High-Performance Physics Simulations Using Multi-Core CPUs and GPGPUs in a Volunteer Computing Context

摘要

This paper presents two conceptually simple methods for parallelizing aParallel Tempering Monte Carlo simulation in a distributed volunteer computingcontext, where computers belonging to the general public are used. The firstmethod uses conventional multi-threading. The second method uses CUDA, agraphics card computing system. Parallel Tempering is described, and challengessuch as parallel random number generation and mapping of Monte Carlo chains todifferent threads are explained. While conventional multi-threading on CPUs iswell-established, GPGPU programming techniques and technologies are stilldeveloping and present several challenges, such as the effective use of arelatively large number of threads. Having multiple chains in ParallelTempering allows parallelization in a manner that is similar to the serialalgorithm. Volunteer computing introduces important constraints to highperformance computing, and we show that both versions of the application areable to adapt themselves to the varying and unpredictable computing resourcesof volunteers' computers, while leaving the machines responsive enough to use.We present experiments to show the scalable performance of these twoapproaches, and indicate that the efficiency of the methods increases withbigger problem sizes.