Advanced notice :
We will be upgrading wiki.geant.org from the current version of Confluence Server to the current LTS version 8.5. During the maintenance window we expect that there will be an outage of 20 minutes.
Maintenance start time: 22/10/2024 16:00 UTC. Maintenance end time: 22/10/2024 18:00 UTC.
h1. Yeah-TCP
Heuristic TCP protocol gained from experimental evaluation of [CUBIC|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBCubicTcp],[HSTCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBHighSpeedTCP], [H-TCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBHamiltonTCP], Africa, [Compound TCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBCompoundTCP].
Algorithm Design Goals:
* Network capacity should be exploited efficiently. This is the most obvious goal, which can be achieved by modifying the congestion window update rules; [YeAH-TCP|YeahTcp] can exploit anyone of the increment rules of other proposals \(e.g. [STCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBScalableTcp],[H-TCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBHamiltonTCP], etc.\).
* The stress induced to the network should be less or equal than that induced by [Reno TCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBTcpReno].
* TCP friendliness with [Reno|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBTcpReno] traffic. A "politically" acceptable algorithm should be able to compete fairly with Reno flows, avoiding starvation of competing flows, and simultaneously exploiting the link capacity.
* The algorithm should be internally and RTT fair.
* Performance should not be substantially impaired by non-congestion related \(random\) packet loss events; random packet loss cannot be ruled out even in case of high speed optical backbones. Reasonable values of this loss depend on the technological context, but we verify that even a loss rate in the order of 10^-7 can give rise to sensitive performance degradation.
* [ Small link buffers|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBNetworkBufferSizing] should not prevent high performance. It is not feasible to design buffer size equal to the bandwidth-delay product in [high BDP links|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBLongFatNetworks] as required by standard Reno congestion control. This goal can be achieved by adopting a decrease policy in case of packet loss similar to the Westwood algorithm.
[YeAH-TCP|YeahTcp] has two different operating modes, "Fast" and "Slow", like Africa TCP. During the "Fast" mode, [YeAH-TCP|YeahTcp] increments the congestion window according to an aggressive rule \(e.g. [STCP|%SCRIPTURL{"view"}%cgibintwikiviewPERTKBScalableTcp] rule\). The state is decided according to the estimated number of packets in the bottleneck queue \(inferred from the measured RTTs\).
h2. References
* _YeAH-TCP: Yet Another Highspeed TCP,_ A. Baiocchi, A. Castellani, F. Vacirca, PFLDnet 2007, Feb. 2007 \([PDF|pfldnet2007:paper^YeAH_TCP.pdf]\)
-- Main.ChrisWelti - 26 Feb 2007