Anomaly detection and its application in grid

Abstract

Size and complexity of grid computing systems increase rapidly. Failures can have adverse effect on application executing on the system. Thus, failure information should be provided to the scheduler to make scheduling decisions that reduce effect of failures. In this paper, we survey anomaly detection methods and present in details window-based method together with experiments to evaluate it. The experiments show that window-based method is efficient in detecting anomalies. It may be implemented in a monitoring service. We introduce architecture of services to make failure information available for other services in grid. Using these services, grid schedulers and job/execution managers can get additional information for matching and handling task failures.