Improving End-to-End Performance of the Web Using Server Volumes and Proxy Filters

1. Improving End-to-End Performance of the Web Using Server Volumes and Proxy Filters Edith Cohen, Balachander Krishnamurthy, and Jennifer Rexford ACM SIGCOMM Conference 1998 전산학과 CALab 황인철

2. Contents • Introduction • Exchanging Filter and Volume Information • Server Volumes • Web Proxy Applications • Conclusion • Discussion

3. Introduction(1/2) • The exponential rate of growth of the WWW • A dramatic increase in Internet traffic • A significant degradation in user-perceived latency

4. Introduction(2/2) • Previous research for these problems • Focused on improving the performance of individual components • Without considering any other components • In this paper • Focuses on end-to-end information exchange by piggybacking • Between the proxy and the server

5. Exchanging Filter and Volume Information- Piggybacking Protocol(1/3) • The server • Has the information about each resource, including size and content type as well as the frequency of modification • The proxy • Has the information about its domain of clients and their access pattern • To bridge the knowledge gap between servers and proxies • piggybacking

6. Exchanging Filter and Volume Information- Piggybacking Protocol(2/3) • Server volume • The group of related resources at server-side • Proxy filter • The method to tailor volume information

7. Exchanging Filter and Volume Information- Piggybacking Protocol(3/3)

8. Exchanging Filter and Volume Information- Proxy Filters • Problem of piggybacking protocol • If several accesses occur to the same server, server transfers the same information • The method to reduce piggybacked message using proxy filter • Proxy can control piggyback frequency • Randomized Method • RPV(Recently Piggybacked volumes) • Proxy can customize the contents of each piggyback messages

9. Exchanging Filter and Volume Information- Piggybacking in HTTP 1.1 Proxy -> Server Server -> Proxy

10. Server Volumes- Optimization Criteria • Fraction predicted • The probability that a resource requested at the proxy has appeared in at least one piggyback message in the last T seconds • True prediction fraction • The likelihood that a resource that appears in a piggyback message will be accessed by a client in the next T seconds • Update fraction • The likelihood that a resource requested at the proxy has been predicted in the last T seconds and appeared in a previous request in the last C seconds, where C>T

11. Server Volumes- Directory-Based Volumes • Volume Construction and Maintenance • Based on the heuristic • Resources in the same directory are likely to have related content • Example • One-level volume • www.foo.com/a/b.html & www.foo.com/a/d/e.html • www.foo.com/a/b.html & www.foo.com/c/k.html • Volume elements • A collection of FIFO lists partitioned by resource size and content type

12. Server Volumes- Probability-Based Volumes • Volume Construction and Maintenance • Construct volume by measuring access patterns • Resource s is included in r’s volume if Ps|r >= Pt

13. Server Volumes- Performance Evaluation Directory-based Probability-based

14. Server Volumes- Performance Evaluation Directory-based Probability-based

15. Web Proxy Applications • Cache coherency • Prefetching • Cache replacement • Adaptive freshness interval

16. Conclusions • Using piggyback message • The information can exchange among the various Web components • Server volume • Directory-based • Probability-based

17. Discussion • Good approach • Proxy uses the server-side information • Evaluation • In Real Web Proxy Application • New research for • The proxy filters