請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/60625
標題: | 雲端資料中心網路的壅塞觸發適應性繞送 Congestion-Triggered Adaptive Routing in Cloud Data Center Networks |
作者: | TSUNG-CHENG HOU 侯宗成 |
指導教授: | 廖婉君(Wanjiun Liao) |
關鍵字: | 適應性繞送,資料中心網路,雲端資料中心,QCN, adaptive routing,data center networks,cloud data center,QCN, |
出版年 : | 2013 |
學位: | 碩士 |
摘要: | 適應性繞送,主要針對高附載且傳輸較久的大型網路流相碰而產生的壅塞狀況,希望藉由重新安排部分大型網路流的路徑,來增強整體網路的傳輸效能、降低延遲。我們最主要的貢獻,在於精確判別那些網路流最適合被重新安排路徑,並能夠藉由這些特定網路流的適應性繞送,透過更低的成本和更少副作用,達到更好的效能。有別於其他相關的研究,我們刻意選擇大中較小的網路流作為主要處理標的,並透過模擬證實我們所提出的標定與適應性繞送方法,能提供更好的成效。
為了在短時間內快速減輕壅塞狀況,先前的相關研究選擇相爭網路流中最大流量者,並將之重新繞送到可能不壅塞的路徑上。然而,這些最大網路流多半難以找到足以容納其極大流量需求的路徑,因此導致後續的壅塞、更多連鎖觸發的適應性重繞、以及增長網路延遲。反之,我們策畫設計了一個有效的機制,能夠辨認其餘較小的網路流並將這些較小流量重新繞送,又可以確定搬移這些網路流的大小足夠減輕原來的壅塞。我們機制的所有面向皆基於現存的網路協定來設計,不增加額外的複雜度或成本。 關鍵字:適應性繞送、資料中心網路、雲端資料中心、QCN Adaptive routing in data center networks is targeted at resolving congestions of heavy-loaded and long-lasting elephant flows by trying to re-allocate part of the elephant flows to other paths. While there are several mechanisms proposed, we are able to achieve better performance in terms of higher throughput and lower delay, by minimizing side effects of adaptive routing and lower costs. Our mechanism is based on commodity available IEEE standard of Quantized Congestion Notification which guarantees responsiveness and enough inference of network states. Scalability is sustained because it is a distributed mechanism. The greatest contribution of this work is the concept of “reroute smaller flows”, which means that we are able to accurately identify the flows eligible to be rerouted and prove that by following this concept we can achieve higher system throughput by fewer reroute trials. We justify the disadvantage of reroute largest flows in previous works and the benefits of reroute smaller flows by a custom built packet-level and event-based simulator. For the sake of relieving congestion quickly in a short period of time, previous works chose to select largest flows among contending flows in a congested link to be rerouted. However, these largest flows are less possible to be accommodated by other paths since they are already too large; hence they trigger following congestion, induced reroutes, and increase delay. On the contrary, we devise an effective mechanism to identify the remaining smaller flows to be rerouted and ascertain that these flows are still large enough to be rerouted to relieve congestion. Every aspect of our mechanism is solely based on existing standard, without adding more complexities and costs. Keywords: adaptive routing, data center networks, cloud data center, QCN |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/60625 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 7.11 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。