Configuring a RAID for clustering: One large array or many small arrays?

The setup:

I have two servers and one RAID array.  I am building an active/active SQL 2008 cluster meaning that at all times there will be two instance of SQL running on this cluster.  In case of a server failure both instances of SQL Server will be running on the one remaining server.  

This means I need seven disk letters:
Two for data files
Two for log files
Two for tempdb.
One for quorum

2 x Dell R710 with Perc 5/e
1 x MD3000 w/ 15 x 136GB

The problem:
I am unsure which way to setup the disks for a two node active-active cluster.

1) Do I create one large 14-disk RAID 10 and then carve out the needed 7 drive letters?

2) Do I create 2 2-disk RAID 1 (Logs) and 2 5-disk RAID 5 (Data) then carve out the needed 7 drive letters? (The quorum and tempdb drives will be carved out of a log disk).

My concern with #1 is that both servers will be reading and writing to the same RAID 10 array and will bog it down.  What if server A is writing and server B is reading?  If they each get their own array then they won't compete-- theoretically.

Ideas?
SteveMBOAsked:
Who is Participating?
 
thribhuConnect With a Mentor Commented:

You can use diff combinations , but remember the below standard
1) Data disk should be Raid 1+0 (Best if more writes) or Raid 5 (4+1) better if more reads.
2) Log disk should be Raid 1 prefered for best write performance as usal.
3) Temp disk can be any but better of raid 1

Senario 1:
Data disks -> Raid 1+0(4+4) disks. I prefer two raid groups with 2+2 disks.One for raid group for one drive. you have advantages on this.

In Third raid group(Raid1)(1+1) make one disk for log and one disk for tmp.
In fourth rais group(Raid1)(1+1) make one disk for log and one disk for tmp.

If you do this you will have redendency and better performance.
Rest 3 disks can be used Quoram Hot spare etc.

Senario 2:
If you have more reads tha writes in Data disks then below is good:
Total disks (15 X 136GB)

RG1:(4+1 = 5disks)
First Data Disk -> Raid 5 (4+1)
RG2:(4+1 = 5disks)
Second Data Disk ->Raid 5 (4+1)

RG3 (1+1)
Two Log Disks and -> Raid 1(1+1)

RG4(1+1)
Two Temp Disks -> Raid 1(1+1)

If you want you can carve single raid group x with 2+2 disks for two log and two temp.

The remaining disk can be used for Quoram or Hot spare.

Let me know if any thing more on this............
0
 
SteveMBOAuthor Commented:
Thinking about it, because two servers will each be talking to the MD3000 through a controller and you can't have two controllers talking to the same set of disk... doesn't that mean I have to have at minimum 2 sets of disk?
0
 
tigin44Commented:
current disk controller do access many disk at the same time.. To increase the thrughput of disk subsytem it will be better to group disk for each of the letters you need except the quorum. You can define the quorum as a partition on one of the disks my choise will be the log file disk..

2 x 2 disk raid1 for log
2 x 1 disk raid1 for temp
2 x 4 disk raid 1+0 for data

0
 
SteveMBOAuthor Commented:
Thanks, both of you, for your help.  This solution closely matched what Dells documentation (that I just found) said.

http://i.dell.com/sites/content/business/solutions/power/en/Documents/Configuring_Microsoft_SQL_Server_2005_on_Dell%20Servers_and_Storage.pdf
0
 
thribhuCommented:
WOW, i am proud , i gave a solution that is coated by Dell - coool , i can be a storage architech then.
My second senario is coated by Dell. But this is good when you have more reads .

Raid (1+0) is Best for write performance .

Any way thanks for the Points SteveMBO.
0
All Courses

From novice to tech pro — start learning today.