Evan Daniel wrote: > We should consider doing something about time of day effects, though > I'm not precisely sure what. If we insert one day, and then retrieve > several days later at the same time, I suspect we get a higher success > rate than if we retrieve later at a different time of day, because a > more similar set of nodes will be online.
I pointed to this external website in the past, but I think it's a good time for a reminder: http://crawler.trillinux.org/history.html These are the stats for the G2 (bastard child of gnutella) network. I know that the usage patterns may be different, but I also believe that these detailed usage pattern stats from another p2p network may be insightful. I think that backs pretty well the point raised by Evan that time locality is of importance. There are also histograms for uptime, an animated map of hubs around the world... Also, in G2, hub nodes are encouraged to stay online, just as users of freenet are.
