Insights into the Design of Congestion Control Protocols for Multi-Hop Wireless Mesh Networks

The widespread deployment of multi-hop wireless mesh networks will depend on the performance seen by the user. Unfortunately, the most predominant transport protocol, TCP, performs poorly over such networks, even leading to starvation in some topologies. In this work, we characterize the root causes of starvation in 802.11 scheduled multi-hop wireless networks via simulations. We analyze the performance of three categories of transport protocols. (1) end-to-end protocols that require implicit feedback (TCP SACK), (2) Explicit feedback based protocols (XCP and VCP) and (3) Open-loop protocol (UDP). We ask and answer the following questions in relation to these protocols: (a) Why does starvation occur in different topologies? Is it intrinsic to TCP or, in general, to feedback-based protocols? or does it also occur in the case of open-loop transfers such as CBR over UDP? (a) What is the role of application behavior on transport layer performance in multi-hop wireless mesh networks? (b) Is sharing congestion in the wireless neighborhood essential for avoiding starvation? (c) For explicit feedback based transport protocols, such as XCP and VCP, what performance can be expected when their capacity estimate is inaccurate? Based on the insights derived from the above analysis, we design a rate-based protocol called VRate that uses the two ECN bits for conveying load feedback information. VRate achieves near optimal rates when configured with the correct capacity estimate