entries: 100
 Global mean traffic (queries per hour):3321.9222222222224
 Local mean traffic (queries per hour): 689.6485665846112
 Current advertise probability: 0.15754031587654324
 Current proportion of requests being accepted: 0.99

This is what I expect to see.  I have out going bandwidth limited
to half of the available, 6k/s.  Node is well integrated with a 8G ds with
about 19000 keys.

I also have noticed the network has shifted my nodes specialization 
reciently.  I was down for about 30 hours.  After this my node was found
quickly but after about three days I noticed the dense part of the routing
histo had moved.  Looking that the ds histo the content has not shifted  
much yet.  Nodes changing specialization after a short down my also be
part of the problem...

Ed Tomlinson

On May 16, 2003 03:42 pm, Toad wrote:
> On Fri, May 16, 2003 at 07:29:02PM +0100, Toad wrote:
> > Or alternatively,
> >
> > http://127.0.0.1:8888/servlet/nodestatus/version_histogram.txt
> >
> >  :)))
> > >
> > > (My apologies for the *horrible* perl -- it's not my best language.)
> > > Anyway, this shows that out of the 100 nodes in my routing table, only
> > > 14 of them are running pre-build-579 nodes.  Builds 593-595 dominate
> > > in this part of Freenetland.  (That one build-629 node is probably
> > > pre-pcaching too, but I'm not certain.)
> > >
> > > > Maybe we should make 0.5.2 mandatory?
> > >
> > > I think there's not enough evidence on either side of the pcaching
> > > argument to decide this yet.
> >
> > There won't be until pcaching actually does something. The statistics
> > strongly suggest that pcaching is not doing much because every two hops
> > or so it hits a pre-pcaching node. Or that there is some extremely wierd
> > bug happening somewhere...
>
> So, a theory presents itself: we have
> Set A: N nodes with high traffic, ~3k reqs per hour to ~ 100k reqs per hour
> Set B: M nodes with low traffic, hundreds of reqs per hour or less
>
> N is a LOT smaller than M.
>
> Set A nodes have low advertising (datasource reset) probabilities, but
> Set B nodes have high advertising probabilities, due to the load
> balancing algorithm, which evidently isn't working. Hence a request
> goes out on the network and after a few hops meets a set B node, and
> gets its datasource reset and consequentially hopsSinceReset set to
> zero. Interestingly, the nodes that get little traffic are NOT locally
> overloaded, or they would have lower reset probabilities due to the
> (proportion of requests accepted) ^ 5 which is multiplied into the reset
> probability...
>
> Any suggestions as to what to do about this?
>
> BTW, I want to see some reset probabilities from live nodes to try to
> confirm this statistically. Mail me the top five lines from
> http://127.0.0.1:8888/servlet/nodeinfo/networking/loadstats
>
> > > --
> > > Greg Wooledge                  |   "Truth belongs to everybody."
> > > greg at wooledge.org              |    - The Red Hot Chili Peppers
> > > http://wooledge.org/~greg/     |
>
> _______________________________________________
> devl mailing list
> devl at freenetproject.org
> http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl

Reply via email to