Subject: Re: NetBSD locking on mclpool error
To: Manuel Bouyer <bouyer@antioche.lip6.fr>
From: Chris G. Demetriou <cgd@netbsd.org>
List: netbsd-help
Date: 09/22/1999 13:31:44
Manuel Bouyer <bouyer@antioche.lip6.fr> writes:
> > In a nutshell, this is due to the way clusters are used by many
> > drivers, especially DMA-using drivers like 'ex', etc.
> 
> Can you give details about this ? I'm curious :)

Charles would be better to explain it, since it's his patch, but from
what I understand (i'm not a networking guy!) it's something like:

Drivers structured like like 'ex' use complete mbuf clusters for
incoming data.  If you receive a lot of small packets, they'll each
end up using their own mbuf cluster (even though they're each using
only a small fraction of the mbuf cluster storage).

The patch reduces that waste by moving moving more data into clusters
(and freeing clusters which are then unneeded) in the socket buffer
layer.


cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.