臺灣博碩士論文加值系統

隨著上網人數不斷的激增，寬頻網路的普及，許多大型的內容提供者(ICP)所面臨的最大挑戰就是巨量的使用者需求。單一伺服器已不能滿足這樣龐大的需求，所以一個良好的解決方案就是伺服器叢集系統。伺服器叢集技術已發展了許多年，不過對於今日電子商務及其它的網路服務，L4負載平衡技術不能根據使用者要求的內容做負載平衡是不夠用的，甚至可能會導致嚴重的錯誤。因此能夠根據內容做交換的網頁交換器因應而生。現行的做法最大的缺點就是網頁交換器必需處理回傳給使用者的資料，這造成網頁交換器本身容易形成效能上的瓶頸。本論文針對這個問題提出一個直接路由的網頁交換器架構，並針對HTTP持續連線與session persistence的問題提出二個簡單且有效的方法，並實作於Linux核心中。最後在我們測試直接路由與非直接路由的效能中，我們的方法比現存的方法每秒能處理更多的HTTP需求。

With growing WWW users and popularity of broadband networks, many large-scaled ICPs have faced the need of dealing with huge amount of user requests. However, a single web server is limited by its capacity, so a good solution is to build a server farm. Server farm technology has been popular for years, but without looking into request content, as L4 load balancers do, it may cause redirection errors in some modern e-commerce services. A content-aware web switch can solve this problem. Nowadays, most web switches have to deal with response packets from the servers and make themselves a bottleneck easily. In this paper, we propose a direct routing architecture to build a web switch and two simple but effective methods to achieve persistent connection and session persistence. We implement them into the Linux kernel and compare both direct-routing and non-direct-routing approaches. The result shows that our solutions can deal with more HTTP requests than existing solutions.

Chapter 1 Introduction 1
Chapter 2 Architecture and difficulties in web switch design 4
2.1 The difficulties in web switch design 4
2.2 Mechanisms overview 4
2.2.1 Proxy approach 4
2.2.2 Redirect flows 5
2.2.3 TCP handoff 6
2.3 One-packet TCP state migration to packet filter 8
Chapter 3 Problems of HTTP keepalive connection 13
3.1 Problem statement 13
3.2 Some existing methods 13
3.2.1 Rewriting HTTP header 13
3.2.2 Using HTTP status code - 302 14
3.2.3 Multiple TCP connection handoff 14
3.2.4 Back-end request forwarding 14
3.3 Switch back masquerading in the packet filter 15
Chapter 4 Support server side language's session mechanism 19
4.1 Problem statement 19
4.2 Existing methods 19
4.2.1 Programmer pre-definition 20
4.2.2 Automatic cookie insertion 20
4.2.3 Cookie learning 21
4.3 Cookie name rewriting in the packet filter 21
Chapter 5 Implementation and benchmark 25
5.1 Functional description of DRWS 25
5.2 Implementation environment 26
5.3 Benchmark methodology 26
5.4 Benchmark result 28
5.5 Internal benchmark of DRWS 29
Chapter 6 Conclusion and future work 31

[1] Linux Virtual Server Project. http://www.linuxvirtualserver.org
[2] D. Maltz. TCP Splicing for Application Layer Proxy Performance. IBM Research Report RC-21139, March 1998.
[3] Colby, et al. Method and system for directing a flow between a client and a server. US patent 6,006,264 December 21, 1999.
[4] V. S. Pai, et al. Locality-Aware Request Distribution in Cluster-based Network Servers. Eighth International Conference on Architectural Support for Programming Languages and Operating Systems, San Jose, CA, Oct 1998.
[5] Mohit Aron, Peter Druschel and Willy Zwaenepoel. Efficient Support for P-HTTP in Cluster-Based Web Servers. USENIX 1999 Annual Technical Conference, Monterey, CA, June 1999.
[6] Kernel TCP Virtual Server (KTCPVS). http://www.linuxvirtualserver.org/software/ktcpvs/ktcpvs.html
[7] Mohit Aron, Darren Sanders, Peter Druschel and Willy Zwaenepoel. Scalable Content-aware Request Distribution in Cluster-based Network Servers USENIX 2000 Annual Technical Conference, San Diego, CA, June 2000.
[8] Scalable Content-aware Request Distribution in Cluster-based Network Servers. http://www.crpc.rice.edu/softlib/scalableRD.html
[9] R. Fielding, et al. Hypertext Transfer Protocol - HTTP/1.1. IETF Request for Comments, june1999.
[10] Netfilter. http://www.netfilter.org