mbuf leakage with nfs/zfs? (was: em0 freezes on ZFS server)

[Gerrit Kühn]

On Fri, 26 Feb 2010 17:41:02 +0200 Daniel Braniss <danny at cs.huji.ac.il>
wrote about Re: em0 freezes on ZFS server :

DB> check:
DB> 	ftp://ftp.cs.huji.ac.il/users/danny/freebsd/plot.ps
DB> x is seconds, y is mbus current.

Looks not as bad as mine. I had 37k when I rebooted the machine some
minutes ago (and it's basically idle, just serving a few nfs clients that
don't do much).
But from the values Jeremy has posted and from my own comparsisons here I
would think that something like 5k of mbuf clusters would be normal for my
machine (and probably also for yours).

Some more info from my side:
In the meantime I also tried a different network interface. The
nfe-interface that is onboard causes the same problems, so it is probably
not an em-specific issue.
Furthermore I found this via Google:
<http://lists.freebsd.org/pipermail/freebsd-current/2009-December/014062.html>.
I patched and recompiled my kernel with this, just to try it out. Right
now I have

2264/1321/3585 mbufs in use (current/cache/total)
1239/1017/2256/65000 mbuf clusters in use (current/cache/total/max)
1239/809 mbuf+clusters out of packet secondary zone in use (current/cache)

but the uptime is only 12min so far. In some hours I'll know for certain
if this patch has anything to do with the problem.


cu
  Gerrit