Generating Samples Of Complex Population Rapidly Using GPUs And Multiple CPUs

Coalescent theory was born in the1980s, compared with the classical population genetics which researches population evolution along with time, it traces back time to speculate the events which happened in a particular time of the evolution history, then creates a phylogenetic tree of the entire population and summarizes the evolution of the descendant genealogy with common ancestor. Coalescent theory can be simulated by compute under a given set of population conditions. Huson’s ms program is a widely used coalescent program which needs a long time to accumulate a large number of simulated dates in order to achieve the best results. This is a big challenge to computer performance, which can be solved by parallel computing.We used CPU and GPU parallel computing to parallel the source program, so that it could accelerate run in the parallel computing platform. We used NVIDIA CUDA and multi-threading library standard Pthread to parallel the source program. After the parallel, under the conditions of experiment, firstly, we identified the best thread proportion in this program though which we could obtain the optimal performance of the program; secondly, the maximum speedup achieved by parallel processing in our program was6under the situation that the samples produced by two programs were identical; thirdly, we analyzed the speed factors and constraints during the study, and identified the crucially optimal points; finally, we found that although the increase of CPU performance could get a certain degree of acceleration, but the speedup was limited, compared with CPU, GPU would obtain higher acceleration, so it could obtain better acceleration by using GPU parallel computing under the required of numerous samples.In order to achieve supercomputing on the desktop, this study took advantage of CPU and GPU computing resources, speeded up the simulation of group samples in program. The parallel process and ideas used in our program will provide reference and basis for the parallelization of more complex coalescent program.