Effective Fault Diagnosis in Communication Networks

摘要

Today's communication networks are typically large andrnheterogeneous systems. Fault management is very importantrnto keep the normal operations and ensure the network'srnreliability and availability. In this paper, we present a newrnapproach for fault diagnosis. In traditional fault diagnosis,rnalarms are generated and collected when failure happens inrnthe network. Then passive diagnosis is performed to analyzernthese alarms and possible hypotheses of the fault arernproposed. After that, active diagnosis (test) is conducted tornisolate the fault. We propose to conduct earlier active testrnduring the fault diagnosis process, i.e., integrating passiverndiagnosis and active diagnosis together to get betterrndiagnosis performance. We formulate the fault diagnosisrnproblem as a Markov decision process. And we prove arnsufficient condition for early active testing. The sufficientrncondition is used to form a heuristic policy. And further wernapply rollout algorithm on the heuristic policy to get a betterrnpolicy. The performances of passive diagnosis policy, thernheuristic policy and rollout policy are evaluated byrnsimulation. Our test results show that policy based on thernsufficient condition is always better than passive diagnosis,rnwhich demonstrate that earlier active test provides betterrndiagnosis.