Computer Science Department
School of Computer Science, Carnegie Mellon University
WiMed: An Infrastructure-less Approach
Kaushik Lakshminarayanan, Srinivasan Seshan, Peter Steenkiste
The availability of unlicensed spectrum coupled with the increasing popularity of wireless communication has given rise to a diverse range of wireless technologies that compete for spectrum. In particular, 802.11 devices face a host of problems such as interference with other 802.11 devices (hidden terminals) as well as with technologies like Bluetooth and ZigBee. Inferring the cause of wireless performance problems based on observations from a single node is hard. Past work on wireless diagnosis has made use of monitoring infrastructures with co-ordinated observations from multiple monitors to tackle this problem.
In this paper, we try to answer the question: "how can we enable users to reason about wireless performance variations without requiring elaborate instrumentation and infrastructure support?". We propose WiMed, a tool that uses only local physical layer measurements from commodity 802.11 NICs (at the node being diagnosed) to diagnose a variety of performance-related problems. We have implemented a WiMed prototype using the MadWifi driver for Atheros NICs. Our results show that WiMed can diagnose non-802.11 interference better than existing systems. WiMed's time allocation map shows how the medium is utilized, which is useful in identifying performance bottlenecks.