Deadline-aware load balancing for coflow in datacenter networks

IF 4.6 2区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Computer Networks Pub Date : 2025-10-10 DOI:10.1016/j.comnet.2025.111765

Zhi Gong , Zhichen Wang , Jinbin Hu , Jin Wang , Fayez Alqahtani , Amr Tolba , Kai Jin

{"title":"Deadline-aware load balancing for coflow in datacenter networks","authors":"Zhi Gong , Zhichen Wang , Jinbin Hu , Jin Wang , Fayez Alqahtani , Amr Tolba , Kai Jin","doi":"10.1016/j.comnet.2025.111765","DOIUrl":null,"url":null,"abstract":"<div><div>To accurately capture the characteristics of collective data transmission in modern distributed data parallel applications, the abstraction of Coflow is introduced. Notably, a significant number of Coflows in time-sensitive applications have strict deadlines, which are critical for meeting the application’s timeliness and performance requirements. However, existing Coflow transmission mechanisms fail to fine-grain the classification of Coflows, and current load balancing strategies do not holistically consider the transmission characteristics of Coflows. This inevitably increases Coflow completion time (CCT) and raises the proportion of Coflows missing their deadlines. To address this issue, we propose DLB. Specifically, DLB classifies Coflows into three transmission levels. Coflows with deadlines that can be completed within their respective deadlines are categorized as the first transmission level. Furthermore, DLB prioritizes routing paths based on the queue length at routing ports and the queue growth rate of transmission paths, ensuring that first-level Coflows are directed to the highest-priority routing ports. The evaluation results show that DLB reduces the miss rate up to 33%, 28%, 23%, 15% and 6% over ECMP, LetFlow, Varys, Con-Myopic and JFSR at a load of 0.7 under Web Search workload, respectively.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"273 ","pages":"Article 111765"},"PeriodicalIF":4.6000,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computer Networks","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1389128625007315","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE","Score":null,"Total":0}

引用次数: 0

Abstract

To accurately capture the characteristics of collective data transmission in modern distributed data parallel applications, the abstraction of Coflow is introduced. Notably, a significant number of Coflows in time-sensitive applications have strict deadlines, which are critical for meeting the application’s timeliness and performance requirements. However, existing Coflow transmission mechanisms fail to fine-grain the classification of Coflows, and current load balancing strategies do not holistically consider the transmission characteristics of Coflows. This inevitably increases Coflow completion time (CCT) and raises the proportion of Coflows missing their deadlines. To address this issue, we propose DLB. Specifically, DLB classifies Coflows into three transmission levels. Coflows with deadlines that can be completed within their respective deadlines are categorized as the first transmission level. Furthermore, DLB prioritizes routing paths based on the queue length at routing ports and the queue growth rate of transmission paths, ensuring that first-level Coflows are directed to the highest-priority routing ports. The evaluation results show that DLB reduces the miss rate up to 33%, 28%, 23%, 15% and 6% over ECMP, LetFlow, Varys, Con-Myopic and JFSR at a load of 0.7 under Web Search workload, respectively.

查看原文本刊更多论文

数据中心网络中协同流的截止日期感知负载平衡

为了准确地捕捉现代分布式数据并行应用中集体数据传输的特点，引入了Coflow抽象。值得注意的是，时间敏感型应用程序中的大量coflow具有严格的截止日期，这对于满足应用程序的及时性和性能需求至关重要。然而，现有的Coflow传输机制无法对Coflow进行精细分类，当前的负载均衡策略也没有从整体上考虑Coflow的传输特性。这不可避免地增加了Coflow完成时间（CCT），并增加了Coflow错过截止日期的比例。为了解决这个问题，我们提出了DLB。具体来说，DLB将coflow分为三个传输级别。具有可在其各自截止日期内完成的截止日期的coflow被归类为第一传输级别。此外，DLB根据路由端口的队列长度和传输路径的队列增长率对路由路径进行优先级排序，确保第一级coflow被定向到优先级最高的路由端口。评价结果表明，在Web搜索负载为0.7时，DLB比ECMP、LetFlow、Varys、Con-Myopic和JFSR分别降低了33%、28%、23%、15%和6%的失分率。

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

Computer Networks 工程技术-电信学

CiteScore

10.80

自引率

3.60%

发文量

434

审稿时长

8.6 months

期刊介绍： Computer Networks is an international, archival journal providing a publication vehicle for complete coverage of all topics of interest to those involved in the computer communications networking area. The audience includes researchers, managers and operators of networks as well as designers and implementors. The Editorial Board will consider any material for publication that is of interest to those groups.