Automatic response performance monitoring for real-time software with nondeterministic behaviors

Abstract

We describe a monitoring approach for evaluation of the response performance of services delivered by real-time software systems. Our approach handles certain specification nondeterminism in the behavioral requirements of these systems and is capable of concurrently measuring state-dependent response time intervals. We detect impairments to service performance as response performance failures, i.e., those system response time intervals that statistically exceed some specified maximum delay. While monitoring of behavioral correctness may require a full specification model to detect behavioral failures, our approach detects response time and response performance failures using a reduced timepost-model (TPM). We consider those targets whose: (1) behavior is specified using communicating extended finite state machines; (2) response time objectives are tabular in format. We present an algorithm for deriving an interpretable TPM from these software requirements. We report and comment on an experimental evaluation of the TPM derivation algorithm and the robustness of the approach in the presence of behavioral failures. The target in the evaluation was the call processing program for a small telephone switch.