Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Wed, Nov 07, 2007 at 08:46:35AM +0100, Martijn Plak wrote:
> My 7.0-BETA2 (+ updates) system is crashing regularly on a kmem_map =20
> too small panic too.
> Yesterday I applied the vm_kern.c.2.patch patch and turned off all of =20
> my sysctl tweaks.
> At first, the system was stable for a day. During that day, I ran =20
> make -j4 buildworld several times without trouble.
> Then started a network transfer. This was a usenet download, on 4 TCP =20
> channels, for a total of 10Mbps, from a server at a 'ping distance' =20
> of about 20ms, using hellanzb.py. My guess was that this sort of =20
> transfer puts some memory pressure on the kernel space. Running that =20
> in parallel with the buildworld crashed the system again.
> I hope this helps the investigation. Let me know if I can run other =20
> tests.

> =3D=3D=3D cat /boot/loader.conf
> # Load ZFS and load root system from the RAID-Z array.
> zfs_load=3D"YES"
> vfs.root.mountfrom=3D"zfs:raid/sys/root"
> # Tune ZFS and VM parameters.
> #vfs.zfs.arc_max=3D"64M"
> #kern.maxvnodes=3D"50000"
> #vm.kmem_size_max=3D"512M"
> #vm.kmem_size=3D"512M"
> =3D=3D=3D sysctl vm | grep kmem
> vm.kmem_size_scale: 3
> vm.kmem_size_max: 335544320
> vm.kmem_size_min: 0
> vm.kmem_size: 335544320

The only tuning that is needed (and ZFS will warn now about this) is
vm.kmem_size/vm.kmem_size_max. Can you set those back to 512M and retry?

Pawel Jakub Dawidek http://www.wheel.pl
pjd@FreeBSD.org http://www.FreeBSD.org
FreeBSD committer Am I Evil? Yes, I Am!

Content-Type: application/pgp-signature
Content-Disposition: inline

Version: GnuPG v2.0.4 (FreeBSD)