針對網頁伺服器叢集之基於伺服器狀態的主動式負載平衡

Abstract

在本論文中，我們針對網頁伺服器叢集提出一套負載平衡的方法，為『基於伺服器狀態的主動式負載平衡』，這個方法計算每一個Request在網頁伺服器上所產生的負載值，進而加總計算網頁伺服器的整體負載值。網頁伺服器的負載值分為數個等級，當網頁伺服器的負載值向上或向下跳躍一個等級時，網頁伺服器將主動地向Dispatcher回報目前的負載等級。最後，Dispatcher依照每一網頁伺服器的負載等級分配Request。 對於『基於伺服器狀態的主動式負載平衡』，我們在Dispatcher上設計一個新的排程方法『Remnant and Weight』，此一方法之目的在於促使每一網頁伺服器的負載相同，俾能減少Request的反應時間。實驗結果顯示我們提出的方法的確可以達到網頁伺服器的負載平衡，並減少Request的反應時間。

In this thesis, we propose a method of load balancing, called Active Load Balancing Based on Server States, for web server clusters. This method estimates the load of each request on the web server, and then sums up the total load of all requests for each web server. The total load of a web server is classified into several levels. The web server reports information about its current load to the dispatcher actively when the load level is changed. Finally, the dispatcher forwards the requests to appropriate web servers of the cluster according to the load levels of the web servers. A new scheduling method, called Remnant and Weight, is used in the dispatcher for our load balancing method. We attempt to make the load level of each web server equal so that the response time of requests may be reduced. The simulation results show that we achieve load balancing among web servers and the response time of requests is reduced.