On Sun, Nov 11, 2001 at 05:49:05PM -0500, Gianni Johansson wrote:
<> 
> There are also issues with the connection management which are pretty
> severe.  I kind of wonder whether there's any real routing happening
> in the current network, or whether nodes are just desperately
> connecting to any node that answers them. I've been working on this a
> little since DC.  I know Oskar also has some ideas.  I think that
> between us we can probably get things cleaned up relatively quickly
> (I also doubt that Oskar will share my optimism ;-) ).

I think I have the optimism (IMHO most of 0.4 is better designed then
the previous versions), but just not the time.

The connection purging needs to be more agressive (currently it won't
kill any connections younger then half the timeout) and probably we need
something to target many too many connections to the same node (that
could be done when a new connection is added to OCM).

Also, to avoid the deadlock seen in thelemas logs and dumps, we could
probably use a method like:

ThreadManager.forceRun(long millis)

that waits for pooled thread for some time and then forces the creation
if none is available (calling that with 30000 rather than blockingRun()
from Ticker should keep an overloaded node from indefinite deadlock).

-- 

Oskar Sandberg
oskar at freenetproject.org

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

Reply via email to