Re: [zfs-discuss] LUN expansion choices

2012-11-14 Thread Peter Tribble
On Tue, Nov 13, 2012 at 6:16 PM, Karl Wagner  wrote:
> On 2012-11-13 17:42, Peter Tribble wrote:
>
> > Given storage provisioned off a SAN (I know, but sometimes that's
> >  what you have to work with), what's the best way to expand a pool?
> >
> > Specifically, I can either grow existing LUNs, a]or add new LUNs.
> >
> > As an example, If I have 24x 2TB LUNs, and wish to double the
> > size of the pool, is it better to add 24 additional 2TB LUNs, or
> > get each of the existing LUNs expanded to 4TB each?
>
> This is only my opinion, but I would say you'd be better off expanding your
> current LUNs.
>
> The reason for this is balance. Currently, your data should be spread fairly
> evenly over the LUNs. If you add more, those will be empty, which will
> affect how data is written (data will try to go to those first).
>
> If you just expand your current LUNs, the data will remain balanced, and ZFS
> will just use the additional space.

Maybe, or maybe not. If you think in terms of metaslabs, then there
isn't any difference between creating extra metaslabs by growing an
existing LUN and creating new LUNs. With pooled storage on the
SAN back-end, there's no difference in I/O placement either.

Peripherally, this note by Adam Leventhal may be of interest

http://dtrace.org/blogs/ahl/2012/11/08/zfs-trivia-metaslabs/

-- 
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] LUN expansion choices

2012-11-13 Thread Fajar A. Nugraha
On Wed, Nov 14, 2012 at 1:35 AM, Brian Wilson
 wrote:
> So it depends on your setup. In your case if it's at all painful to grow the 
> LUNs, what I'd probably do is allocate new 4TB LUNs - and replace your 2TB 
> LUNs with them one at a time with zpool replace, and wait for the resliver to 
> finish each time. With autoexpansion on,

Yup, that's the gotcha. AFAIK autoexpand is off by default. You should
be able to use "zpool online -e" to force the expansion though.

> you should get the additional capacity as soon as the resliver for each one 
> is done, and the old 2TB LUNs should be reclaimable as soon as it's 
> reslivered out.

Minor correction: the additional capacity is only usable after a top
level vdev is completely replaced. In case of stripe-of-mirrors, this
is as soon as all vdev in one mirror is replaced. In the case of
raidzX, this is when all vdev is replaced.

-- 
Fajar
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] LUN expansion choices

2012-11-13 Thread Brian Wilson
Not sure if this will make it to the list, but I'll try...

On 11/13/12, Peter Tribble 
 wrote:
> Given storage provisioned off a SAN (I know, but sometimes that's
> what you have to work with), what's the best way to expand a pool?
> 
> Specifically, I can either grow existing LUNs, a]or add new LUNs.
> 
> As an example, If I have 24x 2TB LUNs, and wish to double the
> size of the pool, is it better to add 24 additional 2TB LUNs, or
> get each of the existing LUNs expanded to 4TB each?






The thing I've found about growing the LUN is to evaluate if it saves you time 
and effort in your setup to make it work. So, first I figure out if it's 
possible. Then I figure out how big a PITA it is compared to the PITA of 
allocating new LUNs. For example I've got one SAN backend here where it's easy 
as pie (lun resize blah and then OS steps) and another backend where it 
requires a major undertaking (create temporary horcm files, destroy mirrors, 
run special command to resize mirror, wait, run special command to resize 
source, wait, recreate mirrors, delete temporary horcms, etc etc and *then* the 
OS steps).






What I love about ZFS is that it can handle either approach.






So it depends on your setup. In your case if it's at all painful to grow the 
LUNs, what I'd probably do is allocate new 4TB LUNs - and replace your 2TB LUNs 
with them one at a time with zpool replace, and wait for the resliver to finish 
each time. With autoexpansion on, you should get the additional capacity as 
soon as the resliver for each one is done, and the old 2TB LUNs should be 
reclaimable as soon as it's reslivered out.






That being said, I'm not aware of what any deeper implications of doing that 
are - from what Karl said about balancing the data out as one example.






Cheers,


Brian



> 
> -- 
> -Peter Tribble
> http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/
> ___
> zfs-discuss mailing list
> zfs-discuss@opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

--
-- 




---

Brian Wilson, Solaris SE, UW-Madison DoIT

Room 3114 CS&S 608-263-8047

brian.wilson(a)doit.wisc.edu

'I try to save a life a day. Usually it's my own.' - John Crichton

---
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] LUN expansion choices

2012-11-13 Thread Karl Wagner
 

On 2012-11-13 17:42, Peter Tribble wrote: 

> Given storage
provisioned off a SAN (I know, but sometimes that's
> what you have to
work with), what's the best way to expand a pool?
> 
> Specifically, I
can either grow existing LUNs, a]or add new LUNs.
> 
> As an example, If
I have 24x 2TB LUNs, and wish to double the
> size of the pool, is it
better to add 24 additional 2TB LUNs, or
> get each of the existing LUNs
expanded to 4TB each?

This is only my opinion, but I would say you'd be
better off expanding your current LUNs. 

The reason for this is
balance. Currently, your data should be spread fairly evenly over the
LUNs. If you add more, those will be empty, which will affect how data
is written (data will try to go to those first). 

If you just expand
your current LUNs, the data will remain balanced, and ZFS will just use
the additional space. 

I _think _that's how it would work. Others here
will be able to give a more definitive answer. ___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss