Simulating Batch and Application Level Scheduling Using GridSim and SimGrid

Modern high performance computing (HPC) sys- tems are increasing in the complexity of their design and in the levels of parallelism they offer. Studying and enhancing scheduling in HPC became very interesting for two main as- pects. First, scheduling decisions are taken by different types of schedulers such as batch, application, process, and thread schedulers. Second, simulation has become an important tool to examine the design of HPC systems. Therefore, in this work, we study the simulation of different scheduling levels. We used two well-known simulation toolkits, SimGrid and GridSim, in order to support two different scheduling levels, batch and application level scheduling. Each toolkit is extended to support both levels. Moreover, three different scheduling algorithms for each level are implemented and their performance is examined through a real workload dataset. Finally, a comparison for the extension challenges of the two simulators is conducted.