隨著網路快速地發展許多新的網路服務單靠一台高效能的機器是不夠的,網頁伺服器的叢集系統常用來解決這樣的問題。 所謂的動態內容網頁是指使用者每一次瀏覽所得到的網頁內容都會重新產生一次,可以依使用者的使用狀態不同而生成不同內容的網頁。 在這篇論文中,對網頁負載平衡系統作了一系列的實驗,最主要的是針對在突波式的網頁服務之下,動態網頁對整個負載平衡系統的影響。在其中,包含了四種不同型態的架構,加入了由連接資料庫產生的動態網頁的測試與分析。負載平衡系統若是採用異質的伺服器,由於硬體等級不同,勢必要對客戶端的請求給予不同的權重值,因此在實驗中,首先針對不同等級的單一伺服器作測試,將所得到的結果作為後續實驗中不同等級伺服器的負載平衡系統的權重值。在訂好權重值後,實際對四種負載平衡架構作測試,其中包含了LVS via direct routing, LVS via NAT, LVS via IP tunneling以及另外一種利用HTTP重導向的分派架構。 經由這一連串的實驗所得到的結果將有助於分析動態網頁對整個負載平衡系統的影響與產生的問題,可以依據這樣的結果,在未來發展出更適合動態網頁的網頁伺服器負載平衡系統,也可配合不同等級的伺服器作更適當的調整,讓負載平衡系統的運作更加的完善。
As the Internet grows rapidly, it has also created the need for more and faster web servers capable of serving millions of users. As a matter of fact, services to large number of users can not be able to be handled by a single server. Thus, cluster of servers is used increasingly to build high-capacity web site. In this thesis, a series of experiments with several load balancing methods are performed. There are many configurations to construct an Internet cluster system. The configurations include dispatcher-based approach, parallel filtering approach, and round-robin Domain Name Server (RR-DNS) approach. Many papers have proposed methods to make the load balancing system more efficient. In this thesis,we focus on the effect of dynamic content on a load balancing system. We setup a burst mode service request environment; that is, a considerable amount of requests which are generated in a short period of time. Weighted round robin with weight according to the performance of the servers is used to make heterogeneous servers to cooperate together effectively. We test various cluster configurations with different loading conditions. The advantage and shortage of each configuration are also presented. These experimental studies help us to find out the problems and the cause of the problems in the load balancing architectures with dynamic content web service. With these findings, we can improve the performance of load balancing architectures for the burst mode service requests and for heterogeneous servers in the load balancing system.