Automatic Selection of Nearby Web Servers

Performance of global services such as the world-wide web can be improved by physically distributed replicas. Most replicated systems today request users to select manually a nearby replica (typically from a list of sites), yet users often are not willing or able to make an informed choice. This paper describes an approach to automatic replica selection which works without change to existing web clients and web and FTP servers. We describe the assumptions behind our approach and propose several metrics for estimation of network distance. We examine the feasibility be-hind this approach and compare the time required for replica-selection by each algorithm. Finally, we de-scribe a small change to HTTP that would improve replica selection transparency.