idnits 2.17.1 draft-nishida-tsvwg-sctp-failover-00.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** You're using the IETF Trust Provisions' Section 6.b License Notice from 12 Sep 2009 rather than the newer Notice from 28 Dec 2009. (See https://trustee.ietf.org/license-info/) 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 doesn't use any RFC 2119 keywords, yet seems to have RFC 2119 boilerplate text. -- The document date (February 24, 2010) is 5174 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) -- Possible downref: Non-RFC (?) normative reference: ref. 'CARO02' -- Possible downref: Non-RFC (?) normative reference: ref. 'CARO04' -- Possible downref: Non-RFC (?) normative reference: ref. 'CARO05' -- Possible downref: Non-RFC (?) normative reference: ref. 'FALLON08' -- Possible downref: Non-RFC (?) normative reference: ref. 'GRINNEMO04' -- Possible downref: Non-RFC (?) normative reference: ref. 'IYENGAR06' -- Possible downref: Non-RFC (?) normative reference: ref. 'JUNGMAIER02' -- Possible downref: Non-RFC (?) normative reference: ref. 'NATARAJAN08' ** Downref: Normative reference to an Informational RFC: RFC 4690 Summary: 2 errors (**), 0 flaws (~~), 2 warnings (==), 9 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group Y. Nishida 3 Internet-Draft WIDE Project 4 Intended status: Standards Track Natarajan 5 Expires: August 28, 2010 Cisco Systems 6 February 24, 2010 8 Quick Failover Algorithm in SCTP 9 draft-nishida-tsvwg-sctp-failover-00 11 Abstract 13 One of the major advantages in SCTP is supporting multi-homing 14 communication. If an multi-homed end-point has redundant network 15 connections, sctp sessions can have a good chance to survive from 16 network failures by migrating inactive network to active one. 17 However, if we follow the SCTP standard, there can be significant 18 delay for the network migration. During this migration period, SCTP 19 cannot transmit much data to the destination. This issue drastically 20 impairs the usability of SCTP in some situations. This memo 21 describes the issue of SCTP failover mechanism and discuss its 22 solutions which require minimal modification to the current standard. 24 Status of this Memo 26 This Internet-Draft is submitted to IETF in full conformance with the 27 provisions of BCP 78 and BCP 79. 29 Internet-Drafts are working documents of the Internet Engineering 30 Task Force (IETF), its areas, and its working groups. Note that 31 other groups may also distribute working documents as Internet- 32 Drafts. 34 Internet-Drafts are draft documents valid for a maximum of six months 35 and may be updated, replaced, or obsoleted by other documents at any 36 time. It is inappropriate to use Internet-Drafts as reference 37 material or to cite them other than as "work in progress." 39 The list of current Internet-Drafts can be accessed at 40 http://www.ietf.org/ietf/1id-abstracts.txt. 42 The list of Internet-Draft Shadow Directories can be accessed at 43 http://www.ietf.org/shadow.html. 45 This Internet-Draft will expire on August 28, 2010. 47 Copyright Notice 48 Copyright (c) 2010 IETF Trust and the persons identified as the 49 document authors. All rights reserved. 51 This document is subject to BCP 78 and the IETF Trust's Legal 52 Provisions Relating to IETF Documents 53 (http://trustee.ietf.org/license-info) in effect on the date of 54 publication of this document. Please review these documents 55 carefully, as they describe your rights and restrictions with respect 56 to this document. Code Components extracted from this document must 57 include Simplified BSD License text as described in Section 4.e of 58 the Trust Legal Provisions and are provided without warranty as 59 described in the BSD License. 61 Table of Contents 63 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 64 2. Conventions and Terminology . . . . . . . . . . . . . . . . . 4 65 3. Issue in SCTP Path Management Process . . . . . . . . . . . . 5 66 4. Solutions for Smooth Failover . . . . . . . . . . . . . . . . 6 67 4.1. Reduce Path.Max.Retrans . . . . . . . . . . . . . . . . . 6 68 4.2. Adjust RTO related parameters . . . . . . . . . . . . . . 7 69 4.3. Introduce Potential Failure Status in Failure 70 Detection Algorithm . . . . . . . . . . . . . . . . . . . 7 71 5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 9 72 5.1. Effect of Path Bouncing . . . . . . . . . . . . . . . . . 9 73 5.2. Permanent Failover . . . . . . . . . . . . . . . . . . . . 9 74 6. Security Considerations . . . . . . . . . . . . . . . . . . . 10 75 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 11 76 8. Normative References . . . . . . . . . . . . . . . . . . . . . 12 77 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 13 79 1. Introduction 81 Multihoming support is one of the major advantage of SCTP which is 82 not supported in other transport protocols such as TCP or UDP. If an 83 multi-homed end-point has redundant network interfaces, SCTP sessions 84 can survive from the network failures by migrating inactive path to 85 active one. This feature can be expected to be a driving force for 86 deploying SCTP, however, because of minor issues in the SCTP 87 specification, most of SCTP sessions will have significant delay to 88 failover and will cause significant performance degradation during 89 the failover process. We believe this issue is impairing the 90 usability of SCTP and it is important to address it to make SCTP more 91 efficient and attractive. 93 In this memo, we describe the issue of SCTP failover process and 94 discuss the solutions. Our main focus is to propose a solution that 95 does not require major modification to the current standard. Using 96 Concurrent Multipath Transfer (CMT) [IYENGAR06] allows SCTP to 97 utilize multiple paths simultaneously for data transmission. While 98 CMT can reduce the impact of path failures, CMT is not yet a 99 standard. In addition, some may not want concurrent data transfer 100 feature, but want to use smooth failover feature in SCTP. From this 101 reason, we believe the proposals in this document can be useful and 102 meaningful. 104 2. Conventions and Terminology 106 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 107 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 108 document are to be interpreted as described in [RFC2119]. 110 Since this document describes a potential risk in NewReno, it uses 111 the same terminology and definitions in RFC4690. [RFC4690]. 113 3. Issue in SCTP Path Management Process 115 SCTP can utilize multiple IP addresses for single SCTP association. 116 Each SCTP endpoint exchanges the list of available addresses on the 117 node during initial negotiation. After this, endpoints select one 118 address from the list and define this as the destination of the 119 primary path. Basically, SCTP sends all data through this primary 120 path for normal data transmissions. Also, it sends heartbeat packets 121 to other (non-primary) destinations at a certain interval to check 122 the reachability of the path. 124 If sender has multiple active destination addresses, it can 125 retransmit data to secondary destination address when the 126 transmission to the primary times out. 128 When sender receives the acknowledgment for data or heartbeat packets 129 from one of the destination addresses, it considers the destination 130 is active. If it fails to receive acknowledgments, the error count 131 for the address is increased. If the error counter exceeds the 132 protocol parameter 'Path.Max.Retrans', SCTP endpoint considers the 133 address is inactive. 135 The failover process of SCTP is initiated when the primary path 136 becomes inactive (error counter for the primacy path exceeds 137 Path.Max.Retrans). If the primary path is marked inactive, SCTP 138 chooses new destination address from one of the active destinations 139 and start using this address to send data. If the primary path 140 becomes active again, SCTP uses the primary destination for 141 subsequent data transmissions and stop using non-primary one. 143 An issue in this failover process is that it usually takes 144 significant amount of time before SCTP switches to the new 145 destination. Let's say the primary path on a multi-homed host 146 becomes unavailable and the RTO value for the primary path at that 147 time is around 1 second, it usually takes over 60 seconds before SCTP 148 starts to use the secondary path. This is because the recommended 149 value for Path.Max.Retrans in the standard is 5, which requires 6 150 consecutive timeouts before failover takes place. Before SCTP 151 switches to the secondary address, SCTP keeps trying to send packets 152 to the primary and only retransmitted packets are sent to the 153 secondary can be reached at the receiver. This slow failover process 154 can cause significant performance degradation and will not be 155 acceptable in some situations. 157 4. Solutions for Smooth Failover 159 The following approach are conceivable for the solutions of this 160 issue. 162 4.1. Reduce Path.Max.Retrans 164 If we choose smaller value for Path.Max.Retrans, we can shorten the 165 duration of failover process. In fact, this is recommended in some 166 research results [JUNGMAIER02] [GRINNEMO04] [FALLON08]. For example, 167 if we set Path.Max.Retrans to 0, SCTP switches to another destination 168 on a single timeout. However, smaller value for Path.Max.Retrans 169 might cause spurious failover. In addition, if we use smaller value 170 for Path.Max.Retrans, we may also need to choose smaller value for 171 'Association.Max.Retrans'. The Association.Max.Retrans indicates the 172 threshold for the total number of consecutive error count for the 173 entire SCTP association. If the total of the error count for all 174 paths exceeds this value, the endpoint considers the peer endpoint 175 unreachable and terminates the association. According to the Section 176 8.2 in RFC4960, we should avoid having the value of 177 Association.Max.Retrans larger than the summation of the 178 Path.Max.Retrans of all the destination addresses. Otherwise, even 179 if all the destination addresses become inactive, the endpoint still 180 considers the peer endpoint reachable. The behavior in this 181 situation is not defined in the RFC and depends on each 182 implementation. In order to avoid inconsistent behavior between 183 implementations, we had better use smaller value for 184 Association.Max.Retrans. However, if we choose smaller value for 185 Association.Max.Retrans, associations will prone to be terminated 186 with minor congestion. 188 Another issue is that the interval of heartbeat packet: 'HB.interval' 189 may not be small. (recommended value is 30 seconds) This means once 190 failover takes place, an endpoint might need a certain amount of time 191 to use the primary path again. This can cause undesirable effects in 192 case of spurious failover. If we choose smaller value for 193 HB.interval, the traffic used for path probing in a session will be 194 increased. 196 The advantage of tuning Path.Max.Retrans is that it requires no 197 modification to the current standard, although it needs to ignore 198 several recommendations. In addition, some research results indicate 199 path bouncing caused by spurious failover does not cause serious 200 problems. We discuss the effect of path bouncing in the section 5. 202 4.2. Adjust RTO related parameters 204 As several research results indicate, we can also shorten the 205 duration of failover process by adjusting RTO related parameters 206 [JUNGMAIER02] [FALLON08]. During failover process. RTO keeps being 207 doubled. However, if we can choose smaller value for RTO.max, we can 208 stop the exponential growth of RTO at some point. Also, choosing 209 smaller values for RTO.initial or RTO.min can contribute to keep RTO 210 value small. 212 Similar to reducing Path.Max.Retrans, the advantage of this approach 213 is that it requires no modification to the current standard, although 214 it needs to ignore several recommendations. However, this approach 215 requires to have enough knowledge about the network characteristics 216 between end points. Otherwise, it can introduce adverse side-effects 217 such as spurious timeouts. 219 4.3. Introduce Potential Failure Status in Failure Detection Algorithm 221 As seen above, one difficulty of tuning Path.Max.Retrans is that it 222 is required to meet the following two inconsistent requirements. 224 o In order to respond network failure quickly, we need to mark a 225 path as inactive as soon as we detect failure. 227 o In order to make an association persistent and robust against 228 network failure, we need to be conservative to mark a path as 229 inactive. 231 To satisfy these requirements, we propose to introduce "Potentially- 232 failed" (PF) destination state in failure detection algorithm in 233 SCTP. PF state is the intermediate state between Active and 234 Inactive. It indicates that the path is possibly inactive, but not 235 confirmed yet. By using the PF state, SCTP can respond to network 236 failures quickly, while preserving a conservative policy of marking 237 path as inactive. The idea of using PF state was originally proposed 238 in [NATARAJAN08] for CMT. 240 In this algorithm, when sender receives the acknowledgment for data 241 or heartbeat packets from one of the destination addresses, it 242 considers the destination is Active. If it fails to receive 243 acknowledgments, SCTP endpoint increment the error count for the path 244 and transitions the destination to the PF state. (we might need to 245 have new threshold value for error counter to be conservative to 246 migrate from Active to PF. But, we choose this way for now) 248 If the primary path is marked PF, SCTP chooses new destination 249 address from one of the active destinations and starts using this 250 address to send data. SCTP endpoints should not send any data packet 251 to destinations in the PF state, however, it can send heartbeat 252 packets at a certain interval. To allow quick recover from the PF 253 state, we also propose to introduce a new protocol parameter 254 'PFHB.Interval'. PFHB.interval is used to determine the interval of 255 heartbeat packets. It is recommended that a heartbeat packet is sent 256 once per RTO of each destination address plus PFHB.interval with 257 jittering of +/- 50% of the RTO value. (Preethi: wondering why we 258 need jittering?) It is also recommended to use relatively smaller 259 value than HB.interval for PFHB.interval. 261 If the heartbeat is answered, SCTP marks the path Active again. If 262 unanswered, SCTP increments the error count and use an exponential 263 backoff algorithm to increase the RTO. If the error count exceeds 264 Path.Max.Retrans, the path is marked as Inactive. If all 265 destinations are marked PF, SCTP endpoint can choose one destination 266 to send data to its peer. How SCTP chooses a path is implementation 267 specific. One possibility is to select the destination with the 268 least error count. Once a PF destination is chosen for data 269 transmission, the chosen destination must be transitioned from PF to 270 the Active state. Except the use of PFHB.interval, other rules of 271 sending heartbeats are completely the same as those of the standard. 273 The advantage of this approach is that we can keep the same values 274 for Path.Max.Retrans, Association.Max.Retrans and HB.interval used in 275 the current implementations, while it can respond network failure 276 quickly. In addition, new transmission algorithm becomes effective 277 only when the path is in the PF state. When the primary path is in 278 Active or Inactive, the behavior is completely the same as that of 279 the current standard. When the failure detection threshold is most 280 aggressive (PMR=0), both SCTP and SCTP-PF detect path failure after 281 the first timeout. Specifically, SCTP-PF's failure detection does 282 not involve the PF state transition and is equivalent to SCTP's 283 failure detection procedure. In other words, when PMR=0, both SCTP 284 and SCTP-PF perform similarly during path failure. As PMR increases, 285 SCTP's failure detection takes longer and the performance difference 286 between SCTP and SCTP-PF widens (SCTP-PF performs better)." 288 5. Discussion 290 5.1. Effect of Path Bouncing 292 The methods described above can accelerate failover process. Hence, 293 it might introduce path bouncing effect which keeps changing the data 294 transmission path frequently. This sounds harmful for data transfer, 295 however several research results indicate that there is no serious 296 problem with SCTP in terms of path bouncing effect [CARO04] [CARO05]. 298 There are two main reasons for this. First, SCTP is basically 299 designed for multipath communication, which means SCTP maintains all 300 path related parameters (cwnd, ssthresh, RTT, error count, etc) per 301 each destination address. These parameters cannot be affected by 302 path bouncing. In addition, when SCTP migrates to another path, it 303 starts with minimal cwnd because of slow-start. Hence, there is 304 little chance for packet reordering or duplicating. 306 Second, even if all communication paths between end-nodes share the 307 same bottleneck, the proposed method does not make situations worse. 308 In case of congestion, the current standard tries to transmit data 309 packets to the primary during failover, while the proposed method 310 tries to explore other destinations. In any case, the same amount of 311 data packets sent to the same bottleneck. 313 5.2. Permanent Failover 315 When primary path becomes active again after failover, SCTP migrates 316 back to the primary path. After this, SCTP starts data transfer with 317 minimal cwnd. This is because SCTP must perform slow-start when it 318 migrates to new path. However, this might degrade the communication 319 performance in case that the performance of the alternative path is 320 relatively good. In order to mitigate this effect of slow-start, 321 permanent failover was proposed in [CARO02]. Permanent failover 322 allows SCTP to remain the alternative path even if the primacy path 323 becomes active again. This approach can improve performance in some 324 cases, however, it will require more detail analysis since it might 325 impact on SCTP failover algorithm. Since we prefer to keep the 326 current behavior of the standard as possible, we recommend not to 327 take this approach for now. 329 6. Security Considerations 331 There are no new security considerations introduced in this document. 333 7. IANA Considerations 335 This document does not create any new registries or modify the rules 336 for any existing registries managed by IANA. 338 8. Normative References 340 [CARO02] Caro Jr., A., Iyengar, J., Amer, P., Heinz, G., and R. 341 Stewart, "A Two-level Threshold Recovery Mechanism for 342 SCTP", Tech report, CIS Dept, University of Delaware , 343 7 2002. 345 [CARO04] Caro Jr., A., Amer, P., and R. Stewart, "End-to-End 346 Failover Thresholds for Transport Layer Multihoming", 347 MILCOM 2004 , 11 2004. 349 [CARO05] Caro Jr., A., "End-to-End Fault Tolerance using Transport 350 Layer Multihoming", Ph.D Thesis, University of Delaware , 351 1 2005. 353 [FALLON08] 354 Fallon, S., Jacob, P., Qiao, Y., Murphy, L., Fallon, E., 355 and A. Hanley, "SCTP Switchover Performance Issues in WLAN 356 Environments", IEEE CCNC 2008, 1 2008. 358 [GRINNEMO04] 359 Grinnemo, K-J. and A. Brunstrom, "Peformance of SCTP- 360 controlled failovers in M3UA-based SIGTRAN networks", 361 Advanced Simulation Technologies Conference , 4 2004. 363 [IYENGAR06] 364 Iyengar, J., Amer, P., and R. Stewart, "Concurrent 365 Multipath Transfer using SCTP Multihoming over Independent 366 End-to-end Paths.", IEEE/ACM Trans on Networking 14(5), 367 10 2006. 369 [JUNGMAIER02] 370 Jungmaier, A., Rathgeb, E., and M. Tuexen, "On the use of 371 SCTP in failover scenrarios", World Multiconference on 372 Systemics, Cybernetics and Informatics , 7 2002. 374 [NATARAJAN08] 375 Natarajan, P., Ekiz, N., Iyengar, J., Amer, P., and R. 376 Stewart, "Concurrent Multipath Transfer using SCTP 377 Multihoming: Introducing Potentially-failed Destination 378 State", IFIP Networking , 5 2008. 380 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 381 Requirement Levels", BCP 14, RFC 2119, March 1997. 383 [RFC4690] Klensin, J., Faltstrom, P., Karp, C., and IAB, "Review and 384 Recommendations for Internationalized Domain Names 385 (IDNs)", RFC 4690, September 2006. 387 Authors' Addresses 389 Yoshifumi Nishida 390 WIDE Project 391 Endo 5322 392 Fujisawa, Kanagawa 252-8520 393 Japan 395 Email: nishida@wide.ad.jp 397 Preethi Natarajan 398 Cisco Systems 399 425 E. Tasman Drive 400 San Jose, CA 95134 401 USA 403 Email: prenatar@cisco.com