[BBLISA] RE: storevault / netapp

[Rob Cambra]

On Wed, 27 Feb 2008, Peter Galvin wrote:

> At the moment ZFS cannot expand a storage pool by single disk increments.
> You could for example expand a RAIDZ pool by concatenating another RAIDZ set
> to it, but not by adding a single disk.

Sure you can, "zpool add $poolname $device"

> On 2/27/08 11:44 AM, "Edward Ned Harvey" <bblisa2 at nedharvey.com> wrote:
>
> > The second one, what I'm referring to as "dualparity" is the ability to
> > sustain 2 simultaneous disk failures without data loss.  This is a function
> > of the hardware raid controller, and not the filesystem.  Am I making any
> > bad assumptions here?  I am not aware of any hardware raid controllers that
> > can safely lose two disks at the same time, except the
> > netapp/storevault/other enterprise filers.  What I'm saying is, it's not
> > supported like this on a sun xstore solaris zfs machine.  Right?

Definitely supports multiple simultaneous device failures.  We did our
initial testing on "whitebox" systems/arrays, then decided to use xstore
for production.

> > incrementally add disks, without degradation of redundancy or downtime.  If
> > I've got 6 disks in a dualparity configuration, controlled by hardware DP
> > raid (usable capacity of 4 disks assuming no hotspare), I just slap in one
> > more disk, and increase the size of my FS by 25%.  No downtime, no
> > degradation of redundancy.  Perhaps the system runs slow for a couple hours.
> > This too is not supported in the sun xstore solaris zfs configuration,
> > right?

Definitely supported, no need to recompute.

> > I don't think it's fair to say it's supported in OSX.  In OSX, you cannot
> > format and create a ZFS filesystem, and you can only mount it read-only.
> > Although read/write variations exist, they are very immature at best.

You can definitely create a ZFS filesystem in OSX, as of Tiger I believe,
can't say I run it daily, but we did a test just to qualify this feature.

> >
> > Same is true for Linux.  Although it's open source, it can't be compiled in
> > the linux kernel any time soon.  Because it's written under CDDL, which
> > conflicts with GPL.  See here:  http://en.wikipedia.org/wiki/ZFS#Platforms

Fair enough.  I don't really know much about Linux to be honest.

Also -- ZFS does not need battery backup to deal with hardware crashes
mid-write.  There is a lengthy presentation out there which describes this
in detail.

-rob