Detection and repair of faulty access points

In large-scale infrastructure wireless networks several access points (APs) may be unusable at any given moment in time. Unlike completely failed APs, whose failure can be detected by probes to their wired interface, an AP with a faulty wireless interface or whose antenna has been accidentally shielded can only be diagnosed by the actual use of the wireless interface for data communication. We present several algorithms that detect such failed access points by online analysis of AP usage logs. In particular, we demonstrate that we can exploit device mobility to detect faulty APs. We also present efficient heuristics to select a path for a technician to repair failed access points. We evaluate our algorithms using actual log files from an infrastructure network at Dartmouth College. We find that our best algorithm is able to detect nearly 90% of failed access points simply by processing log files. Compared to a naive approach, our algorithm has more than six times fewer false positives. We are also able to construct tours that are up to an order of magnitude more effective that a straightforward greedy approach. Our algorithms require no modifications to either APs or devices. We believe that these properties make our work immediately applicable to real-world scenarios