Accelerating simulators by parallelization is a promising approach. Yet, it is challenging to parallelize and do it in an efficient and scalable way while preserving timing accuracy. To this end, we present a multi-process simulation framework SplitSim. Motivated by SimBricks, SplitSim provides a method to systematically decompose a large instance of simulation into several pieces and run them in parallel, while safeguarding accurate timing. What’s more, it is flexible in terms of deciding the number and the scope of divided pieces. We are adopting SplitSim to several representative simulators, including NS3 and Gem5, in the network area and computer architecture area respectively.
SplitSim: Large-Scale Simulations for Evaluating Network Systems Research
Under submission, 2024.
[Preprint]
SplitSim: Scalable and Parallel Full Data Center System Simulations
In The Fifth Young Architect Workshop (YArch), Vancouver BC, Canada, 2023.
[Poster] [Video]