New SQL Server Disk Config

[Aknip]

I'm in the process of reconfiguring the arrays on an SQL server. I'm going for top performance and wanted other users opinions.

The software vendor requires the logical drive letters to be as follows:
C - OS
E - SQL transaction log
F - Data Download
H - SQL Database
I - SQL Database
G - Data Files

The software vendor also recommends the block size be 8K for the Database, Log, and Download drives; and 16K for the G(data files) drive.

The current SQL Database is 30+GB & the data on the G drive is currently up to 800GB.
With the current config the first RAID controller has a RAID10 array which contains the C, E, F, H, and I logical drives. The second RAID controller has a RAID5 array which is the G drive. Both the RAID10 and RAID5 currently have a 64K stripe size.

For the reconfigured server I was thinking about using the following configuration:
Controller 1: RAID1(64K stripe) for the C(OS) and F(Download) drives, RAID0(8K stripe) for the E(log) drive.
Controller 2: RAID10(8K stripe) for the H(database) and I(database) drives.
Controller 3: Two RAID5(16K stripe) arrays for G and add logical drive J. There would be 8 drives in each of these RAID5 arrays.

Is it too much to put 3 RAID controllers in one box? It's possible I can put the "G" array in another box.

Thanks for any input!

 

[mrdenny]

You can put however many raid controllers you need into a single server.

Never use RAID 0 for a production server. It offers no redundancy. If you need the speed of RAID 0 use RAID 0+1.

SQL Databases go fine on RAID 5 arrays. Here's a FAQ on disk setups faq962-5747.

Denny
MCSA (2003) / MCDBA (SQL 2000)

--Anything is possible. All it takes is a little research. (Me)

http://www.mrdenny.com

(Not quite so old any more.)

 

[Aknip]

The RAID0 would be used for the transaction log. If we're doing regular database backups and we would lose the transaction log drive, couldn't we just restore from the last backup after replacing the failed drive? I considered doing a RAID1, but thought the performance of the RAID0 would be better for the log. As far as the RAID 0+1, RAID10 (striped mirring) is far better.

http://www.microsoft.com/technet/prodtechnol/sql/2000/deploy/harag04.mspx

 

[mrdenny]

The problem with using the RAID 0 is that while you will only loose the transaction log if there is a failure, the database is totally unavailable until the RAID 0 is repaired, and the database is restored.

With a database of the size you are talking about you are talking about quit a lot of down time.

Denny
MCSA (2003) / MCDBA (SQL 2000)

--Anything is possible. All it takes is a little research. (Me)

http://www.mrdenny.com

(Not quite so old any more.)

 

[Corran007]

Im curious. why do you have two database drives? are you running two seperate instances?

 

[Aknip]

Actually I think that can be changed back to one drive. On the old server we were able to roll the database to another disk if it got too big for the drive.