zfs discuss - Oct 2007 - Different Sized Disks Recommendation

Hi,

I was first attracted to ZFS (and therefore OpenSolaris) because I thought that
ZFS allowed the used of different sized disks in raidz pools without wasted disk
space. Further research has confirmed that this isn''t possible--by
default.

I have seen a little bit of documentation around using ZFS with slices. I think
this might be the answer, but I would like to be sure what the compromise is if
I go down this path.

I am trying to set up a file server for my home office. I would like as much
storage space as possible, and I need to be able to survive one drive failing.

Here is what I have:

DiskA - 160GB
DiskB - 200GB
DiskC - 250GB
DiskD - 500GB
DiskE - 500GB

(I also have a separate 20GB disk that I will be booting from.)

Here is the best way I can see to maximize storage:
-160GB slice on all disks
--Put these into a raidz1 zpool
-40GB slice on disks B through E
--Put these into a raidz1 zpool
-50GB slice on disks C through E
--Put these into a raidz1 zpool
-250GB slice on disks C and E
--Put these into a raidz1 zpool

Then I put the 4 zpools into a striped zpool.

That [i]should[/i] leave me with 1110GB of storage space (out of the 1610GB of
hard disk space).

If a hard drive fails, one of more of my raidz1 zpools with become degraded.
However, I should be able to replace that drive, slice it like the previous
drive and then I should be back in business.

How much extra work am I going to spend trying to maintain a structure this
complicated? If a drive fails and I replace it with a drive of a different size
(bigger or smaller) what kind of hell will I be putting myself through to
reorganize things? What if I want to add a new disk of a different size? Most
importantly, am I putting my data at risk by having (and manipulating) these
extra ''layers''?

How much more difficult will it be to recover if my system completely crashes
and I need to reinstall the O/S?

I much simpler layout might be Disks A through C in one raidz1 zpool and Disks D
and E in another raidz1 zpool (there is a good chance I''ll buy another
500GB disk in the next month). This leaves me with 820GB. It is a lot less
efficient, but the set up is much simpler to maintain. (It will be a little more
difficult to use because I''ll have two file systems, not one.)

I have attached a spreadsheet that calculates how big each slice should be if I
want to be as efficient as possible with the storage.

Any comments?

Thanks!
 
 
This message posted from opensolaris.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: zfs-disks.xls
Type: application/vnd.ms-excel
Size: 24576 bytes
Desc: not available
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20071029/d60a488b/attachment.xls>

ZFS "copies" attribute could be used to make this easy, but with all
the talk of kernel panics on drive loss and non-guaranteed block placement
across different disks, I don''t like ZFS copies.  (see threads like
http://mail.opensolaris.org/pipermail/zfs-discuss/2007-October/043279.html )

The bottom line in my opinion:  Yes what you are talking about is a big hassle
to create, and yes it is a big hassle to maintain in the future.  Instead
I''d buy another cheap ($99) 500GB disk and RAID5/RAIDZ them together
for a clean and easy 1000GB of space.  The end!
 
 
This message posted from opensolaris.org