Statistical Detection of Enterprise Network Problems [Vol. 7, No. 1, 1999]

Statistical Detection of Enterprise Network Problems

M. Thottan
Department of Electrical, Computer and Systems Engineering
Jonsson Engineering Center, 6003
Rensselaer Polytechnic Institute
Troy, NY 12180 USA
Email: thottm_AT_ecse.rpi.edu

C. Ji
Department of Electrical, Computer and Systems Engineering
Jonsson Engineering Center, 6003
Rensselaer Polytechnic Institute
Troy, NY 12180 USA
Email: chuanyi_AT_ecse.rpi.edu

Abstract
The detection of network fault scenarios was achieved using an appropriate subset of Management Information Base (MIB) variables. Anomalous changes in the behavior of the MIB variables was detected using a sequential Generalized Likelihood Ratio (GLR) test. This information was then temporally correlated using a duration filter to provide node level alarms which correlated with observed network faults and performance problems. The algorithm was implemented on data obtained from two different network nodes. The algorithm was optimized using five of the nine fault data sets, and it proved general enough to detect three of the remaining four faults. Consistent results were obtained from the second node as well. Detection of most faults occurred in advance (at least 5 minutes) of the fault suggesting the possibility of prediction and recovery in the future.

Keywords: MIB variables, hypothesis testing, Auto Regressive (AR), Generalized Likelihood Ratio (GLR), change detection, duration filter.

JNSM: Vol. 7, No. 1, 1999 Statistical Detection of Enterprise Network Problems [Vol. 7, No. 1, 1999]

NOTE: only abstract of paper available on-line

Back to JNSM main page