Resize RAID-0?

Resize RAID-0?

[Patrick J. LoPresti]

From reading the documentation and source code, I gather "mdadm
--grow" is not supported for RAID-0 devices.

In my application, I am using md RAID-0 to stripe among several
(identical) hardware RAID-0 chassis.  I would like to extend my setup
by adding another chassis, but apparently, I cannot (?).

The next time I build a system like this, is there any way to get what
I want using Linux?  That is, striping among devices, no parity, but
with the ability to grow in the future?

Thanks!

 - Pat

Re: Resize RAID-0?

[Michael Evans]

On Thu, Mar 11, 2010 at 7:39 PM, Patrick J. LoPresti <lopresti@gmail.com> wrote:
> From reading the documentation and source code, I gather "mdadm
> --grow" is not supported for RAID-0 devices.
>
> In my application, I am using md RAID-0 to stripe among several
> (identical) hardware RAID-0 chassis.  I would like to extend my setup
> by adding another chassis, but apparently, I cannot (?).
>
> The next time I build a system like this, is there any way to get what
> I want using Linux?  That is, striping among devices, no parity, but
> with the ability to grow in the future?
>
> Thanks!
>
>  - Pat
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

I THINK that LVM may allow you to do that.  I know the newer version
of the labels supports /mixing/ how many stripes are in a given
segment of storage. You may be able to, via some slightly non-trivial
complexity but with existing commands and a little free space, shuffle
your data out of an M device set of storage in to an N device set of
storage striping.

I don't actually see any technical reason why restriping raid0 or
another raid level to any other should be impossible to support. I
just don't believe the feature currently exists, and possibly that the
metadata might need changes to allow for different segments of backing
storage and use within the same md container.  However were it even
possible to store two types then only a small critical window would
need backup, and everything else could sync to the largest portion of
non-overlap.  The worse case would be same size to same size reshape;
a literal nightmare performance wise since the whole set would be a
critical section.  Even reserving a fraction of a megabyte at creation
time would at least allow the operation on /that/ to be done as a
non-critical section.  Alternatively if a write-intent map were in use
it could instead be destroyed and it's space used for the critical
section.  The map might then flop between front and back during
different reshapes or shifts.

I just got done glancing at a few locations (website, documentation,
and kernel code) and didn't see anything like a representation of what
the metablock might look like, so I can't say for sure if looks like I
expect from the behavior.
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Resize RAID-0?

[John Robinson]

On 12/03/2010 03:39, Patrick J. LoPresti wrote:
> From reading the documentation and source code, I gather "mdadm
> --grow" is not supported for RAID-0 devices.
> 
> In my application, I am using md RAID-0 to stripe among several
> (identical) hardware RAID-0 chassis.  I would like to extend my setup
> by adding another chassis, but apparently, I cannot (?).

If I remember correctly, there were patches posted to this list recently 
to add support for reshaping RAID-0, and for migrating from RAID-0 to 
other RAID levels, but I don't remember whether they've made it in to 
the kernel/mdadm yet. I recommend a trawl through the last few months' 
archives.

Cheers,

John.

Re: Resize RAID-0?

[Neil Brown]

On Fri, 12 Mar 2010 10:40:58 +0000
John Robinson <john.robinson@anonymous.org.uk> wrote:

> On 12/03/2010 03:39, Patrick J. LoPresti wrote:
> > From reading the documentation and source code, I gather "mdadm
> > --grow" is not supported for RAID-0 devices.
> > 
> > In my application, I am using md RAID-0 to stripe among several
> > (identical) hardware RAID-0 chassis.  I would like to extend my setup
> > by adding another chassis, but apparently, I cannot (?).
> 
> If I remember correctly, there were patches posted to this list recently 
> to add support for reshaping RAID-0, and for migrating from RAID-0 to 
> other RAID levels, but I don't remember whether they've made it in to 
> the kernel/mdadm yet. I recommend a trawl through the last few months' 
> archives.

I'll save you the effort.  No they haven't.

The current plan is to support reshaping RAID-0 but converting the
RAID0 to a degraded RAID4, then reshaping it, then converting back.

But the code isn't ready yet.

NeilBrown

Re: Resize RAID-0?

[Kristleifur Daðason]

On Fri, Mar 12, 2010 at 3:39 AM, Patrick J. LoPresti <lopresti@gmail.com> wrote:
> From reading the documentation and source code, I gather "mdadm
> --grow" is not supported for RAID-0 devices.
>
> In my application, I am using md RAID-0 to stripe among several
> (identical) hardware RAID-0 chassis.  I would like to extend my setup
> by adding another chassis, but apparently, I cannot (?).
>
> The next time I build a system like this, is there any way to get what
> I want using Linux?  That is, striping among devices, no parity, but
> with the ability to grow in the future?

I agree with others on the list - I believe that LVM will perform well
as a flexible stripe wrangler on top of mdadm arrays. For instance
some LVM striping over mdadm parity-based stuff underneath.

Do you want to add a whole new stripe device, so changing the
coarse-scale layout, or grow the underlying data containers?

I had a problem related to growing underlying containers which was
solved on the list a few weeks ago. I had two largish RAID6 arrays and
an mdadm stripe-set over those. I added some disks to both RAID6
subarrays and wanted to expand the stripe-array to use the space
gained. Here's a copy of the thread, "raid0 not growable?":
http://www.issociate.de/board/post/501375/raid0_not_growable?.html

But perhaps you wanted to do something like add a 3rd stripe to a
two-stripe array? (I myself don't know if that's possible or not.)

As you are running striped, I assume that the data is either
perishable or backed up. I'd suggest you look into moving to LVM. (LVM
has its quirks - so look before you leap, perhaps :) ) I suggest you
play a little with LVM in a virtual-machine; In the gas-lit and
magical world of open-source RAID management, there's nothing like
direct experience to tell you what's possible and what doesn't work.
The tools and the power are out there, but you have to do a little
hunting in the forest.

Best of luck!

-- Kristleifur
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Resize RAID-0?

[Patrick J. LoPresti]

First, thanks to everyone for the helpful responses.

On Fri, Mar 12, 2010 at 3:36 AM, Kristleifur Daðason
<kristleifur@gmail.com> wrote:
>
> I agree with others on the list - I believe that LVM will perform well
> as a flexible stripe wrangler on top of mdadm arrays. For instance
> some LVM striping over mdadm parity-based stuff underneath.

I made a typo in my original post, so let me clarify.  In my
configuration, I have a pair of dual-channel FC HBA cards.  I
currently have three hardware RAID-5 chassis connected to three of the
HBA channels, and I am using Linux md to perform RAID-0 striping
across the channels.  This gives me very high performance -- 800
MiB/sec sustained reads and writes, for hundreds of gigabytes, using
XFS -- at very modest CPU utiliziation.

What I want is to connect a fourth, identical hardware RAID chassis to
the fourth HBA channel and then re-stripe the md RAID-0 across all
four units.  I now understand that this is not currently supported.

> Do you want to add a whole new stripe device, so changing the
> coarse-scale layout, or grow the underlying data containers?

New stripe device.

> As you are running striped, I assume that the data is either
> perishable or backed up.

My data is all reproducible.  (The system spends its time processing
large volumes of data that originated somewhere else.)  We are using
underlying RAID-5 chassis to improve MTBF, not to prevent data loss.

> I'd suggest you look into moving to LVM.

Thanks to Michael Evans and to you for this suggestion; I will investigate LVM.

Thanks once again to everyone who responded.  The world of Linux RAID
may be "gas-lit and magical", but in many ways the support is superior
to any commercial product I have seen.

 - Pat
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Resize RAID-0?

[Patrick J. LoPresti]

Addendum (for the archives).

EVMS claims to support this:

http://evms.sourceforge.net/user_guide/#resizeraid0

 - Pat

Re: Resize RAID-0?

[Neil Brown]

On Fri, 12 Mar 2010 16:09:43 -0800
"Patrick J. LoPresti" <lopresti@gmail.com> wrote:

> Addendum (for the archives).
> 
> EVMS claims to support this:
> 
> http://evms.sourceforge.net/user_guide/#resizeraid0

Just for clarification, that is an off-line non-interruptible (better have a
UPS) resize.

There is a command line tool to do this:

  http://unthought.net/raidreconf/

however I wouldn't recommend it, because if anything goes wrong in the middle,
you lose your data.

NeilBrown