Static versus dynamic content

In this test, WebBench is used. The enlarged workload used in Sections 6.3 and 6.4 consists of complete static Web pages. To examine the impact of dynamic Web pages on performance, the static Web pages have been mixed with dynamic Web pages written with PHP in the ratio of 90%:10%

(denoted as 90s10d), 80%:20% (denoted as 80s20d), 70%:30% (denoted as 70s30d), and 60%:40%

(denoted as 60s40d). Eight back-ends are used in this test.

As shown in Figure 19, LVS-CAD with CAHRD gives the best performance. Compared with LVS/WLC, LVS-CAD with CAHRD gives a performance speedup of 13.26–36.72%, LVS-CAD with xLARD/R gives a performance speedup of 2.83–24.16%, and LVS-CAD with LARD/R gives a performance speedup of 0.94–17.24%.

The 70%:30% ratio is the critical point for xLARD/R and CAWLL. Before this point xLARD/R gives a better performance than CAWLL, and after this point CAWLL performs better than xLARD/R.

This also indicates that CAWLL is more suitable for a workload with dynamic Web pages.

0

Different localities of access Different localities of access

Different localities of access Different localities of access

Throughput (req s-1)Throughput (req s-1) Throughput (req s-1)Throughput (req s-1)

Figure 18. Performance of different localities of hot Web pages for the LVS versus LVS-CAD:

(a) 10%; (b) 20%; (c) 30%; (d) 40%.

By combining the advantages of xLARD/R and CAWLL, CAHRD gives the best performance of these request scheduling algorithms.

7. CONCLUSIONS

We have designed and implemented a CAD platform, LVS-CAD, to build efficient and scalable Web clusters. LVS-CAD can extract and analyze the content in requests, then dispatch each request by its content or type of service requested. By applying the proposed connection transfer mechanism, TCP Rebuilding, the connection between the dispatching server and the request-handling server can be transferred efficiently. By applying the proposed content-aware request distribution policies to reduce the frequent connection handoff and increase the cache hit ratio of Web servers, request distribution among Web servers is more balanced and the system resources of the overall Web cluster are thus utilized more efficiently. Together with the implementation of the fast TCP module handshaking and

1400 1600 1800 2000 2200 2400 2600

90s10d 80s20d 70s30d 60s40d

WLC (LVS) CAWLL LARD/R xLARD/R CAHRD Throughput (req s-1)

Different mixed ratio of Web pages

Figure 19. Performance of mixed ratio of dynamic and static contents for the LVS versus LVS-CAD.

multiple TCP Rebuilding, LVS-CAD provides efficient support for HTTP/1.1 persistent connection in Web clusters with content-based request distribution.

Performance results indicate that LVS-CAD outperforms the content-blind dispatching platform, the LVS. By applying TCP Rebuilding and fast TCP module handshaking only, LVS-CAD can improve on the performance of the LVS by 32.4% and is more scalable as the number of back-ends is increased.

Together with the employment of the proposed content-aware request distribution policies, LVS-CAD with xLARD/R performs 58.7–156.49% better than LVS/WLC for scalability test and 14.37–64.28%

better for different localities of hot Web pages. LVS-CAD with CAHRD outperforms LVS/WLC by 13.26–36.72% for a mixed ratio of dynamic and static Web pages.

Based on this platform, several issues can be explored or enhanced in future research, such as the cooperation of cache management among back-end servers, session persistence handling, providing differentiated service, quality of service support, special Web content deployment, efficient methods for analyzing the content of requests, etc. In addition, pre-establishing TCP connections between the front-end and back-ends would be helpful for reducing the TCP connection transfer time.

REFERENCES

1. Andreolini M, Colajanni M, Nuccio M. Kernel-based Web switches providing content-aware routing. Proceedings of the 2nd IEEE International Symposium on Network Computing and Applications, Cambridge, MA, 16–18 April 2003.

IEEE Computer Society Press: Los Alamitos, CA, 2003; 25–32.

2. Andreolini M, Colajanni M, Nuccio M. Scalability of content-aware server switches for cluster-based Web information systems. Proceedings of the 12th International World Wide Web Conference, Budapest, Hungary, 20–24 May 2003.

The International World Wide Web Conference Committee (IW3C2), 2003.

3. Aron M, Druschel P, Zwaenepoel W. Efficient support for P-HTTP in cluster-based Web servers. Proceedings of the Annual Usenix Technical Conference, Monterey, CA, June 1999. USENIX Association: Berkeley, CA, 1999.

4. Aron M, Sanders D, Druschel P, Zwaenepoel W. Scalable content-aware request distribution in cluster-based network servers. Proceedings of the Annual Usenix Technical Conference, San Diego, CA, 18–23 June 2000. USENIX Association:

Berkeley, CA, 2000.

5. Cardellini V, Casalicchio E, Colajanni M, Yu PS. The state of the art in locally distributed Web-server systems.

ACM Computing Surveys 2002; 34(2):263–311.

6. Casalicchio E, Colajanni M. A client-aware dispatching algorithm for Web clusters providing multiple services.

Proceedings of the 10th International World Wide Web Conference, Hong Kong, 1–5 May 2001. The International World Wide Web Conference Committee (IW3C2), 2001; 535–544.

7. Cherkasova L, Karlsson M. Scalable Web server cluster design with workload-aware request distribution strategy WARD.

Proceedings of the 3rd International Workshop on Advanced Issues of E-Commerce and Web-Based Information Systems, San Jose, CA, 21–22 June 2001. IEEE Computer Society Press: Los Alamitos, CA, 2001; 212–221.

8. Pail VS, Aront M, Bangat G, Svendsent M, Druschelt P, Zwaenepoelt W, Nahumq E. Locality-aware request distribution in cluster-based network servers. Proceedings of the 8th International Conference on Architectural Support for Programming Languages and Operating Systems, San Jose, CA, October 1998. ACM Press: New York, 1998; 205–216.

9. Sit Y-F, Wang C-L, Lau F. Cyclone: A high-performance cluster-based Web server with socket cloning. Cluster Computing 2004; 7(1):21–37.

10. Wang L. TCPHA Project Website, 28 April 2004. http://dragon.linux-vs.org/∼dragonfly/htm/tcpha.htm.

11. Yang C-S, Luo M-Y. Efficient support for content-based routing in Web server clusters. Proceedings of the 2nd USENIX Symposium on Internet Technologies and Systems, Boulder, CO, 11–14 October 1999. USENIX Association: Berkeley, CA, 1999.

12. Yang C-S, Luo M-Y. System support for scalable, reliable, and highly manageable Web hosting service. Proceedings of the 3rd USENIX Symposium on Internet Technologies and Systems, San Francisco, CA, 26–28 March 2001. USENIX Association: Berkeley, CA, 2001.

13. Zhang X, Barrientos M, Chen JB, Seltzer M. HACC: An architecture for cluster-based Web servers. Proceedings of the 3rd USENIX Windows NT Symposium, Seattle, WA, 12–13 July 1999. USENIX Association: Berkeley, CA, 1999.

14. Lin Y-D, Tsai P-T, Lin P-C, Tien C-M. Direct Web switch routing with state migration, TCP masquerade, and cookie name rewriting. Proceedings of the Global Telecommunications Conference 2003 (GLOBECOM ’03), San Francisco, CA, 1–5 December 2003, vol. 7. IEEE Computer Society Press: Los Alamitos, CA, 2003; 3663–3667.

15. Maliz D, Bhagwat P. TCP splicing for application layer proxy performance. Technical Report RC-21139, IBM, March 1998.

16. Steven C, Krawczyk JJ, Nair RK, Royce K, Siegel KP, Stevens RC, Wasson S. Method and system for directing a flow between a client and a server. US Patent 6,006,264, 21 December 1999.

17. Liu H-H, Chiang M-L. TCP Rebuilding for content-aware request dispatching in Web clusters. Journal of Internet Technology 2005; 6(2):231–240.

18. Linux Virtual Server Website. http://www.linuxvirtualserver.org/ [18 October 2006].

19. Zhang W, Jin S, Wu Q. Creating Linux virtual servers. Proceedings of the 1999 LinuxExpo Conference, Raleigh, NC, 18–23 May 1999. Available at: http://www.linuxvirtualserver.org/Documents.html.

20. Zhang W. Linux virtual servers for scalable network services. Proceedings of the OTTAWA Linux Symposium, Ottawa, Canada, 19–22 July 2000.

21. Fielding R, Gettys J, Mogul J, Frystyk H, Masinter L, Leach P, Berners-Lee T. Hypertext Transfer Protocol—HTTP/1.1.

Request for Comments 2616, 9 July 2004. Available at:

http://www.w3.org/Protocols/rfc2616/rfc2616.html [1 September 2004].

22. LVS Deployment, 9 July 2004. http://www.linuxvirtualserver.org/deployment.html [2 August 2006].

23. Red Hat. http://www.redhat.com/ [3 October 2006].

24. Mack J. LVS-HOWTO, July 2003.

http://www.austintek.com/LVS/LVS-HOWTO/ [25 February 2006].

25. Postel J. Transmission Control Protocol, Request for Comments 793, September 1981.

26. Kokku R, Rajamony R, Alvisi L, Vin H. Half-pipe anchoring: An efficient technique for multiple connection handoff.

Proceedings of the 10th International Conference on Network Protocols (ICNP 2002), Paris, France, November 2002;

12–21.

27. WebBench Web site. http://www.etestinglabs.com/benchmarks/webbench/webbench.asp [8 February 2005].

28. Apache. http://www.apache.org/ [3 October 2006].