On Tue, Apr 18, 2006 at 12:47:14AM +0200, John B?ckstrand wrote: > Jusa Saari wrote: > >Geometric mean is still vulnerable to this, just less so than arithmethic > >mean. Besides, it is likely much more expensive computationally - or do > >modern day chips come with instructions to calculate the Nth root (could > >be, I haven't really examined their capabilities) ? > > I dont think they do.
I bet it's still faster than pointer chasing in many instances. > > >Is there any particular advantage geometric mean has over median ? > > Median requires sorting, which scales much worse than one single > nth-root operator since it depends on the number of peers. Well yes but there are relatively few peers. I'm willing to switch to median if it's better quality. > > I'm pretty sure any approach is practically fast enough though ;) > > There is one thing this entire approach does not take into account > though: multiple routes since it only gives one measure of load for the > entire node, eg. even if the internet connection is saturated, does not > mean the local LAN is. Uncommon scenario though, but I still feel the > entire approach of trying to measure the connections queue is a bit weird. Hmm. Maybe you have a point. Do you have an alternative? > > --- > John B?ckstrand -- Matthew J Toseland - toad at amphibian.dyndns.org Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so. -------------- 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/devl/attachments/20060421/d6f519ae/attachment.pgp>
