asked on

HP DL380 G8 and P2000 configuration options

Hi, we are in process of implementing dynamics ax 2012r2 in our office and our hardware is:

2 x HP Proliant DL380 G8 1 x HP P2000 with 18 15k SAS Disks 2 x 10gbe Switches

We will be doing a straight forward iSCSI 10gbe setup with Hyper-V and windows server 2012 R2 and wanted some advice on the RAID config. Both the SQL server and the AOS server will be virtualised.

What would be the ideal setup for the P2000?

Initially we started with;

2 Disks for SQL installation (Single VHD) - LUN 10

4 Disks for Database Files (Single vhd) - LUN 20

4 Disks for TLogs (Single VHD) - LUN 30

4 Disks for TempDB (Single VHD) - LUN 40

4 Disks for the AOS server (Single VHD) - LUN 50

Or would OBR10 be more appropriate with 5 separate volumes each with a LUN.

Thanks.

ASKER CERTIFIED SOLUTION

Member_2_231077

membership

This solution is only available to members.

To access this solution, you must be a member of Experts Exchange.

Start Free Trial

dqnet

ASKER

So would you recommend one array 10 disks in Raid 10 for the SQL server and 8 disks in one raid 10 array for the aos server? Then create 4 luns each for the required VHDx's?

--10 disks--
SQL install vhd Lun 1
Database vhd Lun 2
Logs vhd Lun 3
Tempdb vhd Lun 4

-- 8 disks --
Aos Lun 5

I thought a small raid 10 with 4 disks each would provide more performance?

Member_2_231077

I would not put the SQL logs on the same set of spindles as the data, they'll be fairly small so I'd put them on the same disks as AOS.

The more disks in an array the more IOPS available, it's just about linear, 8 disks give twice the performance of 4. You are running twice as much on those 8 disks than on the 4 of course but twice the performance with twice the load should give identical performance either way. But then if one of the VMs isn't busy you get the performance improvement of having more disks.

Let's say you calculate 10 disks needed for application A and 4 for application B, and you assign those disks like that. If you got the calculations right it works well but only very slightly better than if you put all 14 in one array and made two LUNs from it. But what if you got the sums wrong or the situation changes? It's very hard to re-assign the disks in the first instance but the sharing situation automatically lets application B have the IOPS if A doesn't use them. Of course the sharing method can be upset, create a small test LUN and run IOMeter on it and the others will slow to a crawl.

dqnet

ASKER

Thanks!

My only concern is the transaction log is mainly sequential writes and the database random it would be a little bit of an issue with IOPS and latency? Would a mirror RAID 1 just for the TLogs be more appropriate and then increase the RAID 10 to 8 + 8 then buy another spare?

Member_2_231077

You could get an extra pair of disks for that but it doesn't really matter if you do put the transaction logs on something that's mainly random I/O, you just don't get the benefit of it being sequential since it gets chopped up with the rest and ends up random. You wouldn't need an extra spare if they were the same disks as the others since spares can be global so usable by any vdisk.

dqnet

ASKER

Sorry for the delayed answer mate.

We purchased 2 extra disks and are going to operate on this setup, what do you think?

Database 8 disks in raid 10
Tlogs 6 disks in raid 10
Aos server 4 disks in raid 10
Tempdb and sql install 2 disks in raid 1

What you think?

We are now testing strip sizes of 64kb and 128kb at raid level.. Any advice on that?

Thanks mate

Member_2_231077

Sounds fine to me although I'd still share spindles instead of having dedicated disk sets but that's probably just because I'm no good at tuning.

The optimal strip (stripe element) size the database's native block size, e.g. 64K with SQL. A larger strip size will degrade performance by a tiny bit, normally not noticeable. A smaller strip size will degrade performance a lot since two disks will be tied up doing a single database block read.

dqnet

ASKER

Thanks!