High Performance RDMA-Based MPI Implementation over InfiniBand

DMCA

High Performance RDMA-Based MPI Implementation over InfiniBand (2003)

Cached

Download Links

Other Repositories/Bibliography

DBLP

by Jiuxing Liu , Jiesheng Wu , Dhabaleswar K. Panda

Venue:	In 17th Annual ACM International Conference on Supercomputing (ICS ’03
Citations:	126 - 31 self

BibTeX

@INPROCEEDINGS{Liu03highperformance,
    author = {Jiuxing Liu and Jiesheng Wu and Dhabaleswar K. Panda},
    title = {High Performance RDMA-Based MPI Implementation over InfiniBand},
    booktitle = {In 17th Annual ACM International Conference on Supercomputing (ICS ’03},
    year = {2003}
}

OpenURL

Abstract

Although InfiniBand Architecture is relatively new in the high performance computing area, it o#ers many features which help us to improve the performance of communication subsystems. One of these features is Remote Direct Memory Access (RDMA) operations. In this paper, we propose a new design of MPI over InfiniBand which brings the benefit of RDMA to not only large messages, but also small and control messages. We also achieve better scalability by exploiting application communication pattern and combining send/receive operations with RDMA operations. Our RDMA-based MPI implementation currently delivers a latency of 6.8 microseconds for small messages and a peak bandwidth of 871 Million Bytes (831 Mega Bytes) per second. Performance evaluation at the MPI level shows that for small messages, our RDMA-based design can reduce the latency by 24%, increase the bandwidth by over 104%, and reduce the host overhead by up to 22%. For large messages, we improve performance by reducing the time for transferring control messages. We have also shown that our new design is beneficial to MPI collective communication and NAS Parallel Benchmarks.

Keyphrases

high performance rdma-based mpi implementation small message large message new design control message rdma-based design infiniband architecture mpi collective communication application communication pattern mpi level show peak bandwidth high performance mega byte na parallel benchmark performance evaluation host overhead send receive operation rdma-based mpi implementation communication subsystem remote direct memory access rdma operation million byte er many feature

DMCA