SQL Server 2012 Availability Groups (is RAID 0 OK for logs?)

Question

We are planning upgrade from our current 2008 R2 solution which has 2 instances running on a 2-node cluster.  The instances 'face' each other, so that we leverage both hardware, even though both sets of hardware are adequate to run both instances together.

One of our requirements for the 2012 solutions is to upgrade to Enterprise, and take advantage of Availability Groups to create a passive replica in our D-R site.  And plan for a SQL BI solution, which will leverage Analysis Services.  Currently we just use SAN replication to move our backups to the D-R site, but don't maintain an installed server there.  And we have Analysis Services running in a test environment only.  This is going to change.

We are feeling like we want to use just Availability Groups, and lose the failover cluster, since we get basically the same protection with a synchronized write.  This will also make things more consistent and simple for integrating our D-R site as an async replica.

Question I am wondering about is this:  We have a DAS setup (Dell MD3220) with 24 x 146 GB 15K drives for our SQL storage.  Wondering how to best divide this up. 

Currently, we break this up into separate RAID 1 (2 disk) for system dbs and SQL logs,  RAID 10 (4disk) for DB Logs, and RAID 10 (6disk) for DB Databases (3 logical disks for each instance) x2 instances = 6 logical out of the 24 physical disks.

We do have ample room for doubling our storage requirement with the same hardware, so our plan would be to use the same setup, but separate the disks, as they will not be part of the cluster config.  We would use our SAN instead for a witness disk as we already do for the WSFC config.

Here are the 2 thoughts I currently have:

A. Do the same and just split it between servers (not as WSFC resource): RAID 1 (2 disk), RAID 10 (4 disk) db log, RAID 10 (6 disk) db data, and place Analysis Data on the db data disk.

B. Then I think about the log write time (as a split transaction) between the 2 servers, and the fact they are replicas anyway.   I wonder if we should not consider the following: 4 logical disks: RAID 1 (2 disk), RAID 0 (2 disk) db log, RAID 10 (6 disk) db data, RIAD 1 (2 disk) analysis data.

Anyone have thoughts about: 1. Should we break up the analysis disks load from the transactional stuff?  2. Is it reasonable to use RAID 0 when we have a synchronized replica for logs to improve write-performance.

Thanks in advance for any wisdom,

Ryan

---

Ryan

Answer 1

We are planning upgrade from our current 2008 R2 solution which has 2 instances running on a 2-node cluster.  The instances 'face' each other, so that we leverage both hardware, even though both sets of hardware are adequate to run both instances together.

One of our requirements for the 2012 solutions is to upgrade to Enterprise, and take advantage of Availability Groups to create a passive replica in our D-R site.  And plan for a SQL BI solution, which will leverage Analysis Services.  Currently we just use SAN replication to move our backups to the D-R site, but don't maintain an installed server there.  And we have Analysis Services running in a test environment only.  This is going to change.

We are feeling like we want to use just Availability Groups, and lose the failover cluster, since we get basically the same protection with a synchronized write.  This will also make things more consistent and simple for integrating our D-R site as an async replica.

Question I am wondering about is this:  We have a DAS setup (Dell MD3220) with 24 x 146 GB 15K drives for our SQL storage.  Wondering how to best divide this up. 

Currently, we break this up into separate RAID 1 (2 disk) for system dbs and SQL logs,  RAID 10 (4disk) for DB Logs, and RAID 10 (6disk) for DB Databases (3 logical disks for each instance) x2 instances = 6 logical out of the 24 physical disks.

We do have ample room for doubling our storage requirement with the same hardware, so our plan would be to use the same setup, but separate the disks, as they will not be part of the cluster config.  We would use our SAN instead for a witness disk as we already do for the WSFC config.

Here are the 2 thoughts I currently have:

A. Do the same and just split it between servers (not as WSFC resource): RAID 1 (2 disk), RAID 10 (4 disk) db log, RAID 10 (6 disk) db data, and place Analysis Data on the db data disk.

B. Then I think about the log write time (as a split transaction) between the 2 servers, and the fact they are replicas anyway.   I wonder if we should not consider the following: 4 logical disks: RAID 1 (2 disk), RAID 0 (2 disk) db log, RAID 10 (6 disk) db data, RIAD 1 (2 disk) analysis data.

Anyone have thoughts about: 1. Should we break up the analysis disks load from the transactional stuff?  2. Is it reasonable to use RAID 0 when we have a synchronized replica for logs to improve write-performance.

Thanks in advance for any wisdom,

Ryan

---

Ryan

Answer 2

Hi Rvan,

Thanks for your post!
I hope some who familiar with this issue could help you.

Thanks,
Maggie

---

Please remember to mark the replies as answers if they help and unmark them if they provide no help. This can be beneficial to other community members reading the thread.

Answer 3

Hi!

Technically the first thing you should ensure with the config is that only 1 LDF file is placed on the volume if you are looking for speed. The reason behind it is that LDFs are written sequentially, which is by far faster on a disk than random IOs. If you have more than one file though the sequential IO is lost becasue the disk has to jump between the files...

Regarding the disk layout RAID 0 is for sure feasible. I would not use it though unless you have performance problems. If you loose one disk in a RAID 0 you lose the volume, which means that (after replacing the disk) you need to recreate all the databases on that server... That's a lot of work... And disks do tend to fail...

Lucifer