Evaluating Dynamic Communicators and One-Sided Operations for Current MPI Libraries

Abstract

In this paper we evaluate the current status and performance of several MPI implementations regarding two chapters of the MPI-2 specification. First, we analyze whether the performance using dynamically created communicators is comparable to the approach presented in MPI-1 using a static communicator for different MPI libraries. We then evaluate whether the communication performance of one-sided communication on current machines represents a benefit or a disadvantage to the end-user compared to the more conventional two-sided communication.

Keywords

Benchmarking dynamic process management MPI one-sided communication

Get full access to this article

View all access options for this article.

References

Gabriel, E. , Resch, M. , and Ruehle, R. 2001. Implementing and benchmarking derived data types for metacomputing. High Performance Computing and Networking, B. Hertzberger , A. Hoekstra , and R. William , editors, Springer-Berlag, Berlin , pp. 493–502.

Gabriel, E. , Fagg, G.E. , and Dongarra, J.J. 2003. Evaluating the Performance of MPI-2 dynamic communicators and onesided operations. Recent Advances in Parallel Virtual Machine and Message Passing Interface, Lecture Notes in Computer Science Vol. 2840, J.J. Dongarra , D. Laforenza , and S. Orlande , editors, Springer-Verlag, Berlin , pp. 88–97.

Gropp, W. and Lusk, E. 1999. Reproducible measurements of MPI performance characteristics. Recent Advances in Parallel Virtual Machine and Message Passing Interface, J.J. Dongarra , E. Luque , and T. Margalef , editors, Springer Verlag, Berlin , pp. 11–18.

Hempel, R. 1999. Basic message passing benchmarks, methodology and pitfalls . Presentation at SPEC Workshop, Wuppertal, Germany, September.

Message Passing Interface Forum . 1995. MPI: A Message-Passing Interface Standard (version 1.1). Technical Report, June, http://www.mpi-forum.org.

Message Passing Interface Forum . 1997. MPI-2: Extensions to the Message Passing Interface, July 18.

Mierendorff, H. , Cassirer, K. , and Schwamborn, H. 2000. Working with MPI benchmark suites on ccNUMA architectures. Recent Advances in Parallel Virtual Machine and Message Passing Interface, J. Dongarra , P. Kacsuk , and N. Podhorszki , editors, Springer-Verlag, Berlin , pp. 18–26.

Rabenseifner, R. and Koniges, A.E. 2000. Effective File-I/O Bandwidth Benchmark. Proceedings of the Euro-Par 2000, A. Bode , T. Ludwig , and R. Wissmueller , editors, Springer Verlag, Berlin , pp. 1273–1283.

Reussner, R.H. , Sanders, P. , Prechelt, L. , and Mller, M. 1999. SKaMPI: a detailed accurate MPI benchmark. Recent Advances in Parallel Virtual Machine and Message Passing Interface, V. Alexandrov and J.J. Dongarra , editors, Springer-Verlag, Berlin , pp. 52–59.

10.

Thakur, R. , Gropp, W. , and Toonen, B. 2004. Minimizing synchronization overhead in the implementation of MPI one-sided communication. Recent Advances in Parallel Virtual Machine and Message Passing Interface, Lecture Notes in Computer Science Vol. 3241, D. Kranzlmueller , P. Kacsuk , and J.J. Dongarra , editors, Springer-Verlag, Berlin , pp. 57–67.