Traffic Dynamics-Aware Probe Selection for Fault Detection in Networks

Abstract

Fault detection in modern networks is done with a set of specially instrumented nodes which send probes to find faults. These probes generate additional traffic in network and compete with other regular traffic for bandwidth. In this paper we consider the problem of dynamically adapting the probes based on traffic dynamics experienced by nodes. We propose to profile the links and nodes to get aggregate I/O statistics in a time window and use it as an instantaneous measure of congestion. We consider the network with I/O statistics to generate a weighted graph and formulate an optimization problem to find a set of probes covering whole network with minimum weight. By showing that finding minimum weight probes maps to a known NP complete problem, we propose three greedy algorithms for selecting probes. With both simulation and real graphs of Internet Service Provider (ISP) networks, we perform five sets of experiments and show that proposed algorithms can dynamically adapt to changes in traffic dynamics and also can select probes in large networks in reasonable time.