From: Jenny Liu (jen@cs.washington.edu)
Date: Wed Oct 20 2004 - 02:35:45 PDT
When TCP fails to control congestion, a more fine-grained congestion control/notification scheme is the next logical step. The authors present the eXplicit Control Protocol to replace TCP. XCP separates the tasks of 1) maximizing bandwidth utilization and 2) determining and enforcing fairness. XCP offers a more pre-emptive congestion control scheme, which saves connections from having to aggressively decrease data flow once the network becomes congested and taking a long time to recover bandwidth utilization. The experimental results in the paper show some dramatic improvements in performance when XCP is used in place of TCP.
XCP may also have some drawbacks. XCP places a heftier portion of the responsibility for congestion control onto routers. Not only does XCP require routers to keep per-link state data, it also exacts an increased computational toll from a router per packet processed.
XCP also places an additional packet header on packets, increasing the overhead required to send a packet over the network.
Despite these drawbacks, the empirical data presented by the authors suggests that XCP does in fact offer performance no(t significantly) worse than TCP in all cases and significantly better than TCP in some cases. Unfortunately, history is working against the deployment of XCP. The TCP protocol is so entrenched that the deployment of a whole new (and better) control protocol (even one that can 'play nice' with TCP) may not ever happen.
This archive was generated by hypermail 2.1.6 : Wed Oct 20 2004 - 02:35:50 PDT