End-to-end performance prediction for the Internet (Work in progress)

Many applications designed for wide-area systems use replication to provide a service at multiple locations. From which site should a client choose to obtain service? A client can specify its needs in terms of communication latency, bandwidth, or error rate. This specification must be matched against the expected performance available from each site. I present the results of two experiments that measured end-to-end performance, and discuss how the results can be used for prediction.