Network Working Group Y.N. Nishida
Internet-Draft WIDE Project
Intended status: Standards Track P. Natarajan
Expires: January 07, 2012 Cisco Systems
July 06, 2011

Quick Failover Algorithm in SCTP
draft-nishida-tsvwg-sctp-failover-03

Abstract

One of the major advantages in SCTP is supporting multi-homing communication. If a multi-homed end-point has redundant network connections, SCTP sessions can have a good chance to survive from network failures by migrating inactive network to active one. However, if we follow the SCTP standard, there can be significant delay for the network migration. During this migration period, SCTP cannot transmit much data to the destination. This issue drastically impairs the usability of SCTP in some situations. This memo describes the issue of SCTP failover mechanism and discuss its solutions which require minimal modification to the current standard.

Status of this Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on January 07, 2012.

Copyright Notice

Copyright (c) 2011 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.


Table of Contents

1. Introduction

The Stream Control Transmission Protocol (SCTP) [RFC4960] natively supports multihoming at the transport layer -- an SCTP association can bind to multiple IP addresses at each endpoint. SCTP's multihoming features include failure detection and failover procedures to provide network interface redundancy and improved end-to-end fault tolerance.

In SCTP's current failure detection procedure, the sender must experience Path.Max.Retrans (PMR) number of consecutive timeouts on a destination before detecting path failure. The sender fails over to an alternate active destination only after failure detection. Until failover, the sender transmits data on the failed path, degrading SCTP performance. Concurrent Multipath Transfer (CMT) [IYENGAR06] is an extension to SCTP and allows the sender to transmit data on multiple paths simultaneously. Research [NATARAJAN09] shows that the current failure detection procedure worsens CMT performance during failover and can be significantly improved by employing a better failover algorithm.

This document proposes an alternative failure detection procedure for SCTP (and CMT) that improves SCTP (CMT) performance during failover.

2. Conventions and Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].

3. Issue in SCTP Path Management Process

SCTP can utilize multiple IP addresses for a single SCTP association. Each SCTP endpoint exchanges the list of available addresses on the node during initial negotiation. After this, endpoints select one address from the list and define this as the primary destination. During normal transmission, SCTP sends all data to the primary destination. Also, it sends heartbeat packets to other (non-primary) destinations at a certain interval to check the reachability of the path.

If sender has multiple active destination addresses, it can retransmit data to secondary destination address when the transmission to the primary times out.

When sender receives the acknowledgment for data or heartbeat packets from one of the destination addresses, it considers the destination is active. If it fails to receive acknowledgments, the error count for the address is increased. If the error counter exceeds the protocol parameter 'Path.Max.Retrans', SCTP endpoint considers the address is inactive.

The failover process of SCTP is initiated when the primary path becomes inactive (error counter for the primacy path exceeds Path.Max.Retrans). If the primary path is marked inactive, SCTP chooses new destination address from one of the active destinations and start using this address to send data. If the primary path becomes active again, SCTP uses the primary destination for subsequent data transmissions and stop using non-primary one.

An issue in this failover process is that it usually takes significant amount of time before SCTP switches to the new destination. Let's say the primary path on a multi-homed host becomes unavailable and the RTO value for the primary path at that time is around 1 second, it usually takes over 60 seconds before SCTP starts to use the secondary path. This is because the recommended value for Path.Max.Retrans in the standard is 5, which requires 6 consecutive timeouts before failover takes place. Before SCTP switches to the secondary address, SCTP keeps trying to send packets to the primary and only retransmitted packets are sent to the secondary can be reached at the receiver. This slow failover process can cause significant performance degradation and will not be acceptable in some situations.

4. Solutions for Smooth Failover

The following approach are conceivable for the solutions of this issue.

4.1. Reduce Path.Max.Retrans

If we choose smaller value for Path.Max.Retrans, we can shorten the duration of failover process. In fact, this is recommended in some research results [JUNGMAIER02] [GRINNEMO04] [FALLON08]. For example, if we set Path.Max.Retrans to 0, SCTP switches to another destination on a single timeout. However, smaller value for Path.Max.Retrans might cause spurious failover. In addition, if we use smaller value for Path.Max.Retrans, we may also need to choose smaller value for 'Association.Max.Retrans'. The Association.Max.Retrans indicates the threshold for the total number of consecutive error count for the entire SCTP association. If the total of the error count for all paths exceeds this value, the endpoint considers the peer endpoint unreachable and terminates the association. According to the Section 8.2 in [RFC4960], we should avoid having the value of Association.Max.Retrans larger than the summation of the Path.Max.Retrans of all the destination addresses. Otherwise, even if all the destination addresses become inactive, the endpoint still considers the peer endpoint reachable. The behavior in this situation is not defined in the RFC and depends on each implementation. In order to avoid inconsistent behavior between implementations, we had better use smaller value for Association.Max.Retrans. However, if we choose smaller value for Association.Max.Retrans, associations will prone to be terminated with minor congestion.

Another issue is that the interval of heartbeat packet: 'HB.interval' may not be small. (recommended value is 30 seconds) This means once failover takes place, an endpoint might need a certain amount of time to use the primary path again. This can cause undesirable effects in case of spurious failover. If we choose smaller value for HB.interval, the traffic used for path probing in a session will be increased.

The advantage of tuning Path.Max.Retrans is that it requires no modification to the current standard, although it needs to ignore several recommendations. In addition, some research results indicate path bouncing caused by spurious failover does not cause serious problems. We discuss the effect of path bouncing in the section 5.

4.2. Adjust RTO related parameters

As several research results indicate, we can also shorten the duration of failover process by adjusting RTO related parameters [JUNGMAIER02] [FALLON08]. During failover process. RTO keeps being doubled. However, if we can choose smaller value for RTO.max, we can stop the exponential growth of RTO at some point. Also, choosing smaller values for RTO.initial or RTO.min can contribute to keep RTO value small.

Similar to reducing Path.Max.Retrans, the advantage of this approach is that it requires no modification to the current standard, although it needs to ignore several recommendations. However, this approach requires to have enough knowledge about the network characteristics between end points. Otherwise, it can introduce adverse side-effects such as spurious timeouts.

4.3. Introducing Potentially-failed Destination State in Failure Detection Algorithm

Our proposal stems from the following two observations about SCTP's failure detection procedure:

From the above observations it is clear that tweaking the PMR value involves the following tradeoff -- a lower value improves performance but increases the chances of spurious failure detection, whereas a higher value degrades performance and reduces spurious failure detection in a wide range of path conditions. Thus, tweaking the association's PMR value is an incomplete solution to address performance impact during failure.

We propose a new "Potentially-failed" (PF) destination state in SCTP's path management procedure. The PF state was originally proposed to improve CMT performance [NATARAJAN09]. The PF state is an intermediate state between Active and Failed states. SCTP's failure detection procedure is modified to include the PF state. The new failure detection algorithm assumes that loss detected by a timeout implies either severe congestion or failure en-route. After a single timeout on a path, a sender is unsure, and marks the corresponding destination as PF. A PF destination is not used for data transmission except in special cases (discussed below). The new failure detection algorithm requires only sender-side changes. Details are:

5. Discussion

5.1. Effect of Path Bouncing

The methods described above can accelerate failover process. Hence, it might introduce path bouncing effect which keeps changing the data transmission path frequently. This sounds harmful for data transfer, however several research results indicate that there is no serious problem with SCTP in terms of path bouncing effect [CARO04] [CARO05].

There are two main reasons for this. First, SCTP is basically designed for multipath communication, which means SCTP maintains all path related parameters (cwnd, ssthresh, RTT, error count, etc) per each destination address. These parameters cannot be affected by path bouncing. In addition, when SCTP migrates to another path, it starts with minimal cwnd because of slow-start. Hence, there is little chance for packet reordering or duplicating.

Second, even if all communication paths between end-nodes share the same bottleneck, the proposed method does not make situations worse. In case of congestion, the current standard tries to transmit data packets to the primary during failover, while the proposed method tries to explore other destinations. In any case, the same amount of data packets sent to the same bottleneck.

5.2. Permanent Failover

When primary path becomes active again after failover, SCTP migrates back to the primary path. After this, SCTP starts data transfer with minimal cwnd. This is because SCTP must perform slow-start when it migrates to new path. However, this might degrade the communication performance in case that the performance of the alternative path is relatively good. In order to mitigate this effect of slow-start, permanent failover was proposed in [CARO02]. Permanent failover allows SCTP to remain the alternative path even if the primacy path becomes active again. This approach can improve performance in some cases, however, it will require more detail analysis since it might impact on SCTP failover algorithm. Since we prefer to keep the current behavior of the standard as possible, we recommend not to take this approach for now.

6. Security Considerations

There are no new security considerations introduced in this document.

7. IANA Considerations

This document does not create any new registries or modify the rules for any existing registries managed by IANA.

8. References

8.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC4960] Stewart, R., "Stream Control Transmission Protocol", RFC 4960, September 2007.

8.2. Informative References

[IYENGAR06] Iyengar, J., Amer, P. and R. Stewart, "Concurrent Multipath Transfer using SCTP Multihoming over Independent End-to-end Paths.", IEEE/ACM Trans on Networking 14(5), 10 2006.
[NATARAJAN09] Natarajan, P., Ekiz, N., Amer, P. and R. Stewart, "Concurrent Multipath Transfer during Path Failure ", Computer Communications , 5 2009.
[JUNGMAIER02] Jungmaier, A., Rathgeb, E. and M. Tuexen, "On the use of SCTP in failover scenarios ", World Multiconference on Systemics, Cybernetics and Informatics , 7 2002.
[GRINNEMO04] Grinnemo, K-J and A. Brunstrom, "Performance of SCTP-controlled failovers in M3UA-based SIGTRAN networks ", Advanced Simulation Technologies Conference , 4 2004.
[FALLON08] Fallon, S., Jacob, P., Qiao, Y., Murphy, L., Fallon, E. and A. Hanley, "SCTP Switchover Performance Issues in WLAN Environments ", IEEE CCNC 2008, 1 2008.
[CARO04] Caro Jr., A., Amer, P. and R. Stewart, "End-to-End Failover Thresholds for Transport Layer Multihoming ", MILCOM 2004 , 11 2004.
[CARO05] Caro Jr., A., "End-to-End Fault Tolerance using Transport Layer Multihoming ", Ph.D Thesis, University of Delaware , 1 2005.
[CARO02] Caro Jr., A., Iyengar, J., Amer, P., Heinz, G. and R. Stewart, "A Two-level Threshold Recovery Mechanism for SCTP ", Tech report, CIS Dept, University of Delaware , 7 2002.

Authors' Addresses

Yoshifumi Nishida WIDE Project Endo 5322 Fujisawa, Kanagawa 252-8520 Japan EMail: nishida@wide.ad.jp
Preethi Natarajan Cisco Systems 510 McCarthy Blvd Milpitas, CA 95035 USA EMail: prenatar@cisco.com