TCP-DCR: Making TCP Robust to Non-congestion Events

Abstract

In this paper, we propose and evaluate TCP-DCR. TCP-DCR makes simple modifications to the TCP congestion control algorithm to make it more robust to non-congestion events. The key idea here is to delay the congestion response of TCP for a short interval of time τ, thereby creating room for local recovery mechanisms to handle any non-congestion events that may have occurred. If at the end of the delay t, the event is not handled, then it is treated as a congestion loss. We evaluate TCP-DCR through analysis and simulations. The evaluation is done for three scenarios — a wireless network with channel errors, a wired network with packet reordering and a network with zero non-congestion events. The simulation results show that significant performance improvements can be achieved by using TCP-DCR in the presence of non-congestion events with zero or marginal impact in the absence of non-congestion events. TCP-DCR remains fair to the native implementations of TCP that respond to congestion immediately after receiving three dupacks. TCP-DCR is a simple, effective scheme providing a unified solution to several problems with minimal implementation overhead.

This work is supported in part by a grant from The Texas Higher Education Board, by NSF grant ANI-0087372 and by Intel Corp.

An extended version of this paper is available in [22]

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.