Retransmission policies with transport layer multihoming

Abstract

Abstract-We evaluate several retransmission policies for transport protocols that support multihoming, such as SCTP. We find that schemes that attempt to improve the chance of success by retransmitting to an alternate peer IP address often degrade performance. Our results show that for better performance, new data transmissions and retransmissions should be sent to the same peer IP address. We also find that our Multiple Fast Retransmit algorithm further improves performance by reducing the number of timeouts. Since our results assume reachability of all peer IP addresses, we conclude with suggestions for scenarios where failures are possible. We suggest compromising some of the performance improvements to avoid performance degradation during failures.

Retransmission policies for multihomed transport protocols

“…This paper combines and extends results published by the authors in three incremental conference publications [5][6][7], thereby documenting the complete development of this research. Section 2 demonstrates the problem with SCTPs current retransmission policy (AllRtxAlt) by comparing it to an alternative policy, AllRtxSame (All Retransmissions to Same).…”

“…Thus, Karn's algorithm can be eliminated, and successful retransmissions can be used to update the RTT estimate and maintain a more accurate RTO value. This feature is especially useful in alleviating the stale RTO problem of AllRtxAlt and FrSameRtoAlt [5].…”

“…2 At this point, MFR state stores the highest packet currently outstanding, n. This way, if the retransmission of x is also lost, the sender can detect the loss with another four missing reports. However, this time only sacks for packets greater than n can serve as missing reports, because the sacks up to n were already in flight when x was fast retransmitted the first time [5].…”

“…The extra heartbeats introduced by HAR try to alleviate the stale RTO problem of AllRtxAlt and FrSameRtoAlt. With HAR, a sender updates an alternate destination's RTT estimate more frequently, thus resulting in a better RTT estimate on which to base the RTO value [5].…”

End-to-end failover thresholds for transport layer multihoming

“…The sender uses the following retransmission scheme: (1) send fast retransmissions to the same peer IP address as new data transmissions, (2) send timeout retransmissions to a non-failed alternate peer IP address (if one exists), and (3) employ our Multiple Fast Retransmit algorithm [3]. This scheme differs from the current scheme in RFC2960, but results in [4] show this scheme to perform better.…”