> From: Matthew Toseland <toad at amphibian.dyndns.org>
> Date: 28 Oct 2002 17:00:13 +0000
> 
> Well... we've had reports of the development branch causing 100% cpu
> usage and high loads, but on the other hand the load balancing code is
> promising. If you don't mind living on the edge, upgrade to the
> development branch (replace your freenet.jar with
> http://freenetproject.org/snapshots/freenet-unstable-latest.jar), set
> doLoadBalance=yes in your freenet.conf and tell us how it goes.

Another thing that could be done is to turn down the default announce
time delay so that newer nodes are brought into the fold faster.

I can safely say that my node is getting hammered as well. 138% of what
it should be (250 threads, 5978 requests per hour and my current routing
time is sometimes more then 2 seconds and sometimes reporting 0ms) and
getting lots and lots and lots of RNFs. This is with build 525.

I don't think that the solution to the network overloading is through
more effective throttling though. An overloaded Freenet does not decay
gracefully it would seem. A more preferable failure mode would be longer
waits rather then 99.9% RNFs.

This does indicate a weakness though. Freenet may be able to handle a
good slshdotting of requests but not a good slashdotting of
new/transient nodes. Something to remember for future PR pushes :] It
may be a good reason to have many frequent small point releases with
less significant changes or maybe even have wider coverage of the
pre-release to draw in a the newcomers over an extended period of time.
Hopefully the current network crippling rush does not damage Freenet's
credibility further and people stick with it until it can adapt.

Mike


_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to