DiffQ - Differential Backlog Congestion Control for Wireless Multi-hop Networks
There has been a recent spur of interest in the use of wireless multi-hop networks especially in the context of a replacement for traditional wired networks in small to medium-sized enterprises. However, as such networks increase in size, data rates, and number of users, it was quickly realized that contemporary congestion control protocols designed for wired networks,
especially TCP, do not translate well into the wireless medium.
With TCP being so ubiquitous, it was a natural first step to take a top-down approach and identify the problems that plague TCP in such wireless environments. Several key problem areas have been identified; to name a few -- TCP's reliance on losses as a congestion signal, TCP's interaction with wireless routing protocols, TCP's aggressive probing for bandwidth and self-interference between TCP data and ACK packets.
As a response to these problems, the research community has proposed several enhancements to TCP, e.g. TCP-FeW, TCP-Veno, TCP-Westwood, TCP-ELFN. Note that such protocols focus on improving the achievable capacity in such networks. However, as the number of such concurrent TCP flows increase, the focus has now shifted to improving the fairness in such networks. The wireless medium is shared among neighboring nodes; thus bandwidth must be allocated fairly among neighboring flows that do not necessarily share the same link. Protocols like TCP face severe unfairness or even starvation of a large number of flows.
A second challenge is the recent interest in exploiting the broadcast feature of wireless networks, leading to the proposal of many creative
protocols including opportunistic routing and network coding. These protocols enable the use of many diverse, yet dynamically changing routing paths. Congestion control for these protocols using traditional end-to-end protocols such as
TCP may result in too conservative rate control.
Inspired by existing theoretical solutions of cross-layer optimization, we develop a protocol, called DiffQ, for congestion control in wireless multi-hop networks. DiffQ can support congestion control for network flows that use either single-path or opportunistic multi-path routing. The protocol is currently implemented in Linux 2.6 series and tested in a network of 46 IEEE 802.11 wireless nodes. It is observed that DiffQ greatly
improves the efficiency and fairness of existing transport protocols that use application-level multi-path routing and single-path routing.
Advisor:Dr. Injong Rhee; Dr. Rudra Dutta; Dr. Do Young Eun; Dr. Khaled Harfoush
School:North Carolina State University
School Location:USA - North Carolina
Source Type:Master's Thesis
Date of Publication:07/08/2008