Leveraging torus topology with deadlock recovery for cost-efficient on-chip network

Minjeong Shin, John Kim
{"title":"Leveraging torus topology with deadlock recovery for cost-efficient on-chip network","authors":"Minjeong Shin, John Kim","doi":"10.1109/ICCD.2011.6081371","DOIUrl":null,"url":null,"abstract":"On-chip networks are becoming more important as the number of on-chip components continue to increase. 2D mesh topology is a commonly assumed topology for on-chip networks but in this work, we make the argument that 2D torus can provide a more cost-efficient on-chip network since the on-chip network datapath is reduced by 2× while providing the same bisection bandwidth as a mesh network. Our results show that 2D torus can achieve an improvement of up to 1.9× over a 2D mesh in performance per watt metric. However, routing deadlock can occur in a torus network with the wrap-around channel and requires additional virtual channels for deadlock avoidance. In this work, we propose deadlock recovery with tokens (DRT) in on-chip networks that exploits on-chip networks - exploiting the abundant wires available while minimizing the need for additional buffers. As a result, deadlocks can be exactly detected without having to rely on a timeout mechanism and when needed, recover from the deadlock. We show how DRT results in minimal loss in performance, compared with deadlock avoidance using virtual channels, while reducing the on-chip network complexity.","PeriodicalId":354015,"journal":{"name":"2011 IEEE 29th International Conference on Computer Design (ICCD)","volume":"238 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2011-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"13","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2011 IEEE 29th International Conference on Computer Design (ICCD)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICCD.2011.6081371","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 13

Abstract

On-chip networks are becoming more important as the number of on-chip components continue to increase. 2D mesh topology is a commonly assumed topology for on-chip networks but in this work, we make the argument that 2D torus can provide a more cost-efficient on-chip network since the on-chip network datapath is reduced by 2× while providing the same bisection bandwidth as a mesh network. Our results show that 2D torus can achieve an improvement of up to 1.9× over a 2D mesh in performance per watt metric. However, routing deadlock can occur in a torus network with the wrap-around channel and requires additional virtual channels for deadlock avoidance. In this work, we propose deadlock recovery with tokens (DRT) in on-chip networks that exploits on-chip networks - exploiting the abundant wires available while minimizing the need for additional buffers. As a result, deadlocks can be exactly detected without having to rely on a timeout mechanism and when needed, recover from the deadlock. We show how DRT results in minimal loss in performance, compared with deadlock avoidance using virtual channels, while reducing the on-chip network complexity.
利用环面拓扑与死锁恢复为成本效益的片上网络
随着片上元件数量的不断增加,片上网络变得越来越重要。二维网格拓扑是片上网络通常假设的拓扑结构,但在这项工作中,我们认为二维环面可以提供更具成本效益的片上网络,因为片上网络数据路径减少了2倍,同时提供与网格网络相同的等分带宽。我们的研究结果表明,二维环面在每瓦性能方面可以比二维网格提高1.9倍。但是,路由死锁可能发生在具有环绕通道的环面网络中,并且需要额外的虚拟通道来避免死锁。在这项工作中,我们提出在利用片上网络的片上网络中使用令牌(DRT)进行死锁恢复-利用丰富的可用线路,同时最大限度地减少对额外缓冲区的需求。因此,可以准确地检测死锁,而不必依赖于超时机制,并在需要时从死锁中恢复。与使用虚拟通道避免死锁相比,我们展示了DRT如何在降低片上网络复杂性的同时,将性能损失降到最低。
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 求助全文
来源期刊
自引率
0.00%
发文量
0
×
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
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
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 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信