On the Relation Between SACK Delay and SCTP

On the Relation Between SACK Delay and SCTP Failover Performance for Different Traffic Distributions Johan Eklund, Anna Brunstrom, Karl-Johan Grinnemo Johan. Eklund@kau. se Department of Computer Science Karlstad University, Sweden Karlstad University Department of Computer Science

Outline • SCTP • Multihoming – Failover • Key Factors impacting failover performance • Experimental study – Scenario – Results – Analysis • Conclusion Karlstad University Department of Computer Science

Stream Control Transmission Protocol (SCTP) A new reliable transport protocol (Evolution from TCP) – RFC 4960 – Initially designed for signaling traffic • Robustness • Timing requirements New features: – Message boundary preservation – Multistreaming – Multihoming Option: unordered data delivery Karlstad University Department of Computer Science

Multihoming Primary path If 1 ISP If 1 Internet If 2 Sender ISP Alternate/backup path Karlstad University Department of Computer Science Receiver

Multihoming/Failover Timeout!! ISP Internet Sender ISP Karlstad University Department of Computer Science ISP Receiver

Failover procedure Receiver Primary interface Sender Primary Backup interface Receiver Backup Interface Path Failure Timeout 1 Timeout 2 Timeout PMR+1 Now failover happens Karlstad University Department of Computer Science Failover time

Challenge • Differentiate congestion from failure – avoid spurious failovers • Failover transparent to application – Max transfer time ~1. 0 s – Failover time ~2. 0 s Karlstad University Department of Computer Science (signaling)

Key Factors for Failover Time • RTO-timer at the time of failure – Low RTO => short failover time low maximum transfer time for a message • Exponential backoff of the congestion control – RTO-timer is doubled for every timeout Karlstad University Department of Computer Science

Parameters impacting the RTO-timer • Path Delay – RTO is based on RTT, dependant on the Path delay • Traffic intensity • …. . • SCTP parameters – SACK Delay (Acknowledgments (SACKs) delayed to reduce the network load ) – …. . Karlstad University Department of Computer Science

SACK Delay SACK delay 200 ms SACK delay 0 ms Sender Receiver Delayed SACK (y) RTT x+y Karlstad University Department of Computer Science

Our study on an Emulated network Network Emulator Primary path Sender Alternate path Network Emulator Controller Karlstad University Department of Computer Science Receiver

Impact of SACK Delay on failover time Single messages PMR 2, Path delay 40 ms 2500 Failover time (ms) 2000 SACK delay 0 ms SACK delay 40 ms SACK delay 200 ms 1500 1000 500 0 0 10 20 30 40 50 60 Mean message interval Karlstad University Department of Computer Science 70 80 90

Impact of SACK Delay on max transfer time PMR 2, Path delay 40 ms Max transfer time (ms) 2000 SACK delay 40 SACK delay 200 1500 1000 500 0 0 10 Karlstad University Department of Computer Science 20 30 40 50 60 Mean message interval 70 80 90

Why this impact? PMR 2, Path delay 40 ms 250 SACK delay 40 SACK delay 200 RTO at failure time 200 150 100 50 0 0 10 20 30 40 50 Mean message interval Karlstad University Department of Computer Science 60 70 80 90

Are the results valid for all types of traffic? Bursty traffic PMR 2, Path delay 40 ms 2000 SACK delay 40 SACK delay 200 Failover time (ms) 1500 1000 500 0 0 10 20 30 40 50 Mean message interval Karlstad University Department of Computer Science 60 70 80 90

Reduce the SACK delay to zero… • Provides best efficiency • Comes at a cost –more traffic in form of SACKs 2000 1 SACK delay 0 SACK delay 40 SACK delay 200 SACKs/data packets Failover time (ms) 2500 1000 0. 6 0. 4 0. 2 500 0 SACK delay 40 0. 8 SACK delay 200 0 10 20 30 40 50 60 70 Mean message interval(ms) Karlstad University Department of Computer Science 80 90 0 0 10 20 30 40 50 60 Mean message interval (ms) 70 80 90

Conclusions • SACK delay at default level may significantly reduce SCTP failover performance • Disabling SACK delay significantly increases the traffic load • A SACK delay close to zero improves the failover performance at a moderate traffic load increase Karlstad University Department of Computer Science

Thanks for your attention! Questions? Johan. Eklund@kau. se Karlstad University Department of Computer Science