On Sat, Nov 08, 2003 at 11:48:26AM -0500, Andrew Rodland wrote: > Tom Kaitchuck wrote: > > > Second a few lines later there is something like: > > double tDNF = etDNF.guessTime(k) * htl; > > double pNotConnectFailedOrSearchFailed = > > (1 - pConnectFailed) * (1- pSearchFailed); > > estimate += pNotConnectFailedOrSearchFailed > > * (pDNF - pLegitDNF) * (tDNF + requestFailTime); > > Now, if I read this correctly, (I'm assuming that this code is used to > > obtain the next node to route to for each node in the chain.) We find the > > probability of a legit DNF and then multiply that by the estimated time to > > re-request that. (tDNF + requestFailTime). However tDNF is multiplied by > > HTL, (Which I assume to be the current HTL). The problem with this is > > that, if we are not the originating node, we are not going to re-request > > it if it fails. So instead it will be re-requested from the original node. > > So shouldn't we be trying to estimate how long that would take? IE: double > > tDNF = etDNF.guessTime(k) * MAXhtl; > > You've got about half a point there. :) > > tDNF is the estimated time it will actually take for us to get the DNF back. > It's normalized to HTL=1 on report, and then multiplied out by the HTL on > request. > > requestFailTime is the number for the estimated time that it would take to > rerequest it. You're right in saying that that's *not* what we're going to > do, but I'm not entirely sure how that needs to be fixed. Note also that > the requestFailTime isn't normalized against HTL (because hopefully, it > shouldn't be linear in HTL), so we can't just correct it to maxHTL. > > However, it does look like we'd be underestimating, at least by a little, > the cost of a "false DNF" (pDNF - pLegitDNF), which could result in > not-so-good routing.
How so? > > --hobbs -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
signature.asc
Description: Digital signature
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
