Summary
SKaMPI-5 is a micro-benchmark for MPI implementations, designed to be easily extensible.
Besides a “global” parallel file system at least some parallel machines offer (many) hard disks which are local to (processors or) computing nodes. While the MPI2 standard includes functions for doing disk IO, MPI is unable to use this resource because by (our) definition the local disks are not visible to remote processors/nodes. But it is possible to write thin layers of code which allow this kind of access. If used with many hard disks, theoretically very high disk IO bandwidths are possible. In the present paper we have a closer look at one such approach, and give an overview of the extension of SKaMPI-5 for the evaluation of this. We also present a few selected results of benchmark runs for MPI-IO, giving an impression of what is possible with the MPI-IO measurements available in SKaMPI-5. In addition we point out several problems one is facing.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Similar content being viewed by others
References
W. Augustin, M.-O. Straub, and T. Worsch. Benchmarking mpi one-sided communication with SKaMPI-5. In W.E. Nagel, W. Jäger, and M. Resch, editors, High Performance Computing in Science and Engineering ’05, pages 329–340. Springer-Verlag, 2005. ISBN 978-3540283775.
A. Ching, A. Choudhary, K. Coloma, W. Liao, R. Ross, and W. Gropp. Noncontiguous access through MPI-IO. In Proc. of the IEEE/ACM Int. Symp. on Cluster Computing and the Grid, pages 104–111, 2003.
W. Gropp, E. Lusk, and R. Thakur. Using MPI-2: Advanced Featuers of the Message-Passing Interface. MIT Press, 1999.
W. Loewe, T. McLarty, and C. Morrone. IOR Benchmark. ftp://ftp.llnl.gov/pub/siop/ior/, 2007.
J. Mathes. Unterstützung paralleler Datei-Ein-/Ausgabe unter MPI. Diploma thesis, Fak. f. Informatik, Univ. Karlsruhe, 2008.
J. Nunez. Los Alamos National Lab MPI-IO Test, User’s Guide, version 1.0, 2006.
A. Perogiannakis. Leistungsmessung paralleler Ein-/Ausgabe in MPI-Bibliotheken (MOI-IO). Diploma thesis, Fak. f. Informatik, Univ. Karlsruhe, 2006.
R. Rabenseifner, A.E. Koniges, J.-P. Prost, and R. Hedges. The parallel effective I/O bandwidth benchmark: b_eff_io. In Christophe Cerin and Hai Jin, editors, Parallel I/O for Cluster Computing, chapter 4, pages 107–132. Kogan Page Ltd, 2004. ISBN 1-903996-50-3.
S. Saini, D. Talcott, R. Thakur, P. Adamidis, R. Rabenseifner, and R. Ciotti. Parallel I/O performance characterization of Columbia and NEC SX-8 superclusters. In Proc. IPDPS 2007, 2007.
T. Worsch, R. Reussner, and W. Augustin. Benchmarking collective operations with SKaMPI. In E. Krause and W. Jäger, editors, High Performance Computing in Science and Engineering ’02, pages 491–502. Springer-Verlag, 2003. ISBN 3-540-43860-2.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 2009 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Mathes, J., Perogiannakis, A., Worsch, T. (2009). Doing IO with MPI and Benchmarking It with SKaMPI-5. In: Nagel, W.E., Kröner, D.B., Resch, M.M. (eds) High Performance Computing in Science and Engineering '08. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-88303-6_41
Download citation
DOI: https://doi.org/10.1007/978-3-540-88303-6_41
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-88301-2
Online ISBN: 978-3-540-88303-6
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)