在網路邊端設備上為網頁服務品質進行之需求排程

Abstract

在網路邊端的環境裡，因頻寬的限制，使得當瀏覽網頁人數眾多時，容易造成塞車而延緩了瀏覽的時間，所以必須對網路頻寬做一些管制措施，以改善使用者感覺的延遲時間。然而，目前一般的頻寬管理器通常僅能針對IP或者應用程式類別限制，無法進一步對應用程式的內容（例如：網頁的URL、內容型態）做分類配置頻寬，提供更佳的服務品質。為了管理通過邊端設備的網頁流量及提供更有彈性、與內容有關的分類配置策略，我們提供了可在消費者端網路設備設置，不需要更動任何伺服器或使用者端軟體的端點對端點網頁服務品質架構。為了達到內容分類的目的，我們提出了針對網頁需求分類的機制。此外，我們更提出了一個網頁需求排程的演算法，根據另一方向網頁回應內容頻寬的使用情況來決定下一個該釋放哪一個網頁需求及其釋放時間，以達到有效利用頻寬的目的。實驗結果證實我們的方法確實能有效限制不同分類的頻寬，並能準確填補頻寬空隙，大量縮短了瀏覽網頁的時間。

In the network edge environment, network congestion occurs frequently when there are too many users browsing Web pages through a bandwidth-limited link. Therefore, the Web traffics passing through an edge device should be managed so as to improve user-perceived latency. However, traditional bandwidth management, i.e. network quality of service, is based on the policies of class 5-tuples, which is content-blind. That is, the administrator cannot define classification policies according to URL, HTTP header fields, and HTTP payload. In order to manage the Web traffics passing through an edge device and provide more-flexible, content-aware classification policies, we present an end-to-end Web QoS framework at consumer-side edge devices without modifying any client or server software. An HTTP request classification mechanism is presented to meet to requirements of content-aware classification. In addition, the presented request scheduling algorithm decides which queued HTTP request to release to the destination Web server next and when to release it so as to control the bandwidth utilization by the HTTP responses in the opposite direction. The evaluation results show that the request scheduling algorithm achieves good Web service differentiation and accurate release time estimation. Our system can manage Web traffics well and improve user-perceived latency.