To provide consistent high performance the Askalon project includes a perfomance prediction service. The main tasks of this prediction service are:
- System level prediction: Predicting the future performance of Grid components. This includes network predictions (e.g. delay, available bandwidth) and computer system predictions (e.g. CPU load, memory usage, available storage I/O bandwidth)
- Application level prediction: Predicting the run-time of atomic activities (applications, data transfers). This prediction uses the system level prediction to some extent.
- Workflow level prediction: Predicting the run-time of scientific (sub-)workflows (workflow of scientific applications) on the Grid. This prediction builds on predictions provided by the application level prediction.
The prediction service will be used by the Askalon (meta-) scheduler to derive intelligent scheduling decisions.
The main research topics within the scope of performance prediction are application and workflow level predictions. System level predictions are based on data provided by systems that are already used in the GRID community (e.g. Network Weather Service). In case of application level predictions we want to investigate statistical approaches that are based on historical data (measurements).
- Prof. Dr. Thomas Fahringer
- Farrukh Nadeem