Detecting performance change in enterprise application versions using resource profiles

Abstract

Performance characteristics (i.e., response time, throughput, resource utilization) of enterprise applications change for each version due to feature additions, bug fixes or configuration changes. Therefore, performance needs to be continuously evaluated to detect performance changes (i.e., improvements or regressions). This work proposes a performance change detection process by creating and versioning resource profiles for each application version that is being built. Resource profiles are models that describe the resource demand per transaction for each component of an enterprise application and their control flow. Combined with workload and hardware environment models, resource profiles can be used to predict performance. Performance changes can be identified by comparing the performance metrics resulting from predictions of different resource profile versions (e.g., by observing an increase or decrease of response time). The source of changes in the resulting performance metrics can be identified by comparing the profiles of different application versions. We propose and evaluate an integration of these capabilities into a deployment pipeline of a continuous delivery process.