Performance Monitoring and Analysis of Workflows in ASKALON
Performance analysis is a service that unifies the performance monitoring, instrumentation and analysis for Grid applications and supports the performance bottleneck interpretation.
To understand the performance of WFs on the Grid, we need to collect and analyze a variety of types of data relevant to the execution of the WFs from many sources. A Grid performance analysis service for scientific WFs, utilizing the unified resource monitoring middleware, collects monitoring data from the WF control and invocation services, and performance measurements obtained through the dynamic instrumentation of WF activities, and uses WF graphs to monitor and analyze the performance of WFs during the runtime. Relevant data of WFs including WF graphs and performance metrics are stored, and we develop techniques for comparing the performance of different constructs of WFs. The Grid performance analysis service which combines dynamic instrumentation, activity execution monitoring, and performance analysis of WFs in a single system presents a dynamic and flexible way to conduct the performance monitoring and analysis of scientific WFs. This work is integrated into the SCALEA-G system .
We also target automatic instrumentation and bottleneck detection (e.g., excessive synchronisation and communication times, load imbalance, inefficiency, or non-scalability) within Grid workflows, based on the online data provided by the Monitoring service, or the offline data organised and managed by the Information service. This work continues our experience on developing the Aksum bottleneck detection tool for parallel programs.
Further information about our research on performance monitoring and analysis themes.