Co-processing SPMD Computation on GPUs and CPUs on Shared Memory System
School of Informatics and Computing, Pervasive Technology Institute, Indiana University, Bloomington, USA
@article{citation-0,
title={Co-processing SPMD Computation on GPUs and CPUs on Shared Memory System},
author={Li, Hui and Fox, Geoffrey and Laszewski, Gregor and Guo, Zhenhua and Qiu, Judy}
}
Heterogeneous parallel system with multiprocessors and accelerators are becoming ubiquitous due to better cost-performance and energy-efficiency. These heterogeneous processor architectures have different instruction sets and are optimized for either task latency or throughput purposes. Challenges occur in regard to programmability and performance when executing SPMD computations on heterogeneous architectures simultaneously. In order to meet these challenges, we implemented a MapReduce runtime system to co-process SPMD job on GPUs and CPUs on shared memory system. We are proposing a heterogeneous MapReduce programming interface for the developer and leverage the two-level scheduling approach in order to efficiently schedule tasks with heterogeneous granularities on the GPUs and CPUs. Experimental results of C-means clustering, matrix multiplication and word count indicate that using all CPU cores increase the GPU performance by 11.5%, 5.1%, and 41.9% respectively.
April 13, 2013 by hgpu
