EuroPVM/MPI 2008: Recent Advances in Parallel Virtual Machine and Message Passing Interface pp 84-93 | Cite as
A Simple, Pipelined Algorithm for Large, Irregular All-gather Problems
Abstract
We present and evaluate a new, simple, pipelined algorithm for large, irregular all-gather problems, useful for the implementation of the MPI_Allgatherv collective operation of MPI. The algorithm can be viewed as an adaptation of a linear ring algorithm for regular all-gather problems for single-ported, clustered multiprocessors to the irregular problem. Compared to the standard ring algorithm, whose performance is dominated by the largest data size broadcast by a process (times the number of processes), the performance of the new algorithm depends only on the total amount of data over all processes. The new algorithm has been implemented within different MPI libraries. Benchmark results on NEC SX-8, Linux clusters with InfiniBand and Gigabit Ethernet, Blue Gene/P, and SiCortex systems show huge performance gains in accordance with the expected behavior.
Keywords
Block Size Idle Time Collective Operation Communication Round Linear RingPreview
Unable to display preview. Download preview PDF.
References
- 1.Balaji, P., Buntinas, D., Balay, S., Smith, B.F., Thakur, R., Gropp, W.: Nonuniformly communicating noncontiguous data: A case study with PETSc and MPI. In: 21st International Parallel and Distributed Processing Symposium (IPDPS 2007), pp. 1–10 (2007)Google Scholar
- 2.Benson, G.D., Chu, C.-W., Huang, Q., Caglar, S.G.: A comparison of MPICH allgather algorithms on switched networks. In: Dongarra, J., Laforenza, D., Orlando, S. (eds.) EuroPVM/MPI 2003. LNCS, vol. 2840, pp. 335–343. Springer, Heidelberg (2003)CrossRefGoogle Scholar
- 3.Bruck, J., Ho, C.-T., Kipnis, S., Upfal, E., Weathersby, D.: Efficient algorithms for all-to-all communications in multiport message-passing systems. IEEE Transactions on Parallel and Distributed Systems 8(11), 1143–1156 (1997)CrossRefGoogle Scholar
- 4.Gropp, W., Lusk, E.: Reproducible measurements of MPI performance characteristics. In: Margalef, T., Dongarra, J., Luque, E. (eds.) PVM/MPI 1999. LNCS, vol. 1697, pp. 11–18. Springer, Heidelberg (1999)CrossRefGoogle Scholar
- 5.Hedetniemi, S.M., Hedetniemi, T., Liestman, A.L.: A survey of gossiping and broadcasting in communication networks. Networks 18, 319–349 (1988)MathSciNetCrossRefMATHGoogle Scholar
- 6.Krumme, D.W., Cybenko, G., Venkataraman, K.N.: Gossiping in minimal time. SIAM Journal on Computing 21(1), 111–139 (1992)MathSciNetCrossRefMATHGoogle Scholar
- 7.Mamidala, A.R., Vishnu, A., Panda, D.K.: Efficient shared memory and RDMA based design for mpi_allgather over InfiniBand. In: Mohr, B., Träff, J.L., Worringen, J., Dongarra, J. (eds.) PVM/MPI 2006. LNCS, vol. 4192, pp. 66–75. Springer, Heidelberg (2006)CrossRefGoogle Scholar
- 8.Snir, M., Otto, S., Huss-Lederman, S., Walker, D., Dongarra, J.: MPI – The Complete Reference, 2nd edn., vol. 1, The MPI Core. MIT Press, Cambridge (1998)Google Scholar
- 9.Thakur, R., Gropp, W.D., Rabenseifner, R.: Improving the performance of collective operations in MPICH. International Journal on High Performance Computing Applications 19, 49–66 (2004)CrossRefGoogle Scholar
- 10.Träff, J.L.: Efficient allgather for regular SMP-clusters. In: Mohr, B., Träff, J.L., Worringen, J., Dongarra, J. (eds.) PVM/MPI 2006. LNCS, vol. 4192, pp. 58–65. Springer, Heidelberg (2006)CrossRefGoogle Scholar
- 11.Träff, J.L.: Relationships between regular and irregular collective communication operations on clustered multiprocessors. Parallel Processing Letters (2008) (Forthcoming)Google Scholar
- 12.Träff, J.L., Gropp, W., Thakur, R.: Self-consistent MPI performance requirements. In: Cappello, F., Herault, T., Dongarra, J. (eds.) PVM/MPI 2007. LNCS, vol. 4757, pp. 36–45. Springer, Heidelberg (2007)CrossRefGoogle Scholar