多伺服器環境下以記憶體重複使用為考量的使用者分配之研究

User Distributions in multiple-server platform with the consideration of User balancing and memory reusability.

林明德(Ming-Te Lin)

指導教授 ： 許秉瑜

國立中央大學/管理學院/企業管理學系/碩士(2012年)

摘要

由於網際網路的蓬勃發展,使用網路服務的人口迅速地增加,因此必須使用 更多的伺服器(Application Server)同時提供服務,來滿足大量的服務需求。但 相對的,同時使用多台伺服器若不能有效地提供負載平衡(load balancing),會 造成有些伺服器過載或者閒置,因而導致服務品質的不佳。系統中伺服器個數的 多寡將是資源配置的主要考慮因素,因此,該如何在資源有限的情況下發揮最大 的經濟效益,一直都是企業營運的目標之一。
本研究目的是以 LAPO(Load Adjusted Profile Oriented User Allocation Algorithm)演算法改善 POCA(Profile Oriented Clustering Algorithm)演算法在求 取最佳組合解時,必須耗費大量的計算時間,以及能對使用者在 Application server 分配上更均勻,確實達到 load balancing 之目的。由於 LAPO 在有設置條件的情況 下,能有效的掌控 Application server 所分配的使用者個數,因此能減少計算最佳 組合解的次數,有效地增進執行效率。實驗數據來自 SAP ERP ECC 6.0 系統,並 且在實際驗證過後,確定 LAPO 的效率較好、結果更符合企業運用的實際情境。

關鍵字

負載平衡 ； ERP 作業選項 ； 效能評估

並列摘要

Due to the rapid growth of internet in popularity and size, internet traffics have increased significantly. In order to fulfill the vast web service demands, more and more application servers need to be hosted simultaneously. However, without utilizing load balance effectively, the use of multiple sever hosting will not only reduce the quality of services but also decrease the utilization of system. Therefore, how to allocate limited resources in order to optimize the performance of the system becomes a major concern for companies.
The first objective of this paper is to clarify the characteristics of each end user in the system according to their historical logging data, such as user’s transaction data and the system resources consume by this transaction. We used SAP system as our research platform and the data sets were extracted from National Central University’s SAP systems. Detail of the framework and the procedure of this research are illustrated in Chapter 3.
The second objective of this paper is to improve the POCA（Profile Oriented Clustering Algorithm）by applying LAPO（Load Adjusted Profile Oriented User Allocation Algorithm）. LAPO can monitor and control the number of users in each application servers more effectively than POCA which enable LAPO to reduce huge amount of calculation time. As such, LAPO make distribution of users much more evenly and efficiency which also can help multiple server systems to achieve better load balancing.

並列關鍵字

Performance evaluation ； Transaction ； Load balancing

參考文獻

1. SAP, TADM10_1 : SAP NetWeaver AS Implementation & Operation 1, 2008.

2. Saeed Sharifian, Seyed A. Motamedi, Mohammad K. Akbari, “A content-based load balancing algorithm with admission control for cluster web servers”, Future Generation Computer Systems, 24, pp.775–787, 2008.

3. SAP, TADM10_2 : SAP NetWeaver AS Implementation & Operation 1, 2008.

4. P. Krueger and R. Chawla,“The Stealth Distributed Schedular”, Procedding 11th Int’l Conference Distributed Computing Systems, IEEE CS Press, Los Alamitos, Calif., Order No. 2144, pp. 336-343, May 1991.

5. M. Livny and M. Melman, “Load Balancing in Homogeneous Broadcast Distributed Systems”, Computer Network Performance Symposium, pp. 336-343, May 1991.