Understanding and reducing Web delays

low performance costs e-commerce Web sites as much as $4.35 billion annuall y in lost revenue. Perceived latency - the amount of time between when a us er issues a request and receives a response-is a critical issue. Research into improving performance falls into two categories: work on serv ers and work on networks and protocols. On the server side, previous work h as focused on techniques for improving server performance. Such studies sho w how Web servers behave under a range of loads. These studies often sugges t enhancements to application implementations and the operating systems tho se servers run. On the network side, research has focused on improving network infrastructu re performance for Internet applications. Studies focusing on network dynam ics have resulted in several enhancements to HTTP, including data compressi on, persistent connections, and pipelining. These improvements are all part of HTTP 1.1. However, little work has been done on common latency sources that cause the overall delays that frustrate end users. The future of performance improve ments lies in developing additional techniques to help implement efficient, scalable, and stable improvements that enhance the end-user experience.