A High-Performance Superpipeline Protocol for InfiniBand

Alexandre Denis 1, 2
1 RUNTIME - Efficient runtime systems for parallel architectures
Inria Bordeaux - Sud-Ouest, UB - Université de Bordeaux, CNRS - Centre National de la Recherche Scientifique : UMR5800
Abstract : InfiniBand high performance networks require that the buffers used for sending or receiving data are registered. Since memory registration is an expensive operation, some communication libraries use caching (rcache) to amortize its cost, and copy data into pre-registered buffers for small messages. In this paper, we present a software protocol for InfiniBand that always uses a memory copy, and amortizes the cost of this copy with a superpipeline to overlap the memory copy and the RDMA. We propose a performance model of our protocol to study its behavior and optimize its parameters. We have implemented our protocol in the NewMadeleine communication library. The results of MPI benchmarks show a significant improvement in cache-unfriendly applications that do not reuse the same memory blocks all over the time, without degradation for cache-friendly applications.
Document type :
Conference papers
Complete list of metadatas

Cited literature [13 references]  Display  Hide  Download

https://hal.inria.fr/inria-00586015
Contributor : Alexandre Denis <>
Submitted on : Thursday, April 14, 2011 - 2:08:43 PM
Last modification on : Thursday, January 11, 2018 - 6:22:12 AM
Long-term archiving on : Thursday, November 8, 2012 - 4:35:33 PM

File

main.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00586015, version 1

Citation

Alexandre Denis. A High-Performance Superpipeline Protocol for InfiniBand. Euro-Par 2011, Aug 2011, Bordeaux, France. pp.276-287. ⟨inria-00586015⟩

Share

Metrics

Record views

472

Files downloads

2996