在網站閘道器上提供差別服務品質之請求排程

Abstract

隨著網站流量的大幅成長，網站伺服器的負載也變得愈重，進而造成使用者感受到的等待時間也愈長，網站經營者希望能夠採用差別服務來改善特定使用者得到的網站吞吐量以及縮短其感受到的等待時間。本文提出一個部署在網站閘道器上的HTTP請求排程演算法來提供差別服務品質，其不需要修改任何在客戶端或伺服器端的軟體。以閘道器輸出連結的觀點來看，處理請求的時間是根據其回應的大小而不是該請求本身的大小，所以我們的演算法看起來好像是非持續傳送式的¬而不像傳統的封包排程演算法大部分都是持續傳送式的。我們分別以回應的大小以及窗式控制的機制去仿效DRR來決定請求的順序以及其釋放的時間。此順序是依據請求的回應大小以及事先定義讓每個服務類別得到網站吞吐量的比例之服務比重，而釋放時間則是依據網站伺服器的處理速度。在系統評估中，當三個服務類別的服務比重被指定為6:3:1時，不論存取網頁的大小為何，此服務品質閘道器可讓它們如預期般的獲得網站總吞吐量的60%、30%及10%。並且，擁有最大權重之服務類別得到的網站吞吐量以及使用者感受到的等待時間分別最多改善為沒有差別服務品質時的176%及69%。

With the explosive growth of Web traffic, the load on a Web server become heavier, leading to the longer user-perceived latency. Website operators would like to employ service differentiation to offer better throughput and user-perceived latency for some specific users. This work presents an HTTP request scheduling algorithm deployed at the website gateway to enable the Web quality of service without modifying any client or server software. Unlike traditional packet scheduling algorithms which are mostly work-conservative, our algorithm appears to be non-work-conservative from the viewpoint of the output link of the gateway because the service time of a request depends on the size of its response, not the size of the request itself. We emulate deficit round robin and the window control mechanism to decide the order and release time of requests, respectively. The order is according to the response size of the requests and the pre-defined service weights, to which the throughput of the service classes will be proportional, whereas the release time is according to the service rate of the Web server. In the evaluation, when the weight ratio 6:3:1 is assigned to three service classes, QoS website gateway lets them get 60%, 30%, and 10% of the overall throughput as expected, regardless whatever page sizes. In addition, the throughput and the user-perceived latency of the class with the largest weight can be improved by up to 176% and 69% of the QoS-disabled values, respectively.