> What is a good value for NMBCLUSTERS on a busy server?  Is there an upper
> limit?
> 
> I was using the default (512, I think), then upped it to 2048, but I'm still
> getting warnings in syslog and the networking stuff stops working every now
> and then.  I'm going to try out 8192 - should I bump it even higher?  (This
> machine has 384 MB of RAM, if that matters.)

AFAIK, the only downside to raising it is that more memory is
reserved for mbufs.  I'd raise it to 16384 or maybe even higher.
Where I work, we have a sparc running NetBSD that was generating
those messages all of the time.  It was running squid and is now
running at 8192.  It's now running the apache proxy/cache, and we
are not running out of mbufs anymore, but with squid and 8192, we
were seeing that.

-allen