10. Conclusion and Future Work

Today, understanding the performance characteristics of Internet services is critical to evolving and engineering Internet services to match changing demand levels, client populations, and global network characteristics. Existing tools for evaluating web service performance typically rely on active probing to a fixed set of URLs or on web page instrumentation that monitors download performance to a client and transmits a summary back to a server. This paper presents, EtE monitor, a novel approach to measuring web site performance. Our system passively collects packet traces from the server site to determine service performance characteristics. We introduce a two-pass heuristic method and a statistical filtering mechanism to accurately reconstruct composition of individual page and performance characteristics integrated across all client accesses.

Relative to existing approaches, EtE monitor offers the following benefits: i) a breakdown between the network and server overhead of retrieving a web page, ii) longitudinal information for all client accesses, not just the subset probed by a third party, iii) characteristics of accesses that are aborted by clients, and iv) quantification of the benefits of network and browser caches on server performance. Our initial implementation and performance analysis across two sample sites confirm the utility of our approach. We are currently investigating the use of our tool to understand the client performance on a per-network region. This analysis can aid in the placement of wide-area replicas or in the choice of an appropriate content distribution network. Finally, our architecture is general to analyzing the performance of multi-tiered web services. For example, application-specific log processing can be used to reconstruct the breakdown of latency across tiers for communication between a load balancing switch and a front end web server, or communication between a web server and the storage tier/database system.