Single Data Copying for MPI Communication Optimization on Shared Memory System

Authors:
Qiankun Miao;Guangzhong Sun;Jiulong Shan;Guoliang Chen
Affiliations:
Anhui Province-MOST Key Co-Lab of High Performance Computing, and Its Applications, Department of Computer Science, University of Science and Technology of China, Hefei, 230027, P.R. China;Anhui Province-MOST Key Co-Lab of High Performance Computing, and Its Applications, Department of Computer Science, University of Science and Technology of China, Hefei, 230027, P.R. China;Microprocessor Technology Lab, Intel China Research Center, Beijing, China;Anhui Province-MOST Key Co-Lab of High Performance Computing, and Its Applications, Department of Computer Science, University of Science and Technology of China, Hefei, 230027, P.R. China
Venue:
ICCS '07 Proceedings of the 7th international conference on Computational Science, Part I: ICCS 2007
Year:
2007

Citing 5
Cited 0

A high-performance, portable implementation of the MPI message passing interface standard

Parallel Computing
A high-performance MPI implementation on a shared-memory vector supercomputer

Parallel Computing
Compile/run-time support for threaded MPI execution on multiprogrammed shared memory machines

Proceedings of the seventh ACM SIGPLAN symposium on Principles and practice of parallel programming
Multicore CPUs for the Masses

Queue - Multiprocessors
Data Transfers between Processes in an SMP System: Performance Study and Application to MPI

ICPP '06 Proceedings of the 2006 International Conference on Parallel Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Shared memory system is an important platform for high performance computing. In traditional parallel programming, message passing interface (MPI) is widely used. But current implementation of MPI doesn't take full advantage of shared memory for communication. A double data copying method is used to copy data to and from system buffer for message passing. In this paper, we propose a novel method to design and implement the communication protocol for MPI on shared memory system. The double data copying method is replaced by a single data copying method, thus, message is transferred without the system buffer. We compare the new communication protocol with that in MPICH an implementation of MPI. Our performance measurements indicate that the new communication protocol outperforms MPICH with lower latency. For Point-to-Point communication, the new protocol performs up to about 15 times faster than MPICH, and it performs up to about 300 times faster than MPICH for collective communication.