有效率地支援影音串流服務的內容感知需求分配之叢集式網頁伺服器

An Efficient Web Cluster with Content-Aware Request Distribution for Streaming Service

林政緯(Cheng-Wei Lin)

指導教授 ： 姜美玲

國立暨南國際大學/管理學院/資訊管理學系/碩士(2009年)

摘要

近年來，隨著網際網路的爆炸性發展與網路寬頻的普及，加上多媒體影音技術的進步，使得網站提供的服務更加的多元化，網頁內容也不只侷限於靜態網頁及動態網頁，更包含了影音串流服務，無名小站(Wretch)和You Tube即是典型的範例。為了處理大量且多樣性的服務需求，網路伺服器負載也隨之加重，傳統的單一伺服器已經無法應付龐大的工作量。網頁叢集式系統是由一台前端伺服器與多台後端伺服器所組成，因為其具有高可靠性和高擴充性的優點，且能夠同時處理多樣性的網頁服務需求，已成為一個具有成本效益的解決方法。
由於網路伺服器在提供多媒體影音串流服務時，伺服器必須先與客戶端建立串流連線後才能開始提供服務，串流連線持續的時間會隨著影片播放時間的長度而增加，而且在提供服務的過程中，伺服器也無法將此連線轉移給其他台伺服器來接續提供服務。此種服務提供的方式，會因為不同後端伺服器提供的影片長度不同，而造成彼此之間的負載不平衡，最後造成整個叢集式系統的效能降低。為了改善此服務提供的方式，在本論文中我們提出了RTSP Handoff機制，透過此機制能夠讓叢集式系統在提供片長較長的串流影片服務時，能夠將影片邏輯上分成不同的片段，由不同的後端伺服器來接續提供服務給客戶端，但客戶端在觀賞影片的過程中不會感受到伺服器的改變。
實驗結果證明，在LVS-CAD叢集式網頁伺服器平台上搭配我們的RTSP Handoff機制，在提供多樣性的網頁服務時，能夠得到較好的效能。實驗結果顯示，當系統處於重度工作量時，使用我們的機制會比沒有使用時提升37.19%的產出；當系統處於綜合工作量時，使用我們的機制會比沒有使用時提升29.55%的產出。此外，當系統處於重度工作量時，使用我們的機制會比沒有使用時減少34.35%的平均回應時間；當系統處於綜合工作量時，使用我們的機制會比沒有使用時減少34.59%的平均回應時間。

關鍵字

叢集式網頁伺服器 ； RTSP Handoff ； 多媒體串流 ； 內容感知需求分配

並列摘要

In recent years, due to the explosive growth of the internet, the growing popularity of broadband network, and the advance of the multimedia technology, the web sites usually can provide not only static and dynamic web page services but also multimedia streaming services. The Wretch and You Tube are the representatives. For dealing with the huge amount of requests from clients, the traditional single server is not sufficient to handle heavy workload. The web cluster system that is composed of a front-end server and multiple real servers has become a cost-effective way to serve the huge amount and various types of service demands because of the advantages of high availability and scalability.
For providing the multimedia streaming service, the real server needs to establish a streaming connection with the client at first and then transmits the streaming data. The real servers need to maintain the connection until the entire streaming service is finished. In the process of providing streaming service, the real servers can not handoff the existing connection to other real server to continue the streaming service. However, this might cause load unbalance and degrade the performance of the whole web cluster. In order to achieve better load balance and load sharing in the web cluster system when providing the streaming service, we have proposed and implemented the RTSP Handoff mechanism. By our proposed mechanism, the long runtime film can be logically partitioned into several sections and each section of the film can be served by different real server. During the process, the client would not sense the change of servers.
Experimental results demonstrate that the LVS-CAD web cluster with our proposed RTSP Handoff mechanism can achieve better performance when providing multiple web services. Precisely, the LVS-CAD web cluster with our proposed RTSP Handoff mechanism can achieve 37.19% better throughput under the heavy streaming workload, and 29.55% under the synthetic workload than the one without our proposed mechanism. Besides, the average response time of the LVS-CAD web cluster when using our proposed RTSP Handoff mechanism can be reduced by 34.35% under the heavy streaming workload, and 34.59% under the synthetic workload than the one without our proposed mechanism.

並列關鍵字

Web Cluster ； RTSP Handoff ； Multimedia Streaming ； Content-aware Request Distribution

參考文獻

[1] mp4ra.org - MP4 Registration authority, “References, MPEG-4 Registration authority”, http://www.mp4ra.org/specs.html. Retrieved on 2009-06-14.

[2] ISO (2006-04), ISO Base Media File Format white paper, http://www.chiariglione.org/mpeg/technologies/mp04-ff/index.htm.

[3] Apple Computer, “MPEG-4 Fact Sheet”, http://images.apple.com/quicktime/pdf/MPEG4_v3.pdf.

[4] Ho-Han Liu, Mei-Ling Chiang, and Men-Chao Wu, “Efficient Support for Content-Aware Request Distribution and Persistent Connection in Web Clusters,” Software Practice & Experience, Volume 37, Issue 11, Pages 1215-1241, Sep. 2007.

[5] Mei-Ling Chiang and Chun-Hung Wu, “A Kernel Mechanism for Efficiently Supporting Quality of Service in Web Cluster System”, Master Thesis, Department of Information Management, National Chi-Nan University, 2008.