標題: 叢集式網路伺服器分派架構之研究
A Dispatching Architecture with Hash Mechanism for Cluster-based Network Servers
作者: 鄭傑誠
Chieh-Cheng Cheng
簡榮宏
Rong-Hong Jan
資訊科學與工程研究所
關鍵字: 叢集式;雜湊;等待時間;負載配置;cluster-based;Hash;response time;load allocation mechanisms
公開日期: 2001
摘要: 在叢集式網路伺服器系統中,分配的機制扮演著極重要的角色。傳統上一個好的分配機制必須滿足下列基本條件:較低的運算量(low overhead)、伺服器負載平衡(load balancing)和較高的命中率(high hit rate)。在本篇論文中,我們利用雜湊(hash)機制提出兩個有效的負載配置方法,第一個方法稱之為雜湊再雜湊(Hash-Rehash, HRH)。主要是將每個使用者的需求經由雜湊函式計算以決定要傳送的伺服器。當被決定的伺服器發生故障時,我們則再運用雜湊函式重新決定另一台伺服器。第二個方法稱之為連線個數輔助的雜湊再雜湊分配機制(HRHCNA)。跟HRH不同的地方在於HRHCNA加上了另一個判斷機制;即利用伺服器的連線個數判斷是否要把需求重新再雜湊一次,以決定另一台伺服器。HRH優點在於方法簡單且可以應用於使用者端。HRHCNA則是利用伺服器的連線狀態,藉此讓伺服器負載能更加平衡。實驗結果顯示,這兩種方法都能有效的降低使用者的等待時間。
The load allocation mechanisms are very important in the cluster-based network servers. A good allocation method should have the following characteristics: low overhead, load balancing, and high hit rate. In this thesis, we will propose two load allocation mechanisms that apply hash method. The first load allocation mechanism is called as Hash-Rehash (HRH) scheme. The other load allocation mechanism is called as HRH with connection number aided dispatcher scheme (HRHCNA). Both HRH and HRHCNA can accomplish the load location with lower overhead and higher hit rate. The simulation results show that the proposed allocation mechanisms can reduce the response time, effectively.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT900394028
http://hdl.handle.net/11536/68551
Appears in Collections:Thesis