Non-threaded and threaded approaches to Multirail communication with uDAPL

Jie Cai; Alistair P. Rendell; Peter E. Strazdins

doi:10.1109/NPC.2009.10

Non-threaded and threaded approaches to Multirail communication with uDAPL

Jie Cai, Alistair P. Rendell, Peter E. Strazdins

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Citations (Scopus)

Abstract

uDAPL is a portable and platform independent communication library that provides RDMA as well as send/recv operations. Some well-known software has attempted to take advantage of uDAPL's portability, such as Open MPI, MVAPICH2, Intel MPI, and Cluster OpenMP. However, network bandwidth limitations can still be a bottleneck for applications using these software. Engaging a "Multirail" network is a method to by-pass this. In this paper, we design a non-threaded and a threaded approach to improve the performance of uDAPL over multirail configured clusters. The two approaches are evaluated on an InfiniBand cluster with different multirail configurations. The results show that the threaded approach improves by 33% and 148% the uni-directional bandwidth on the multi-port and the multi-HCA configured network respectively, and the nonthreaded approach improves ∼90% of the uni-directional bandwidth on the multi-HCA configured network. A similar improvement is achieved for the bi-directional bandwidth.

Original language	English
Title of host publication	NPC 2009 - 6th International Conference on Network and Parallel Computing
Pages	233-239
Number of pages	7
DOIs	https://doi.org/10.1109/NPC.2009.10
Publication status	Published - 1 Dec 2009
Externally published	Yes
Event	NPC 2009 - 6th International Conference on Network and Parallel Computing - Gold Coast, QLD, Australia Duration: 19 Oct 2009 → 21 Oct 2009

Conference

Conference	NPC 2009 - 6th International Conference on Network and Parallel Computing
Country/Territory	Australia
City	Gold Coast, QLD
Period	19/10/09 → 21/10/09

Access to Document

10.1109/NPC.2009.10Licence: In Copyright

Cite this

@inproceedings{ef71e1b8595b463ea3ffc1e5c5b0e1ac,

title = "Non-threaded and threaded approaches to Multirail communication with uDAPL",

abstract = "uDAPL is a portable and platform independent communication library that provides RDMA as well as send/recv operations. Some well-known software has attempted to take advantage of uDAPL's portability, such as Open MPI, MVAPICH2, Intel MPI, and Cluster OpenMP. However, network bandwidth limitations can still be a bottleneck for applications using these software. Engaging a {"}Multirail{"} network is a method to by-pass this. In this paper, we design a non-threaded and a threaded approach to improve the performance of uDAPL over multirail configured clusters. The two approaches are evaluated on an InfiniBand cluster with different multirail configurations. The results show that the threaded approach improves by 33% and 148% the uni-directional bandwidth on the multi-port and the multi-HCA configured network respectively, and the nonthreaded approach improves ∼90% of the uni-directional bandwidth on the multi-HCA configured network. A similar improvement is achieved for the bi-directional bandwidth.",

author = "Jie Cai and Rendell, {Alistair P.} and Strazdins, {Peter E.}",

year = "2009",

month = dec,

day = "1",

doi = "10.1109/NPC.2009.10",

language = "English",

isbn = "9780769538372",

pages = "233--239",

booktitle = "NPC 2009 - 6th International Conference on Network and Parallel Computing",

note = "NPC 2009 - 6th International Conference on Network and Parallel Computing ; Conference date: 19-10-2009 Through 21-10-2009",

}

TY - GEN

T1 - Non-threaded and threaded approaches to Multirail communication with uDAPL

AU - Cai, Jie

AU - Rendell, Alistair P.

AU - Strazdins, Peter E.

PY - 2009/12/1

Y1 - 2009/12/1

N2 - uDAPL is a portable and platform independent communication library that provides RDMA as well as send/recv operations. Some well-known software has attempted to take advantage of uDAPL's portability, such as Open MPI, MVAPICH2, Intel MPI, and Cluster OpenMP. However, network bandwidth limitations can still be a bottleneck for applications using these software. Engaging a "Multirail" network is a method to by-pass this. In this paper, we design a non-threaded and a threaded approach to improve the performance of uDAPL over multirail configured clusters. The two approaches are evaluated on an InfiniBand cluster with different multirail configurations. The results show that the threaded approach improves by 33% and 148% the uni-directional bandwidth on the multi-port and the multi-HCA configured network respectively, and the nonthreaded approach improves ∼90% of the uni-directional bandwidth on the multi-HCA configured network. A similar improvement is achieved for the bi-directional bandwidth.

AB - uDAPL is a portable and platform independent communication library that provides RDMA as well as send/recv operations. Some well-known software has attempted to take advantage of uDAPL's portability, such as Open MPI, MVAPICH2, Intel MPI, and Cluster OpenMP. However, network bandwidth limitations can still be a bottleneck for applications using these software. Engaging a "Multirail" network is a method to by-pass this. In this paper, we design a non-threaded and a threaded approach to improve the performance of uDAPL over multirail configured clusters. The two approaches are evaluated on an InfiniBand cluster with different multirail configurations. The results show that the threaded approach improves by 33% and 148% the uni-directional bandwidth on the multi-port and the multi-HCA configured network respectively, and the nonthreaded approach improves ∼90% of the uni-directional bandwidth on the multi-HCA configured network. A similar improvement is achieved for the bi-directional bandwidth.

UR - http://www.scopus.com/inward/record.url?scp=73449136872&partnerID=8YFLogxK

UR - http://purl.org/au-research/grants/ARC/LP0669726

U2 - 10.1109/NPC.2009.10

DO - 10.1109/NPC.2009.10

M3 - Conference contribution

AN - SCOPUS:73449136872

SN - 9780769538372

SP - 233

EP - 239

BT - NPC 2009 - 6th International Conference on Network and Parallel Computing

T2 - NPC 2009 - 6th International Conference on Network and Parallel Computing

Y2 - 19 October 2009 through 21 October 2009

ER -

Non-threaded and threaded approaches to Multirail communication with uDAPL

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Cite this