On Sun, Sep 02, 2001 at 12:38:02PM -0700, Scott G. Miller wrote:
< > 
> > That's patching the syndrome, not the problem. Chances are that even if
> > your requests get through locally, you won't see a reply.
> Why not?  Do we expect another limit to be hit elsewhere?

Yes, several.

When a message is received it is put on the Ticker heap, which uses
blockingRun so won't run until threads are cleared up. The even when
that get's off, the reply (which could be on a new connection) may not
get back in, and the reply also goes via the Ticker heap.

The things to do about this problem are:

- Balance the setting for connection limit and the limit to the number
of threads better (you should need 3-4 times as many threads as live
incoming connections, plus those you wish to use for FCP and other) -
you should hit the incoming connection limit before the threadlimit.

- Set down the connectionTimeout value so idle connections are cleared
off sooner.

- Add better/more aggressive pruning of idle connections when the
connection limit approaches.

- Debug my code to check that there isn't some issue keeping old
connections from being used so that new threads are created
unnecessarily.

-- 
'DeCSS would be fine. Where is it?'
'Here,' Montag touched his head.
'Ah,' Granger smiled and nodded.

Oskar Sandberg
oskar at freenetproject.org

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

Reply via email to