tmcd35 Posted February 21, 2010 Posted February 21, 2010 A little advice, how would you set this up? The proposed system: Windows 2008R2 Standard Server 16 x 450Gb 15k SAS Hard Drives For legacy reasons I need to set his up as a NAS with SMB2 shared folders for various data (Home Dirs, Profiles, VM Images, ISO's, etc) rather than using a SAN. My thinking is 12 Drives in RAID-50 with two global hot spares. This will give me 1 large 5.4Tb volume in Windows. Do I - keep it as 1 large volume and just create the shared folders as needed? or - Split the volume into smaller partitions - 1 for home dirs, 1 for VM Images, etc? or - use a different RAID setup to present smaller LUNs for each shared folder? If it helps I'm thinking of using two network connections to the NAS. 1x1Gbps connection from my core network for access to the home dirs, profiles, etc. 1x2Gbps (trunk) connection from a seperate 'storage network' for the Virtual Machine data.
SYNACK Posted February 21, 2010 Posted February 21, 2010 (edited) I think your maths may be a little out there: RAID 5+0 with 2*spare = 7drives*450GB - 450GB parity = 2.7TB mirrored with 2 hot spares I would be tempted to use RAID6 with hotspares for a larger data space RAID6 with 2 hot spares = 14drives*450GB - 2drives parity = 5.4TB with 2 hotspares If this is ever going to be switched over to a SAN style setup, I would use smaller partitions that could be alloacted to different servers more easiy, all on the one large RAID volume though to make the best use of the disks. Just wondering about the legacy bit, you could use a SAN then use an existing server to offer this out as SMB2 without much hassle or depending one the hardware a SAN may offer both modes anyway. Edited February 21, 2010 by SYNACK
tmcd35 Posted February 21, 2010 Author Posted February 21, 2010 (edited) I'm looking at building a 16-drive storage array primarily for use to store Virtual Machines. I want 2 drives as hot-spares, better safe than sorry, leaving 14 drives for the array. Using either RAID-50 or RAID-6, 2 drives will be lost to parity. So 12 drives would hold data, 12*450Gb = 5.4Tb. Is there a reason why I can't have a 14 drive RAID-50 array? I'm lead to believe that RAID-50 could be marginally faster than RAID-6. But RAID-6 could offer better redundancy. As for the legacy bit - that's my three existing servers. The processors don't support virtualisation, so I can't run Hyper-V. I think that rules out Cluster Shared Volumes. Since I want the VM's all in one central location that any server can access, I'm looking at making the fastest NAS box I can I think it makes more sense having one large RAID volume. I'm just wondering if there'd be any performance gains by partitioning that volume, or would it be better leaving it as a single whole? EDIT: ops, re-read what I first wrote and it wasn't what I ment at all I'm not looking at 12 drives total with 2 being hot spares, leaving 8 for data (after parity) as that would not (as SYNACK rightly points out) make 5.4Tb. Edited February 21, 2010 by tmcd35
SYNACK Posted February 21, 2010 Posted February 21, 2010 The numbers above are for 14 drives (16 minus hotspares) in a RAID50 array, RAID5+0 means A RAID 5 set mirrored to an equal number of drives. In this case removing the spares leaves 14 drives, to mirror these equally you need to split that number in half leaving 7 drives in RAID5 one of which is used for parity leaving the capacity of 6 drives total for storage space. As long as you have Windows Server enterprise on your servers you can use clustered file system, there is no requirement for VM extentions for this feature. As this is all running on a server anyway it could be offered through a server as a SMB2 share from a SAN giving you the same level of support but greater abilities in future, some medium speced SAN will also offer SMB shares which would negate the need for the server. I don't think that there would be any performance gains from partitioning it up more. If you are set on the server solution perhaps Windows Storage Server may be a better fit as this offers SIFS (Single Instance File Storage) deduping files and saving space anlong with lots of other tools to manage the files better. 1
Midget Posted February 21, 2010 Posted February 21, 2010 synack i think youre thinking of a mirrored RAID5 (RAID 5+1?) not a striped RAID5 http://upload.wikimedia.org/wikipedia/commons/9/9d/RAID_50.png 1
tmcd35 Posted February 21, 2010 Author Posted February 21, 2010 The numbers above are for 14 drives (16 minus hotspares) in a RAID50 array, RAID5+0 means A RAID 5 set mirrored to an equal number of drives. In this case removing the spares leaves 14 drives, to mirror these equally you need to split that number in half leaving 7 drives in RAID5 one of which is used for parity leaving the capacity of 6 drives total for storage space. RAID-50 is two RAID-5 sets stripped. RAID-51 would be mirrored. RAID-50 has (almost) the same two disk redundancy as RAID-6. 1 redundant disk for each RAID-5 set. It has the same capacity as RAID-6, but the stripping should make it faster than RAID-6. In RAID-6 any two disks can fail. In RAID-50 1 disk from each set can fail (max 2 disks), but if 2 disks fail in the same set - As long as you have Windows Server enterprise on your servers you can use clustered file system, there is no requirement for VM extentions for this feature. I read somewhere that you can only store VM's in a CSV at the moment - can't think why there'd be that limitation. Also you need support for 'Persistant Reservations' which neither of the free SAN solutions (FreeNAS, Openfiler) currently supports. I must come clean and say that this is all based on me building the SAN/NAS solution rather than buying off the self. Reason being it's cheaper and I can get a lot more Tb for the money I don't think that there would be any performance gains from partitioning it up more. I think that was the basic question I had. Thanks. I think using just basic shared folders on 1 single volume is the way to go. If you are set on the server solution perhaps Windows Storage Server may be a better fit as this offers SIFS (Single Instance File Storage) deduping files and saving space anlong with lots of other tools to manage the files better. I looked at it and it's OEM only. I'd have to find a OEM solution like HP with less Tb for the money to use this. The only other option would be to use Windows Storage Server from Technet. MS might not be best pleased with that running on a live server.
Midget Posted February 21, 2010 Posted February 21, 2010 Also don't forget that a "450GB" harddrive is only 419GB so you'll only have about 4.9TB of storage (minus some overheads) RAID50 seems far too complex and introducing too many points of failure for my liking. Especially as you don't really have 2 redundant disks because if 2 disks in one of the RAID5s fails you've lost that array, and with it also being RAID0'd you'd recover the total sum of bugger all. I would go with RAID6
tmcd35 Posted February 21, 2010 Author Posted February 21, 2010 RAID50 seems far too complex and introducing too many points of failure for my liking. Especially as you don't really have 2 redundant disks because if 2 disks in one of the RAID5s fails you've lost that array, and with it also being RAID0'd you'd recover the total sum of bugger all. Aye, but is that not what the 2 hot spares are for and part of the reason we do DR backups of the data? Got to be extremely unlikely for a 2nd disk to go in the same set before the rebuild as completed. Possible, but worth the risk?
Midget Posted February 21, 2010 Posted February 21, 2010 I used to do data recovery, so I've seen many many RAID5 arrays where 2 disks have failed near to the same time. Due to experience once one disk in an array fails, I will be expecting the others to fail soon after so prepare for it. We had one where the 2nd disk failed during the rebuild causing LOTS of fun.
tmcd35 Posted February 21, 2010 Author Posted February 21, 2010 I suppose that's true regardless of which RAID I choose to use. I've built into to my budget plan an annual disk replacement strategy. I'll completely replace 3 or 4 disks in the array with new ones each year regardless of whether the disks themselves have failed. If they haven't failed then the old disks would be kept for emergancies - in case the hot spares are ever used.
SYNACK Posted February 21, 2010 Posted February 21, 2010 synack i think youre thinking of a mirrored RAID5 (RAID 5+1?) not a striped RAID5 Yeap, my bad I was thinking of that setup
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now