caused by frequent switchover 139 are provided in Appendix. 141 2. Conventions and Terminology 143 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 144 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 145 document are to be interpreted as described in [RFC2119]. 147 3. Issues with the SCTP Path Management 149 This section describes issues in the current SCTP to be fixed by the 150 approach described in this document. 152 SCTP can utilize multiple IP addresses for a single SCTP association. 153 Each SCTP endpoint exchanges the list of its usable addresses during 154 initial negotiation with its peer. Then the endpoints select one 155 address from the peer's list and define this as the primary 156 destination. During normal transmission, SCTP sends all user data to 157 the primary destination. Also, it sends heartbeat packets to all 158 idle destinations at a certain interval to check the reachability of 159 the path. Idle destinations normally include all non-primary 160 destinations. 162 If a sender has multiple active destination addresses, it can 163 retransmit data to secondary destination address, when the 164 transmission to the primary times out. 166 When a sender receives an acknowledgment for DATA or HEARTBEAT chunks 167 sent to one of the destination addresses, it considers that 168 destination to be active. If it fails to receive acknowledgments, 169 the error count for the address is increased. If the error counter 170 exceeds the protocol parameter 'Path.Max.Retrans', SCTP endpoint 171 considers the address to be inactive. 173 The failover process of SCTP is initiated when the primary path 174 becomes inactive (error counter for the primary path exceeds 175 Path.Max.Retrans). If the primary path is marked inactive, SCTP 176 chooses a new destination address from one of the active destinations 177 and start using this address to send data to. If the primary path 178 becomes active again, SCTP uses the primary destination for 179 subsequent data transmissions and stop using non-primary one. 181 One issue with this failover process is that it usually takes 182 significant amount of time before SCTP switches to the new 183 destination. Let's say the primary path on a multi-homed host 184 becomes unavailable and the RTO value for the primary path at that 185 time is around 1 second, it usually takes over 60 seconds before SCTP 186 starts to use the secondary path. This is because the recommended 187 value for Path.Max.Retrans in the standard is 5, which requires 6 188 consecutive timeouts before failover takes place. Before SCTP 189 switches to the secondary address, SCTP keeps trying to send packets 190 to the primary and only retransmitted packets are sent to the 191 secondary and can thus be reached at the receiver. This slow 192 failover process can cause significant performance degradation and 193 will not be acceptable in some situations. 195 Another issue is that once the primary path is active again, the 196 traffic is switched back. This is not optimal in some situations. 197 This is further discussed in Section 4.3. 199 4. SCTP with Potentially-Failed Destination State (SCTP-PF) 201 To address the issues described in Section 3, this section updates 202 SCTP path management scheme with the Potentially Failed state and 203 associated Quick Failover operation. We use the term SCTP-PF to 204 denote the resulting SCTP path management operation. 206 4.1. SCTP-PF Concept 208 SCTP-PF as defined stems from the following two observations about 209 SCTP's failure detection procedure: 211 o To minimize performance impact during failover, the sender should 212 avoid transmitting data to the failed destination as early as 213 possible. In the current SCTP path management scheme, the sender 214 stops transmitting data to a destination only after the 215 destination is marked Failed (inactive). Thus, a smaller PMR 216 value is ideal so that the sender transitions a destination to the 217 Failed (inactive) state quicker. 219 o Smaller PMR values increase the chances of spurious failure 220 detection where the sender incorrectly marks a destination as 221 Failed (inactive) during periods of temporary congestion. As 222 [RFC4960] recommends for a coupling of the PMR value and the AMR 223 value such spurious failure detection risks to carry over to 224 spurious association failure detection and closure. Larger PMR 225 values are preferable to avoid spurious failure detection. 227 From the above observations it is clear that tuning the PMR value 228 involves the following tradeoff -- a lower value improves performance 229 but increases the chances of spurious failure detection, whereas a 230 higher value degrades performance and reduces spurious failure 231 detection in a wide range of path conditions. Thus, tuning the 232 association's PMR value is an incomplete solution to address 233 performance impact during failure. 235 This new method introduces a new "Potentially-Failed" (PF) 236 destination state in SCTP's path management procedure. The PF state 237 was originally proposed to improve CMT performance [NATARAJAN09]. 238 The PF state is an intermediate state between Active and Failed 239 states. SCTP's failure detection procedure is modified to include 240 the PF state. The new failure detection algorithm assumes that loss 241 detected by a timeout implies either severe congestion or failure en- 242 route. After a number of consecutive timeouts on a path, the sender 243 is unsure, and marks the corresponding destination as PF. A PF 244 destination is not used for data transmission except in special cases 245 (discussed below). The new failure detection algorithm requires only 246 sender-side changes. 248 4.2. SCTP-PF Algorithm Detail 250 SCTP PF operation is specified as follows: 252 1. The sender maintains a new tunable parameter called Potentially- 253 Failed.Max.Retrans (PFMR). The RECOMMENDED value of PFMR = 0 254 when Quick Failover is used. When PFMR is larger or equal to 255 PMR, Quick Failover is turned off. 257 2. The error counter of an active destination address is 258 incremented as specified in [RFC4960]. This means that the 259 error counter of the destination address will be incremented 260 each time the T3-rtx timer expires, or at times where a 261 HEARTBEAT sent to an idle, active address is not acknowledged 262 within an RTO. When the value in the destination address error 263 counter exceeds PFMR, the endpoint MUST mark the destination 264 transport address as PF. 266 3. The sender SHOULD avoid data transmission to PF destinations. 267 When the destinations are all in PF state or some in PF state 268 and some in inactive state, the sender MUST choose one 269 destination in PF state and transmit data to this destination. 270 The sender SHOULD choose the destination in PF state with the 271 lowest error count (fewest consecutive timeouts) for data 272 transmission and transmit data to this destination. When there 273 are multiple PF destinations with same error count, the sender 274 SHOULD let the choice among the multiple PF destination with 275 equal error count be based on the [RFC4960], section 6.4.1, 276 principles of choosing most divergent source-destination pairs 277 when executing (potentially consecutive) retransmission. This 278 means that the sender SHOULD attempt to pick the most divergent 279 source - destination pair from the last source - destination 280 pair on which data were transmitted or retransmitted. Rules for 281 picking the most divergent source-destination pair are an 282 implementation decision and are not specified within this 283 document. A sender may choose to deploy other strategies than 284 the above when choosing among multiple PF destinations with 285 equal error count. In all cases the sender MUST NOT change the 286 state of chosen destination and it MUST NOT clear the 287 destination's error counter as a result of choosing the 288 destination for data transmission. 290 4. Heartbeats SHOULD be sent to PF destination(s) once per RTO. 291 This means the sender MUST ignore HB.interval for PF 292 destinations. If an heartbeat is unanswered, the sender SHOULD 293 increment the error counter and exponentially back off the RTO 294 value. If error counter is less than PMR, the sender SHOULD 295 transmit another heartbeat immediately after T3-timer 296 expiration. When data is transmitted to a PF destination, the 297 transmission of heartbeats may be omitted as SACK or T3-rtx 298 timer expiration can provide equivalent information. It is 299 RECOMMENDED that heartbeats be send to PF destinations 300 regardless of whether the Path Heartbeat function (Section 8.3 301 of [RFC4960]) is enabled for the destination address or not. 303 5. When the sender receives an heartbeat ACK from a PF destination, 304 the sender MUST clear the destination's error counter and 305 transition the PF destination back to Active state. When the 306 sender resumes data transmission on the destination it MUST do 307 this following the prescriptions of Section 7.2 of [RFC4960]. 309 6. Additional (PMR - PFMR) consecutive timeouts on a PF destination 310 confirm the path failure, upon which the destination transitions 311 to the Inactive state. As described in [RFC4960], the sender 312 (i) SHOULD notify ULP about this state transition, and (ii) 313 transmit heartbeats to the Inactive destination at a lower 314 frequency as described in Section 8.3 of [RFC4960] (when this 315 function is enabled for the destination address). 317 7. When all destinations are in inactive state (association dormant 318 state) the sender MUST also choose one destination to transmit 319 data to. The sender SHOULD choose the destination in inactive 320 state with the lowest error count (fewest consecutive timeouts) 321 for data transmission and transmit data to this destination. 322 When there are multiple destinations with same error count in 323 inactive state, the sender SHOULD attempt to pick the most 324 divergent source - destination pair from the last source - 325 destination pair on which data were transmitted or retransmitted 326 following [RFC4960]. Rules for picking the most divergent 327 source-destination pair are an implementation decision and are 328 not specified within this document. Therefore, a sender SHOULD 329 allow for incrementing the destination error counters up to some 330 reasonable limit larger than PMR+1, thus changing the 331 prescriptions of [RFC4960], section 8.3, in this respect. The 332 exact limit to apply is not specified in this document but it is 333 considered reasonable to require for such to be an order of 334 magnitude higher than the PMR value. A sender MAY choose to 335 deploy other strategies than the above. For example, a sender 336 could choose to prioritize the last active destination during 337 dormant state. The strategy to prioritize the last active 338 destination is optimal when some paths are permanently inactive, 339 but suboptimal when paths' instability is transient. While the 340 increment of the error counters above PMR+1 is a prerequisite 341 for the error counter values to serve to guide the path 342 selection in dormant state, then it is noted that by virtue of 343 the introduction of the Potentially Failed state, one may deploy 344 higher values of PMR without compromising the efficiency of the 345 failover operation, and thus making the increase of path error 346 counters above PMR+1 less critical as the dormant state will be 347 less likely to happen. The downside of increasing the PMR value 348 relative to the AMR value, however, is that the per destination 349 address failure detection and notification of such to ULP 350 thereby is weakened. In all cases the sender MUST NOT change 351 the state of the chosen destination and it MUST NOT clear the 352 destination's error counter as a result of choosing the 353 destination for data transmission. 355 8. ACKs for chunks that have been transmitted to multiple 356 destinations (i.e., a chunk which has been retransmitted to a 357 different destination than the destination to which the chunk 358 was first transmitted) SHOULD NOT clear the error count of an 359 inactive destination and SHOULD NOT transition a PF destination 360 back to Active state, since a sender cannot disambiguate whether 361 the ACK was for the original transmission or the 362 retransmission(s). The same ambiguity concerns the related 363 congestion window growth. The bytes of a newly acknowledged 364 chunk which has been transmitted to multiple destinations SHOULD 365 be considered for contribution to the congestion window growth 366 towards the destination where the chunk was last sent. The 367 contribution of the acked bytes to the window growth is subject 368 to the prescriptions described in Section 7.2 of [RFC4960] is 369 fulfilled. A SCTP sender MAY apply a different approach for 370 both the error count handling and the congestion control growth 371 handling based on unequivocally information on which destination 372 (including multiple destinations) the chunk reached. This 373 document makes no reference to what such unequivocally 374 information could consist of, neither how such unequivocally 375 information could be obtained. The implementation of such an 376 alternative approach is left to implementations. 378 9. ACKs for chunks which has been transmitted to one destination 379 address only MUST clear the error counter of the destination 380 address and MUST transition a PF destination back to Active 381 state. This situation can happen when new data is sent to a 382 destination address in PF state. It can also happen in 383 situations where the destination address is in PF state due to 384 the occurrence of a spurious T3-rtx timer and ACKs start to 385 arrive for data sent prior to occurrence of the spurious T3-rtx 386 and data has not yet been retransmitted towards other 387 destinations. This document does not specify special handling 388 for detection of or reaction to spurious T3-rtx timeouts, e.g., 389 for special operation vis-a-vis the congestion control handling 390 or data retransmission operation towards a destination address 391 which undergoes a transition from active to PF to active state 392 due to a spurious T3-rtx timeout. But it is noted that this is 393 an area which would benefit from additional attention, 394 experimentation and specification for Single Homed SCTP as well 395 as for Multi Homed SCTP protocol operation. 397 10. SCTP stack SHOULD provide the ULP with the means to expose the 398 PF state of its destinations as well as the means to notify the 399 state transitions from Active to PF, and vice-versa. When doing 400 this, such SCTP stack MUST provide the ULP with the means to 401 suppress exposure of PF state and association state transitions 402 as well. 404 4.3. Optional Feature: Permanent Failover 406 In [RFC4960], an SCTP sender migrates the traffic back to the 407 original primary destination once this destination becomes active 408 again. As the CWND towards the original primary destination has to 409 be rebuilt once data transfer resumes, the switch back to use the 410 original primary path is not always optimal. Indeed [CARO02] shows 411 that the switch back to the original primary may degrade SCTP 412 performance compared to continuing data transmission on the same 413 path, especially, but not only, in scenarios where this path's 414 characteristics are better. In order to mitigate this performance 415 degradation, Permanent Failover operation was proposed in [CARO02]. 416 When SCTP changes the destination due to failover, Permanent Failover 417 operation allows SCTP sender to continue data transmission on the new 418 working path even if the old primary destination becomes active 419 again. This is achieved by having SCTP perform a switch over of the 420 primary path to the alternative working path rather than having SCTP 421 switch back data transfer to the (previous) primary path. 423 The manner of switch over operation that is most optimal in a given 424 scenario depends on the relative quality of a set primary path versus 425 the quality of alternative paths available as well as it depends on 426 the extent to which it is desired for the mode of operation to 427 enforce traffic distribution over a number of network paths. I.e., 428 load distribution of traffic from multiple SCTP associations may be 429 sought to be enforced by distribution of the set primary paths with 430 [RFC4960] switchback operation. However as [RFC4960] switchback 431 behavior is suboptimal in certain situations, especially in scenarios 432 where a number of equally good paths are available, it is recommended 433 for SCTP to support also, as alternative behavior, the Permanent 434 Failover switch over modes of operation. 436 The Permanent Failover operation requires only sender side changes. 437 The details are: 439 1. The sender maintains a new tunable parameter, called 440 Primary.Switchover.Max.Retrans (PSMR). The PSMR MUST be set 441 greater or equal to the PFMR value. Implementations MUST reject 442 any other values of PSMR. 444 2. When the path error counter on a set primary path exceeds PSMR, 445 the SCTP implementation MUST autonomously select and set a new 446 primary path. 448 3. The primary path selected by the SCTP implementation MUST be the 449 path which at the given time would be chosen for data transfer. 450 A previously failed primary path MAY come in use as data transfer 451 path as per normal path selection when the present data transfer 452 path fails. 454 4. The recommended value of PSMR is PFMR when Permanent Failover is 455 used. This means that no forced switchback to a previously 456 failed primary path is performed. An implementation of Permanent 457 Failover MUST support the setting of PSMR = PFMR. An 458 implementation of Permanent Failover MAY support setting of PSMR 459 > PFMR. 461 5. It MUST be possible to disable the Permanent Failover and obtain 462 the standard switchback operation of [RFC4960]. 464 This specifications RECOMMENDS a default configuration that uses 465 standard RFC4960 switchback, i.e., switch back to the old primary 466 destination once the destination becomes active again. However, to 467 support optimal operation in a wider range of network scenarios, an 468 implementation MAY implement Permanent Failover operation as detailed 469 above and MAY enable it based on network configurations or users' 470 requests. 472 5. Socket API Considerations 474 This section describes how the socket API defined in [RFC6458] is 475 extended to provide a way for the application to control and observe 476 the quick failover behavior. 478 Please note that this section is informational only. 480 A socket API implementation based on [RFC6458] is, by means of the 481 existing SCTP_PEER_ADDR_CHANGE event, extended to provide the event 482 notification when a peer address enters or leaves the potentially 483 failed state as well as the socket API implementation is extended to 484 expose the potentially failed state of a peer address in the existing 485 SCTP_GET_PEER_ADDR_INFO structure. 487 Furthermore, two new read/write socket options for the level 488 IPPROTO_SCTP and the name SCTP_PEER_ADDR_THLDS and 489 SCTP_EXPOSE_POTENTIALLY_FAILED_STATE are defined as described below. 490 The first socket option is used to control the values of the PFMR and 491 PSMR parameters described in Section 4. The second one controls the 492 exposition of the potentially failed path state. 494 Support for the SCTP_PEER_ADDR_THLDS and 495 SCTP_EXPOSE_POTENTIALLY_FAILED_STATE socket options need also to be 496 added to the function sctp_opt_info(). 498 5.1. Support for the Potentially Failed Path State 500 As defined in [RFC6458], the SCTP_PEER_ADDR_CHANGE event is provided 501 if the status of a peer address changes. In addition to the state 502 changes described in [RFC6458], this event is also provided, if a 503 peer address enters or leaves the potentially failed state. The 504 notification as defined in [RFC6458] uses the following structure: 506 struct sctp_paddr_change { 507 uint16_t spc_type; 508 uint16_t spc_flags; 509 uint32_t spc_length; 510 struct sockaddr_storage spc_aaddr; 511 uint32_t spc_state; 512 uint32_t spc_error; 513 sctp_assoc_t spc_assoc_id; 514 } 516 [RFC6458] defines the constants SCTP_ADDR_AVAILABLE, 517 SCTP_ADDR_UNREACHABLE, SCTP_ADDR_REMOVED, SCTP_ADDR_ADDED, and 518 SCTP_ADDR_MADE_PRIM to be provided in the spc_state field. This 519 document defines in addition to that the new constant 520 SCTP_ADDR_POTENTIALLY_FAILED, which is reported if the affected 521 address becomes potentially failed. 523 The SCTP_GET_PEER_ADDR_INFO socket option defined in [RFC6458] can be 524 used to query the state of a peer address. It uses the following 525 structure: 527 struct sctp_paddrinfo { 528 sctp_assoc_t spinfo_assoc_id; 529 struct sockaddr_storage spinfo_address; 530 int32_t spinfo_state; 531 uint32_t spinfo_cwnd; 532 uint32_t spinfo_srtt; 533 uint32_t spinfo_rto; 534 uint32_t spinfo_mtu; 535 }; 537 [RFC6458] defines the constants SCTP_UNCONFIRMED, SCTP_ACTIVE, and 538 SCTP_INACTIVE to be provided in the spinfo_state field. This 539 document defines in addition to that the new constant 540 SCTP_POTENTIALLY_FAILED, which is reported if the peer address is 541 potentially failed. 543 5.2. Peer Address Thresholds (SCTP_PEER_ADDR_THLDS) Socket Option 545 Applications can control the quick failover behavior by getting or 546 setting the number of consecutive timeouts before a peer address is 547 considered potentially failed or unreachable and before the primary 548 path is changed automatically. This socket option uses the level 549 IPPROTO_SCTP and the name SCTP_PEER_ADDR_THLDS. 551 The following structure is used to access and modify the thresholds: 553 struct sctp_paddrthlds { 554 sctp_assoc_t spt_assoc_id; 555 struct sockaddr_storage spt_address; 556 uint16_t spt_pathmaxrxt; 557 uint16_t spt_pathpfthld; 558 uint16_t spt_pathcpthld; 559 }; 561 spt_assoc_id: This parameter is ignored for one-to-one style 562 sockets. For one-to-many style sockets the application may fill 563 in an association identifier or SCTP_FUTURE_ASSOC. It is an error 564 to use SCTP_{CURRENT|ALL}_ASSOC in spt_assoc_id. 566 spt_address: This specifies which peer address is of interest. If a 567 wildcard address is provided, this socket option applies to all 568 current and future peer addresses. 570 spt_pathmaxrxt: Each peer address of interest is considered 571 unreachable, if its path error counter exceeds spt_pathmaxrxt. 573 spt_pathpfthld: Each peer address of interest is considered 574 potentially failed, if its path error counter exceeds 575 spt_pathpfthld. 577 spt_pathcpthld: Each peer address of interest is not considered the 578 primary remote address anymore, if its path error counter exceeds 579 spt_pathcpthld. Using a value of 0xffff disables the selection of 580 a new primary peer address. If an implementation does not support 581 the automatically selection of a new primary address, it should 582 indicate an error with errno set to EINVAL if a value different 583 from 0xffff is used in spt_pathcpthld. Setting of spt_pathcpthld 584 < spt_pathpfthld should be rejected with errno set to EINVAL. An 585 implementation MAY support only setting of spt_pathcpthld = 586 spt_pathpfthld and spt_pathcpthld = 0xffff. In this case it shall 587 reject setting of other values with errno set to EINVAL. 589 5.3. Exposing the Potentially Failed Path State 590 (SCTP_EXPOSE_POTENTIALLY_FAILED_STATE) Socket Option 592 Applications can control the exposure of the potentially failed path 593 state in the SCTP_PEER_ADDR_CHANGE event and the 594 SCTP_GET_PEER_ADDR_INFO as described in Section 5.1. The default 595 value is implementation specific. 597 This socket option uses the level IPPROTO_SCTP and the name 598 SCTP_EXPOSE_POTENTIALLY_FAILED_STATE. 600 The following structure is used to control the exposition of the 601 potentially failed path state: 603 struct sctp_assoc_value { 604 sctp_assoc_t assoc_id; 605 uint32_t assoc_value; 606 }; 608 assoc_id: This parameter is ignored for one-to-one style sockets. 609 For one-to-many style sockets the application may fill in an 610 association identifier or SCTP_FUTURE_ASSOC. It is an error to 611 use SCTP_{CURRENT|ALL}_ASSOC in assoc_id. 613 assoc_value: The potentially failed path state is exposed if and 614 only if this parameter is non-zero. 616 6. Security Considerations 618 Security considerations for the use of SCTP and its APIs are 619 discussed in [RFC4960] and [RFC6458]. There are no new security 620 considerations introduced in this document. 622 7. IANA Considerations 624 This document does not create any new registries or modify the rules 625 for any existing registries managed by IANA. 627 8. Proposed Change of Status (to be Deleted before Publication) 629 The initial status of this document was Experimental. However, 630 because of its usefulness, simple design and the existence of 631 multiple active implementations, it has been changed to PS by WG 632 consensus. 634 9. References 636 9.1. Normative References 638 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 639 Requirement Levels", BCP 14, RFC 2119, March 1997. 641 [RFC4960] Stewart, R., "Stream Control Transmission Protocol", RFC 642 4960, September 2007. 644 9.2. Informative References 646 [CARO02] Caro Jr., A., Iyengar, J., Amer, P., Heinz, G., and R. 647 Stewart, "A Two-level Threshold Recovery Mechanism for 648 SCTP", Tech report, CIS Dept, University of Delaware , 7 649 2002. 651 [CARO04] Caro Jr., A., Amer, P., and R. Stewart, "End-to-End 652 Failover Thresholds for Transport Layer Multihoming", 653 MILCOM 2004 , 11 2004. 655 [CARO05] Caro Jr., A., "End-to-End Fault Tolerance using Transport 656 Layer Multihoming", Ph.D Thesis, University of Delaware , 657 1 2005. 659 [FALLON08] 660 Fallon, S., Jacob, P., Qiao, Y., Murphy, L., Fallon, E., 661 and A. Hanley, "SCTP Switchover Performance Issues in WLAN 662 Environments", IEEE CCNC 2008, 1 2008. 664 [GRINNEMO04] 665 Grinnemo, K-J. and A. Brunstrom, "Performance of SCTP- 666 controlled failovers in M3UA-based SIGTRAN networks", 667 Advanced Simulation Technologies Conference , 4 2004. 669 [IYENGAR06] 670 Iyengar, J., Amer, P., and R. Stewart, "Concurrent 671 Multipath Transfer using SCTP Multihoming over Independent 672 End-to-end Paths.", IEEE/ACM Trans on Networking 14(5), 10 673 2006. 675 [JUNGMAIER02] 676 Jungmaier, A., Rathgeb, E., and M. Tuexen, "On the use of 677 SCTP in failover scenarios", World Multiconference on 678 Systemics, Cybernetics and Informatics , 7 2002. 680 [NATARAJAN09] 681 Natarajan, P., Ekiz, N., Amer, P., and R. Stewart, 682 "Concurrent Multipath Transfer during Path Failure", 683 Computer Communications , 5 2009. 685 [RFC6458] Stewart, R., Tuexen, M., Poon, K., Lei, P., and V. 686 Yasevich, "Sockets API Extensions for the Stream Control 687 Transmission Protocol (SCTP)", RFC 6458, December 2011. 689 Appendix A. Discussions of Alternative Approaches 691 This section lists alternative approaches for the issues desribed in 692 this document. Although these approaches do not require to update 693 RFC4960, we do not recommend them from the reasons described below. 695 A.1. Reduce Path.Max.Retrans (PMR) 697 Smaller values for Path.Max.Retrans shorten the failover duration. 698 In fact, this is recommended in some research results [JUNGMAIER02] 699 [GRINNEMO04] [FALLON08]. For example, if when Path.Max.Retrans=0, 700 SCTP switches to another destination on a single timeout. This 701 smaller value for Path.Max.Retrans can results in spurious failover, 702 which might be a problem. 704 Unlike SCTP-PF, the interval for heartbeat packets is governed by 705 'HB.interval' even during failover process. 'HB.interval' is usually 706 set in the order of seconds (recommended value is 30 seconds). When 707 the primary path becomes inactive, the next HB can be transmitted 708 only seconds later. Meanwhile, the primary path may have recovered. 709 In such situations, post failover, an endpoint is forced to wait on 710 the order of seconds before the endpoint can resume transmission on 711 the primary path. However, using smaller value for 'HB.interval' 712 might help this situation, but it will be the waste of bandwidth in 713 most cases. 715 In addition, smaller Path.Max.Retrans values also affect 716 'Association.Max.Retrans' values. When the SCTP association's error 717 count (sum of error counts on all ACTIVE paths) exceeds 718 Association.Max.Retrans threshold, the SCTP sender considers the peer 719 endpoint unreachable and terminates the association. Therefore, 720 Section 8.2 in [RFC4960] recommends that Association.Max.Retrans 721 value should not be larger than the summation of the Path.Max.Retrans 722 of each of the destination addresses, else the SCTP sender considers 723 its peer reachable even when all destinations are INACTIVE. To avoid 724 such inconsistent behavior an SCTP implementation SHOULD reduce 725 Association.Max.Retrans accordingly whenever it reduces 726 Path.Max.Retrans. However, smaller Association.Max.Retrans value 727 increases chances of association termination during minor congestion 728 events. 730 A.2. Adjust RTO related parameters 732 As several research results indicate, we can also shorten the 733 duration of failover process by adjusting RTO related parameters 734 [JUNGMAIER02] [FALLON08]. During failover process, RTO keeps being 735 doubled. However, if we can choose smaller value for RTO.max, we can 736 stop the exponential growth of RTO at some point. Also, choosing 737 smaller values for RTO.initial or RTO.min can contribute to keep RTO 738 value small. 740 Similar to reducing Path.Max.Retrans, the advantage of this approach 741 is that it requires no modification to the current specification, 742 although it needs to ignore several recommendations described in the 743 Section 15 of [RFC4960]. However, this approach requires to have 744 enough knowledge about the network characteristics between end 745 points. Otherwise, it can introduce adverse side-effects such as 746 spurious timeouts. 748 Appendix B. Discussions for Path Bouncing Effect 750 The methods described in the document can accelerate the failover 751 process. Hence, they might introduce the path bouncing effect where 752 the sender keeps changing the data transmission path frequently. 753 This sounds harmful to the data transfer, however several research 754 results indicate that there is no serious problem with SCTP in terms 755 of path bouncing effect [CARO04] [CARO05]. 757 There are two main reasons for this. First, SCTP is basically 758 designed for multipath communication, which means SCTP maintains all 759 path related parameters (CWND, ssthresh, RTT, error count, etc) per 760 each destination address. These parameters cannot be affected by 761 path bouncing. In addition, when SCTP migrates the data transfer to 762 another path, it starts with the minimal or the initial CWND. Hence, 763 there is little chance for packet reordering or duplicating. 765 Second, even if all communication paths between the end-nodes share 766 the same bottleneck, the quick failover results in a behavior already 767 allowed by [RFC4960]. 769 Authors' Addresses 771 Yoshifumi Nishida 772 GE Global Research 773 2623 Camino Ramon 774 San Ramon, CA 94583 775 USA 777 Email: nishida@wide.ad.jp 779 Preethi Natarajan 780 Cisco Systems 781 510 McCarthy Blvd 782 Milpitas, CA 95035 783 USA 785 Email: prenatar@cisco.com 787 Armando Caro 788 BBN Technologies 789 10 Moulton St. 790 Cambridge, MA 02138 791 USA 793 Email: acaro@bbn.com 795 Paul D. Amer 796 University of Delaware 797 Computer Science Department - 434 Smith Hall 798 Newark, DE 19716-2586 799 USA 801 Email: amer@udel.edu 802 Karen E. E. Nielsen 803 Ericsson 804 Kistavaegen 25 805 Stockholm 164 80 806 Sweden 808 Email: karen.nielsen@tieto.com