Topology-aware method for optimizing collective communication operations in distributed computer systems
Abstract
Interconnection networks of modern distributed computer systems are now hierarchical. In such systems communication time between processor cores depends on a placement of processors in a system. For example, in large-scale SMP computer clusters first two levels are formed by switches of two-stage fat tree and a third level is presented by a shared memory of computer nodes. In this paper, we describe a dynamic optimization method for collective communication operations on hierarchical computer clusters. Our approach exploits knowledge of the L-level hierarchy and is based on a mapping of intensively communicating processes into the same computer node. Performance results on multicore SMP computer clusters with In?niBand and Gigabit Ethernet networks indicate that the Allgather algorithms based on our approach outperform the original Allgather algorithms.
About the Author
М. M. Kurnosov
Сибирский государственный университет телекоммуникаций и информатики
Russian Federation
Russian Federation
References
1. Хорошевский В.Г. Архитектура вычислительных систем. - М.: МГТУ им. Н.Э. Баумана, 2008. - 520 с.
2. Евреинов Э.В., Хорошевский В.Г. Однородные вычислительные системы. - Новосибирск: Наука, 1978. - 320 с.
3. Rabenseifner R. Automatic MPI Counter Profiling // Proceedings of the 42nd Cray User Group. - Noorwijk, The Netherlands, 2000. - 19 pp.
4. Han D., Jones T.. MPI Profiling // Technical Report UCRL-MI-209658 - Lawrence Livermore National Laboratory, USA, 2004. - 15 pp.
5. Thakur R., Rabenseifner R., and Gropp W. Optimization of collective communication operations in MPICH // Int. Journal of High Performance Computing Applications. - 2005. - Vol. 19, No. 1. - P. 49-66.
6. Balaji P., Buntinas D., Goodell D., Gropp W., Kumar S., Lusk E., Thakur R. and Traff J. L. MPI on a Million Processors // Proc. of the PVM/MPI - Berlin: Springer-Verlag, 2009 .- P. 20-30.
7. Khoroshevsky V., Kurnosov M. Mapping Parallel Programs into Hierarchical Distributed Computer Systems // Proc. of "Software and Data Technologies". − Sofia: INSTICC, 2009. − Vol. 2 .− P. 123-128.
8. Bruck J., Ho C.-T., Kipnis S., Upfal E. and Weathersby D. Efficient algorithms for all-to-all communications in multiport message-passing systems // IEEE Transactions on Parallel and Distributed Systems. - 1997. - Vol. 8 (11). - P. 1143-1156.
9. Karypis G. and Kumar V. A Fast and Highly Quality Multilevel Scheme for Partitioning Irregular Graphs // SIAM Journal on Scientific Computing. - 1999. - Vol. 20, No. 1, P. 359-392.
10. Schloegel K., G. Karypis, V. Kumar. Graph partitioning for high-performance scientific simulations // Sourcebook of parallel computing. - San Franciso: Morgan Kaufmann Publish, 2003. - P. 491-541.
11. Курносов М.Г. Алгоритмы вложения параллельных программ в иерархические распределённые вычислительные системы // Вестник СибГУТИ. − 2009. − № 2 (6). − С. 20-45.
Review
For citations:
M. Kurnosov Topology-aware method for optimizing collective communication operations in distributed computer systems. The Herald of the Siberian State University of Telecommunications and Information Science. 2010;(2):54-65. (In Russ.)
Views:
714
Email this article 