應用於異質性Web叢集系統之Threshold Client–Aware需求分配策略的設計與製作

Abstract

滿足全球化新資訊紀元動輒數萬的服務連線需求，以具成本優勢的叢集系統取代高速主機成為近年來的趨勢，因此，如何提升叢集系統的效能成為重要的研究課題。以極為普及的Web Clusters 為例，藉由實作負載共享機制讓每個叢集節點資源得以充份利用，進而提昇整體系統的效能，需求分配服務策略是扮演關鍵的角色。
為因應Web Clusters後端伺服主機性能不一的情況，所導致的服務效能不均及資源浪費的問題，本論文的目的是提出一套負載共享的方法，來解決異質性叢集系統之服務分配不均的問題，分析每個服務要求連線的類型，依類型來計算每台後端伺服主機的服務能力，當超過所設定的臨界值（threshold）時，即超過該台後端伺服主機的工作負載能力時，就暫時凍結該台後端伺服主機服務的權利，讓其它工作負載較輕的後端伺服主機來提供服務。實驗結果顯示透過適性排程演算法的有效負載共享，使異質性叢集系統之整理效能大大地提昇。Since cluster systems have the advantage of cost superiority than supercomputers, applying cluster systems has been a trend in recent years to meet the requirement of large amount of requests in globalized new information era. Therefore, how to promote performance of cluster systems becomes an important research topic. Scheduling policy plays a key role to adjust request distribution among servers in the cluster system. By using effective load sharing mechanisms, every server node's resources in the cluster system can be sufficiently used, and thus to promote the performance of entire system.
The aim of this paper is to propose an effective load sharing method to make the most use of system resource and to handle service dispatching well for the difference of server node's hardware equipments. In this method, by using threshold control, no server node will be overloaded while the other nodes are idle. As a result, the entire system resource can be effectively shared and thus to promote entire system performance.