Datacenter TCP (DCTCP)

Large datacenter networks pose specific problems for TCP congestion control. For example, bursty traffic from multiple sources and relatively small switch buffers can lead to Incast loss.

DCTCP was developed to address these issues. It employs ECN in a non-traditional way. As a result, DCTCP cannot easily coexist with other TCP variants that react to ECN signals in the traditional way. In particular, in situations where ACKnowledgements are lost (e.g. a congested return path), the sender cannot unambiguously determine the amount of congestion that ECN tried to signal. This is illustrated in Appendix A of RFC 7560, see reference below.

In response to this type of issues, the Prague TCP effort was started in July 2015 to enhance DCTCP so that it can peacefully coexist with other TCP variants.

References

  • draft-ietf-tcpm-dctcp-04, Datacenter TCP (DCTCP): TCP Congestion Control for Datacenters, Stephen Bensley, Lars Eggert, Dave Thaler, Praveen Balasubramanian, Glenn Judd, February 2017
  • RFC 7560, Problem Statement and Requirements for Increased Accuracy in Explicit Congestion Notification (ECN) Feedback, M. Kühlewind, Ed., R. Scheffenegger, B. Briscoe, August 2015

-- Simon Leinen - 2015-08-02 - 2017-02-09

Edit | Attach | Watch | Print version | History: r7 < r6 < r5 < r4 < r3 | Backlinks | Raw View | Raw edit | More topic actions
Topic revision: r7 - 2017-02-09 - SimonLeinen
 
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2004-2009 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.