MadMax017 wrote:Well from the downloaders point of view, the slot must already be open on the server's machine in order for them to even have a hope of getting the file faster from that machine in question. The client machine would only have to perform a "dummy download", possibly download the server's file list to get an idea of the speed of the new server, and if it is a better speed then switch sources.
The problem with this is that there is often a disconnect between busting transfer speed and sustained speed. In fact you this might be intentional at times; a user might want to shoot his file list out really quickly before going back to his normal uploading.
Reliable speed measurement of the type you're suggesting is one of the grails of that branch of computer science. Lots of people are working on it, but not much progress is being made.
I would suggest, though, that a source be chosen based on TTL. If I have two sources that have vastly different TTL measurements, there's an ok chance that one's on the other side of the world from me. Extremely low TTL values might even locate sources inside your own ISP or local net.
But I believe Arne has basically nixed all ideas of this sort.