Improving World Wide Web Latency

The HTTP protocol, as currently used in the World Wide Web, uses a separate TCP connection for each le requested. This adds signi cant and unnecessary overhead, especially in the number of network round trips required. We analyse the costs of this approach and propose simple modi cations to HTTP that, while interoperating with unmodi ed implementations, avoid the unnecessary network costs. We have implemented our modi cations, and our measurements show that they dramatically reduce latencies. We have also investigated the e ectiveness of a scheme to mask network latency by prefetching les likely to be requested next, while the user is browsing through the currently displayed page. Our results indicate a signi cant bene t from prefetching at the cost of an increase in network tra c.