Interconnection Networks in Petascale Computer Systems

Author:

Trobec Roman1,Vasiljević Radivoje2,Tomašević Milo2,Milutinović Veljko3,Beivide Ramon4,Valero Mateo5

Affiliation:

1. Jožef Stefan Institute, Ljubljana, Slovenia

2. University of Belgrade, Serbia

3. Maxeler Technologies and MISANU Belgrade, Serbia

4. University of Cantabria, Spain

5. Barcelona Supercomputing Center, Spain

Abstract

This article provides background information about interconnection networks, an analysis of previous developments, and an overview of the state of the art. The main contribution of this article is to highlight the importance of the interpolation and extrapolation of technological changes and physical constraints in order to predict the optimum future interconnection network. The technological changes are related to three of the most important attributes of interconnection networks: topology, routing, and flow-control algorithms. On the other hand, the physical constraints, that is, port counts, number of communication nodes, and communication speed, determine the realistic properties of the network. We present the state-of-the-art technology for the most commonly used interconnection networks and some background related to often-used network topologies. The interconnection networks of the best-performing petascale parallel computers from past and present Top500 lists are analyzed. The lessons learned from this analysis indicate that computer networks need better performance in future exascale computers. Such an approach leads to the conclusion that a high-radix topology with optical connections for longer links is set to become the optimum interconnect for a number of relevant application domains.

Funder

Slovenian Research Agency

Publisher

Association for Computing Machinery (ACM)

Subject

General Computer Science,Theoretical Computer Science

Cited by 23 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3