Using distributed memory parallel computers and GPU clusters for multidimensional Monte Carlo integration

摘要

The aim of this paper is to show that the multidimensional Monte Carlo integration can be efficiently implementedrnon various distributed memory parallel computers and clusters of multicore nodes using recentlyrndeveloped parallel versions of linear congruential generator and lagged Fibonacci generator pseudorandomrnnumber generators. We show how to accelerate the overall performance by offloading some computationsrnto Graphics Processing Units (GPUs), and we discuss how to transform Message Passing Interfacern(MPI) + OpenMP programs to MPI + OpenMP + CUDA model. We explain how to utilize multiple coresrnof CPUs together with multiple GPU accelerators within a single node and how to achieve reasonable loadrnbalancing of all computational resources of GPU-accelerated multicore nodes. We present and discuss thernresults of experiments performed on the following target architectures: IBM Blue Gene/Q parallel computer,rna cluster of Intel Xeon E5-2660 servers, and a Tesla-based GPU cluster with Intel Xeon X5650 multicorernprocessors. The results are presented from two points of view: strong scaling and weak scaling. We alsorncompare the performance of all considered architectures.