On May 1, 2002 09:36 am, Edgar Friendly wrote:
> ian at hawk.freenetproject.org (Ian Clarke) writes:
> > My understanding is that at the moment QueryRejected messages do not
> > result in a reduction in CP - the question is whether they should, it
> > seems to me that the alternative is that busy nodes get busier until
> > they stop accepting connections completely.
> >
> > Ian.
>
> I've been thinking about this since you posted it.  My initial
> reaction was "hell no", but as I think more about how overloaded nodes
> are getting, I have to wonder if this might be necessary, to have
> longer timeouts than one minute when a node is overloaded for a long
> time.  The reduction in traffic would also be good for really
> overloaded nodes, as my node is so overloaded it's having to refuse
> connections, it can't cope with the its percieved load.
>
> I say "percieved load" because I wonder if there's some bug in the
> load counting.  I had my computer unplugged from the internet for
> several hours and when I came back to it, my freenet node was still
> reporting about 2/5 of its capacity taken up.  Now there's no reason
> for requests to stay around for that long if they can't go anywhere,
> so I have to wonder if there's some leak in threads or something
> that's decreasing nodes' capacity after a while.  I can't debug this
> myself, but will gladly supply any information I can.

I wonder if this is just the overhead needed to support the various
servlets and other core freenet code?  Java likes threads and freenet
really uses them.

Ed Tomlinson

_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://lists.freenetproject.org/mailman/listinfo/devl

Reply via email to