Efficient active probing for fault diagnosis in large scale and noisy networks

  • Authors:
  • Lu Cheng;Xuesong Qiu;Luoming Meng;Yan Qiao;Raouf Boutaba

  • Affiliations:
  • State Key laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China;State Key laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China;State Key laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China;State Key laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China;Cheriton School of Computer Science, University of Waterloo, Waterloo, Canada

  • Venue:
  • INFOCOM'10 Proceedings of the 29th conference on Information communications
  • Year:
  • 2010

Quantified Score

Hi-index 0.01

Visualization

Abstract

Active probing is an effective tool for monitoring networks. By measuring probing responses, we can perform fault diagnosis actively and efficiently without instrumentation on managed entities. In order to reduce the traffic generated by probing messages and the measurement infrastructure costs, an optimal set of probes is desirable. However, the computational complexity for obtaining such an optimal set is very high. Existing works assume single-fault scenarios, apply only to small size networks, or use simplistic methods that are vulnerable to noises. In this paper, by exploiting the conditionally independent property in Bayesian networks, we prove a theorem on the information provided by a set of probes. Based on this theorem and structure property of Bayesian networks, we propose two approaches which can effectively reduce the computation time. A highly efficient adaptive probing algorithm is then presented. Compared with previous techniques, experiments have shown that our approach is more efficient in selecting an optimal set of probes without degrading diagnosis quality in large scale and noisy networks.