An approach to selecting metrics for detecting performance problems in information systems

摘要: other quality of service considerations. We view an information system as a network queues. As such, Early detection o f performance problems is essential to limit their scope and impact. Most commonly, are detected b y applying threshold tests set metrics. For ezample, suppose that disk utilization metric, its value 80%. Then, alarm raised if exceeds Unfortunately, the ad hoc manner in which metrics selected often results false alarms and/or failing detect until serious degradations result. To address this situation, we construct r ules for metric selection based on analytic comparisons statistical power equations five widely used metrics: departure counts (D), number (L), response times (E), servace (S), utilizations (U). These rules ssessed context CPU p aging sub-systems production computer system.