ZFS on a single disk?
Dr Josef Karthauser
joe at tao.org.uk
Fri Mar 4 14:06:29 UTC 2011
On 4 Mar 2011, at 14:00, Jeremy Chadwick wrote:
> On Fri, Mar 04, 2011 at 01:49:15PM +0000, Dr Josef Karthauser wrote:
>> I don't believe that that's an issue anymore. The XEN kernel comes
>> configured with PAE as a default option, and I've seen elsewhere that
>> there is no technical problems running ZFS in a PAE environment. (The
>> PAE docs are out of date when they say we can't use kernel modules,
>> and any 64 bit aware kernel model should run with PAE with no
>> difficulties).
>
> XEN isn't something I'm familiar with, which is obviously a huge part of
> the problem with me trying to give you advice on the matter. :-) The
> only virtualisation "system" I'm familiar with is VMware Workstation,
> which isn't anything like XEN.
>
> What I'm going off of is /sys/i386/conf/PAE vs. /sys/i386/conf/XEN.
>
> Be sure to notice all the "nodevice" lines in /sys/i386/conf/PAE, and
> the comment directly above those. The XEN configuration file has
> "options PAE", which is ultimately what a driver/piece of kernel code
> would use for compile-time detection for supporting/working under PAE.
>
> For example, arcmsr(4) will flat out panic() (intentionally) if PAE
> is used. bge(4) and twa(4) appear to have a 4GB boundary on DMA; I
> don't know the implications of this.
Luckily I care not about device support. As long as the paravirtualised
devices work, and zfs works, I need nothing else.
So, I guess the zfs question boils down to whether (given everything
else being equal with PAE support) I could consider zfs on i386 with
a PAE memory model to be stable enough for a production server (on 8.2).
Joe
More information about the freebsd-fs
mailing list