On October 27, 2003 08:26 pm, Toad wrote:
> > The other thing this sort of usage pattern does is make other nodes
> > see higher values for SendFailed than are real. �The
> > DecayingRunningAverage class is very quick to move to 1.0... �I do have a
> > variant on this class that uses a window of several minutes to determine
> > the number to feed the running average with. �It seems to get more
> > reasonable
> > numbers for SendFailed.
>
> Is that bad though? Maybe we want it to be pretty sensitive at least for
> sendfailed, so it backs off the node while it is rejecting queries?

I think it is bad.  Problem is that once a node's SF reaches 1 it depends on
other nodes having 'bad' estimates before routing will try it again.   The pattern
posted is very common.  This implies that we really want to track how a node
does over a 5-10 minute period.   This change does not stop SF from growing
it just damps it a bit, meaning we continue to try to route to it a little.

Ed
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to