some ZFS questions

[Andrew McRae]

On 25 Aug 2014, at 04:23, Warren Block <wblock at wonkity.com> wrote:

> On Sun, 24 Aug 2014, Michael Sierchio wrote:
> 
>> On Sun, Aug 24, 2014 at 8:00 PM, Warren Block <wblock at wonkity.com> wrote:
>>> On Sun, 24 Aug 2014, Michael Sierchio wrote:
>>> 
>>>> On Sun, Aug 24, 2014 at 6:42 PM,  <kpneal at pobox.com> wrote:
>>>> 
>>>> 
>>>>> If you can use GPT partitioning then you can have GPT labels on
>>>>> partitions
>>>>> spanning entire disks. Then you can rearrange physical disks to your
>>>>> hearts
>>>>> content.
>>>>> 
>>>>> The bonus is that you can put identifying information in the label name
>>>>> to be even more sure you are replacing the correct disk when the time
>>>>> comes.
>>>> 
>>>> 
>>>> You can do this - and it's the approach that FreeNAS takes, but you
>>>> lose the benefits of presenting a whole disk to ZFS - the ability to
>>>> fully use the disk cache when presented a raw device rather than a
>>>> partition.
>>> 
>>> 
>>> I thought that was a Solaris-only limitation.  In fact, I'm pretty sure it
>>> is, and does not apply on FreeBSD.
>> 
>> You mean Solaris-only benefit?
> 
> No, I mean that the disk cache problem was only on Solaris, and does not apply to FreeBSD.  On FreeBSD, there is no difference in performance between partitions versus the whole disk.

Actually the problem never existed in the first place. It’s an urban myth based on a misunderstanding of the fact that when presented with a whole disk, ZFS tries to enable the disk’s write cache. (And that generally has no effect, except for some old Sun-branded disks that shipped with the write cache disabled, as I understand it.)

Here’s the code, still in FreeBSD:
https://svnweb.freebsd.org/base/stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_disk.c?view=markup#l569

Another comment: http://marc.info/?l=zfs-discuss&m=134022989322125