On Wed, Dec 13, 2006 at 09:54:05AM +0000, Michael Rogers wrote: > toad wrote: > >> Any suggestions for a better metric? > > > > The proportion of requests that actually succeed. Load limiting's job is > > to allow as many successes as possible, and if that is not possible > > because of high input load, to throttle the high input load and impose a > > maximum ceiling. > > That makes sense. However we also need to consider throughput in some > way, otherwise a fast reliable network won't look any better than a slow > reliable network.
Okay, so: - The proportion of requests that succeed, local or not. Higher is better. - The maximum number of requests started. Infinity is too high, however if this is high, pSuccess will be low. If a simulation shows high requests started only because of an artefact (lots are being started, none are finishing), then that's an artefact and it needs to be dealt with e.g. by allowing all requests to finish before computing the total psuccess. > > Cheers, > Michael -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: Digital signature URL: <https://emu.freenetproject.org/pipermail/tech/attachments/20061213/b9c6119a/attachment.pgp>
