Kyle Jamieson , and
ACM SenSys 2004, Baltimore, MD, November, 2004.
Network congestion occurs when offered traffic load exceeds available capacity at any point in a network. In wireless sensor networks, congestion causes overall channel quality to degrade and loss rates to rise, leads to buffer drops and increased delays (as in wired networks), and tends to be grossly unfair toward nodes whose data has to traverse a larger number of radio hops.
Congestion control in wired networks is usually done using end-to-end and network-layer mechanisms acting in concert. However, this approach does not solve the problem in wireless networks because concurrent radio transmissions on different ``links'' interact with and affect each other, and because radio channel quality shows high variability over multiple time-scales. We examine three techniques that span different layers of the traditional protocol stack: hop-by-hop flow control, rate limiting source traffic when transit traffic is present, and a prioritized medium access control (MAC) protocol. We implement these techniques and present experimental results from a 55-node in-building wireless sensor network. We demonstrate that the combination of these techniques, Fusion, can improve network efficiency by a factor of three under realistic workloads.
PDF (485 KB)
Our dataset is available online. It contains each packet delivered to the sink of our network as well as detailed per-node statistics on sends, receives, retransmissions, etc. The data is in comma-separated values (CSV) format, with explanations.
Gzipped Tarball (28 MB)