Towards a synthetic load generator framework for validating schedules in HPC environment

Most of the scientific parallel and distributed applications require significant amount of time to complete on huge and expensive computing systems. Often, these applications are executed multiple times, making their scheduling a critical process for the runtime performance. One way to validate effectiveness of a schedule is to run the application on the actual hardware, which takes a lot of time and resources and is largely unfeasible. In this work we present a baseline for an adjustable-scale graph-based parallel workload executor framework for executing custom-scaled parallel applications given as graphs using synthetic load for the tasks, the GRAPPLER. The idea behind GRAPPLER is to operate as a schedule validator whose execution traces can provide valuable insights into runtime bottlenecks of the application, platform and the scheduler.