On the impact of message packetization in networks of workstations with irregular topology

Xavier Molero, F. Silla, V. Santonja, J. Duato
{"title":"On the impact of message packetization in networks of workstations with irregular topology","authors":"Xavier Molero, F. Silla, V. Santonja, J. Duato","doi":"10.1109/EMPDP.2001.904960","DOIUrl":null,"url":null,"abstract":"Networks of workstations (NOWs) are becoming an increasingly popular alternative to parallel computers for those applications with high needs of resources such as memory capacity and input/output storage space, and also for small scale parallel computing. Usually, the software messaging layers in these systems become a bottleneck due to the overhead they introduce. Some proposals like FM and BIP considerably reduce this overhead by splitting long messages into several packets. These proposals have been shown to improve communication performance. However, the effect of message packetization on the network interconnects has not been analyzed yet. In this paper we examine the effect of message packetization from the point of view of the interconnection network in the context of bimodal traffic. Two different routing algorithms have been considered: up*/down* and minimal adaptive routing. Our study shows that when the up */down* routing algorithm is used, message packetization dramatically increases latency and reduces throughput for both long and short messages. On the other hand, if minimal adaptive routing is used, short messages could benefit from message packetization, but at the cost of increasing latency for long messages. In any case, network throughput is considerably reduced.","PeriodicalId":262971,"journal":{"name":"Proceedings Ninth Euromicro Workshop on Parallel and Distributed Processing","volume":"12 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2001-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings Ninth Euromicro Workshop on Parallel and Distributed Processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/EMPDP.2001.904960","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

Abstract

Networks of workstations (NOWs) are becoming an increasingly popular alternative to parallel computers for those applications with high needs of resources such as memory capacity and input/output storage space, and also for small scale parallel computing. Usually, the software messaging layers in these systems become a bottleneck due to the overhead they introduce. Some proposals like FM and BIP considerably reduce this overhead by splitting long messages into several packets. These proposals have been shown to improve communication performance. However, the effect of message packetization on the network interconnects has not been analyzed yet. In this paper we examine the effect of message packetization from the point of view of the interconnection network in the context of bimodal traffic. Two different routing algorithms have been considered: up*/down* and minimal adaptive routing. Our study shows that when the up */down* routing algorithm is used, message packetization dramatically increases latency and reduces throughput for both long and short messages. On the other hand, if minimal adaptive routing is used, short messages could benefit from message packetization, but at the cost of increasing latency for long messages. In any case, network throughput is considerably reduced.
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
不规则拓扑工作站网络中消息分组的影响
对于内存容量和输入/输出存储空间等资源需求高的应用程序以及小规模并行计算,工作站网络(NOWs)正在成为并行计算机的一种日益流行的替代方案。通常,这些系统中的软件消息传递层由于它们带来的开销而成为瓶颈。一些建议,如FM和BIP,通过将长消息分成几个数据包,大大减少了这种开销。这些建议已被证明可以改善通信性能。然而,报文分组对网络互连的影响尚未得到分析。在本文中,我们从双峰流量背景下的互连网络的角度研究了消息分组的影响。考虑了两种不同的路由算法:上/下和最小自适应路由。我们的研究表明,当使用上/下路由算法时,消息分组大大增加了延迟,并降低了长消息和短消息的吞吐量。另一方面,如果使用最小的自适应路由,短消息可以从消息分组中获益,但代价是增加长消息的延迟。在任何情况下,网络吞吐量都会大大降低。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
自引率
0.00%
发文量
0
期刊最新文献
Adding flexibility and real-time performance by adapting a single processor industrial application to a multiprocessor platform Heterogeneous matrix-matrix multiplication or partitioning a square into rectangles: NP-completeness and approximation algorithms Parallel simulated annealing for the delivery problem Implementing on-line techniques to allocate file resources in large distributed systems Visual data-parallel programming for signal processing applications
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1