The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|S. Li, T. Hoefler, M. Snir:
|NUMA-Aware Shared Memory Collective Communication for MPI
(Vol , Nr. , In Proceedings of the 22nd international symposium on High-performance parallel and distributed computing, presented in New York City, NY, USA, pages 85--96, ACM, ISSN: , ISBN: 978-1-4503-1910-2, Jun. 2013, )
Nominated for Best Paper Award at HPDC'13 (3/20)
AbstractAs the number of cores per node keeps increasing, it becomes increasingly important for MPI to leverage shared-memory for intra-node communication. This paper investigates the design and optimizations of MPI collectives for clusters of NUMA nodes. We develop performance models for collective communication using shared memory, and develop several algorithms for various collectives. Experiments are conducted on both Xeon X5650 and Opteron 6100 InfiniBand clusters. The measurements are in agreement with the model and indicate that different algorithms dominate for short vectors and long vectors. We compare our shared memory allreduce with several traditional MPI implementations, i.e., Open MPI, MPICH2, and MVAPICH2, which utilize system shared memory to facilitate inter-process communication. On 16 nodes Xeon cluster and 8 nodes Opteron clusters, our implementation achieves on average 2.5X and 2.3X speedup over MVAPICH2, respectively. Our techniques enable an efficient implementation of collective operations on future multi- and manycore systems.