[rescue] Netra NFS/Enterprise 150 question

[Jonathan C. Patschke]

On Sun, 8 Feb 2009, Michael-John Turner wrote:

> FreeBSD/sparc64 does support ZFS, but a friend of mine (who is
> subscribed to this list :) had a few issues with 7.0-RELEASE on an U60.
> It worked, but he'd get repeatable kernel panics during large copies.

ZFS on FreeBSD requires a good bit of tuning.  Also, like ZFS everywhere,
it requires gobs of memory to perform well and not crash.  Here's what I
have[0] configured on my FreeBSD/amd64 system (running 7.1-RELEASE):

   $ tail -6 /boot/loader.conf
   zfs_load="YES"
   vm.kmem_size_max="1536M"
   vm.kmem_size="1536M"
   vfs.zfs.arc_min="32M"
   vfs.zfs.arc_max="256M"
   vfs.zfs.prefetch_disable=1

The ZFS prefetch stuff was a known source of crashes in 7.0.  I don't
trust it enough to enable it on 7.1, especially since I set up ZFS just
before planning to be half a planet away from the server for a month.

This is on a system with two pools.  One is a zmirror of two 1TB disks
(used for rsnapshot), and one is a plain pool containing a single LUN
exported by a 3ware RAID controller (used for nearly everything else).

I don't have root or swap on ZFS yet, since ZFS is sill rather an
experimental feature on FreeBSD.  Plenty of people have excellent luck
with it, but plenty of other people have kernel panics.  I've, however,
had a much more reliable time with ZFS than with GEOM journalling.

Solaris 11 (as of build 105 or so) is good enough that I'm tempted to dump
FreeBSD entirely and switch back to Solaris once the official release
comes out.  If Solaris Vista^W10 hadn't been such a steaming pile, I'd
have probably never left it.


[0] Experimentally determined on an 8-core system with 8GB memory, 5TB of
     storage managed by ZFS, and light activity.  YMMV.
-- 
Jonathan Patschke ( "I'll be long gone before some smart person ever
Elgin, TX         (  figures out what happened inside this Oval Office."
USA               (                                     --George W. Bush