Ideally, we would like to perform a long-running experiment that would compare end-to-end performance of our optimistic delta mechanism with existing proxy caching. One way to perform such measurements would be to get a set of random URLs, and then request each URL periodically for an extended time using the optimistic delta approach and the existing approach, and compare the average response time. We plan to perform such experiments in the future.
To date, we have focussed on ``microbenchmarks'' to study the two extremes of interest: the case when the client does not have a page cached, and must obtain a full copy of the page (possibly a stale copy followed by a delta or confirmation that it is current); and the case when the client and server have the same copy cached and the server can send a delta. We compare each of these against the response time of an unmodified system that connects directly to the server proxy and does not use deltas.