Hierarchical network coding for collective communication on HPC interconnects

Ahmed Shalaby, Mohamed El Sayed Ragab, Victor Goulart, Ikki Fujiwara, Michihiro Koibuchi

Research output: Contribution to conferencePaper

1 Citation (Scopus)

Abstract

Network bandwidth is a performance concern especially for collective communication because the bisection bandwidth of recent supercomputers is far less than their full bisection bandwidth. In this context we propose to exploit the use of a network coding technique to reduce the number of unicasts and the size of transferred data generated by latency-sensitive collective communication in supercomputers. Our proposed network coding scheme has a hierarchical multicasting structure with intra-group and inter-group unicasts. Quantitative analysis show that the aggregate path hop counts by our hierarchical network coding decrease as much as 94% when compared to conventional unicast-based multicasts. We validate these results by cycle-accurate network simulations. In 1,024-switch networks, the network reduces the execution time of collective communication as much as 64%. We also show that our hierarchical network coding is beneficial for any packet size.

Original languageEnglish
Pages98-102
Number of pages5
DOIs
Publication statusPublished - Jan 1 2014
Event2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014 - Turin, Italy
Duration: Feb 12 2014Feb 14 2014

Other

Other2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014
CountryItaly
CityTurin
Period2/12/142/14/14

Fingerprint

Network coding
Supercomputers
Communication
Bandwidth
Multicasting
Switches
Chemical analysis

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Software

Cite this

Shalaby, A., Ragab, M. E. S., Goulart, V., Fujiwara, I., & Koibuchi, M. (2014). Hierarchical network coding for collective communication on HPC interconnects. 98-102. Paper presented at 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014, Turin, Italy. https://doi.org/10.1109/PDP.2014.58

Hierarchical network coding for collective communication on HPC interconnects. / Shalaby, Ahmed; Ragab, Mohamed El Sayed; Goulart, Victor; Fujiwara, Ikki; Koibuchi, Michihiro.

2014. 98-102 Paper presented at 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014, Turin, Italy.

Research output: Contribution to conferencePaper

Shalaby, A, Ragab, MES, Goulart, V, Fujiwara, I & Koibuchi, M 2014, 'Hierarchical network coding for collective communication on HPC interconnects', Paper presented at 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014, Turin, Italy, 2/12/14 - 2/14/14 pp. 98-102. https://doi.org/10.1109/PDP.2014.58
Shalaby A, Ragab MES, Goulart V, Fujiwara I, Koibuchi M. Hierarchical network coding for collective communication on HPC interconnects. 2014. Paper presented at 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014, Turin, Italy. https://doi.org/10.1109/PDP.2014.58
Shalaby, Ahmed ; Ragab, Mohamed El Sayed ; Goulart, Victor ; Fujiwara, Ikki ; Koibuchi, Michihiro. / Hierarchical network coding for collective communication on HPC interconnects. Paper presented at 2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014, Turin, Italy.5 p.
@conference{55b523941cd647ada97bbc2b8acc92c2,
title = "Hierarchical network coding for collective communication on HPC interconnects",
abstract = "Network bandwidth is a performance concern especially for collective communication because the bisection bandwidth of recent supercomputers is far less than their full bisection bandwidth. In this context we propose to exploit the use of a network coding technique to reduce the number of unicasts and the size of transferred data generated by latency-sensitive collective communication in supercomputers. Our proposed network coding scheme has a hierarchical multicasting structure with intra-group and inter-group unicasts. Quantitative analysis show that the aggregate path hop counts by our hierarchical network coding decrease as much as 94{\%} when compared to conventional unicast-based multicasts. We validate these results by cycle-accurate network simulations. In 1,024-switch networks, the network reduces the execution time of collective communication as much as 64{\%}. We also show that our hierarchical network coding is beneficial for any packet size.",
author = "Ahmed Shalaby and Ragab, {Mohamed El Sayed} and Victor Goulart and Ikki Fujiwara and Michihiro Koibuchi",
year = "2014",
month = "1",
day = "1",
doi = "10.1109/PDP.2014.58",
language = "English",
pages = "98--102",
note = "2014 22nd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2014 ; Conference date: 12-02-2014 Through 14-02-2014",

}

TY - CONF

T1 - Hierarchical network coding for collective communication on HPC interconnects

AU - Shalaby, Ahmed

AU - Ragab, Mohamed El Sayed

AU - Goulart, Victor

AU - Fujiwara, Ikki

AU - Koibuchi, Michihiro

PY - 2014/1/1

Y1 - 2014/1/1

N2 - Network bandwidth is a performance concern especially for collective communication because the bisection bandwidth of recent supercomputers is far less than their full bisection bandwidth. In this context we propose to exploit the use of a network coding technique to reduce the number of unicasts and the size of transferred data generated by latency-sensitive collective communication in supercomputers. Our proposed network coding scheme has a hierarchical multicasting structure with intra-group and inter-group unicasts. Quantitative analysis show that the aggregate path hop counts by our hierarchical network coding decrease as much as 94% when compared to conventional unicast-based multicasts. We validate these results by cycle-accurate network simulations. In 1,024-switch networks, the network reduces the execution time of collective communication as much as 64%. We also show that our hierarchical network coding is beneficial for any packet size.

AB - Network bandwidth is a performance concern especially for collective communication because the bisection bandwidth of recent supercomputers is far less than their full bisection bandwidth. In this context we propose to exploit the use of a network coding technique to reduce the number of unicasts and the size of transferred data generated by latency-sensitive collective communication in supercomputers. Our proposed network coding scheme has a hierarchical multicasting structure with intra-group and inter-group unicasts. Quantitative analysis show that the aggregate path hop counts by our hierarchical network coding decrease as much as 94% when compared to conventional unicast-based multicasts. We validate these results by cycle-accurate network simulations. In 1,024-switch networks, the network reduces the execution time of collective communication as much as 64%. We also show that our hierarchical network coding is beneficial for any packet size.

UR - http://www.scopus.com/inward/record.url?scp=84899432084&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84899432084&partnerID=8YFLogxK

U2 - 10.1109/PDP.2014.58

DO - 10.1109/PDP.2014.58

M3 - Paper

AN - SCOPUS:84899432084

SP - 98

EP - 102

ER -