asked on

RAID 10 performance on PE2900 w/ 8 - 300GB SAS 10k drives / Perc 5i

We recently set up a vmware esxi free edition 3.5 host for our Exchange server an SQL server

I wanted to share my hdd speed tests to see if the speed is correct or similar to what it should be. I have installed a couple other RAID 10 arrays and when testing noticed the graph stays pretty even all the way across.

This implementation seems to Vary widely from over 900Mbps to under 100Mbps.

Is this normal? What are my next steps for troubleshooting?

Below is my configuration, let me know if you need any more info.

- Dell Poweredge 2900
- ESXi 3.5
- Dual Quad core processors
- 20 GB memory
- Perc 5i
- Adaptive read ahead
- Write back
- RAID 10 with 8 drives
- 300GB 10K SAS (Varying manufacturers)

-Two clients
-      Windows Server 2003 (800GB total)
o      Exchange
o      File Shares
-      Windows Server 2003 (200GB total)
o      SQL Server

Im pretty sure the stripe size is 64kb, I did not change from default.

pgm554

You should check this out:

http://msdn.microsoft.com/en-us/library/dd758814%28v=sql.100%29.aspx

If you are running SQL server on 2003 or below,you should probably do a partition alignment.

Needs to be done before the OS install.

SynsorIT

ASKER

I converted the Win2k3 server with Exchange over to the host and ran this test before doing anything else and got the same result.

pgm554

You need to align the partitions before the install,if all you did was convert or import the physical machine and this was never done to the physical machine to begin with when setting up the partitions,you will take up to a 30% hit in i/o performance.

http://www.vmware.com/pdf/esx3_partition_align.pdf

SynsorIT

ASKER

My real question is "is this normal?". I.e. is this similar to what is expected out of a RAID 10 with 8 - 10k drives? Why are there these huge spikes and then lows?

I read the vmware article you referenced. It states that Average increase in throughput is 12% and latency decrease is 10%. The graph i included shows swings of 80% or so.

I did not align the disks so I can see how this could be an issue. However after just spending the whole weekend on doing this upgrade i dont think i want to start over from scratch for a while.

Is this something that can be caused by some other fault or is it normal and not a fault at all?

pgm554

You're dealing with virtual disks here and an older version of Vmware,so things like drivers for the disks are using an older API.

If you are using the free version,why not use the latest and greatest?

My best guest is that it is not normal.

I didn't know about the alignment issues until I went to an M$ SQL Saturday seminar a few weeks ago and the presenter talked about tweaking SQL on a SAN.

2008 out of the box doesn't have that issue.

Supposedly Paragon has an alignment tool that may help,but I haven't tried it.

http://www.paragon-software.com/home/partition-alignment/

SynsorIT

ASKER

Fair enough.

The only reason that i am using esxi 3.5 is that the Poweredge 2900 only shows that on VMware's HCL.

Would it be safe to upgrade to a newer version with the specs i mentioned above?
If, so how difficult is it to upgrade to a new version? Does it require that i move any of my VMs or change the main datastore?

I know this is out of the context of the original question and really appreciate the help.

ASKER CERTIFIED SOLUTION

pgm554

membership

This solution is only available to members.

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

Start Free Trial

SynsorIT

ASKER

At this point we are still using remote backup exec agents and backup up file level to a media server, like i said, we just converted two machines over the weekend.

I am in the process of evaluating backup solutions so we are just sticking with our old system for now.

I will attempt to move to esxi 5. Looks like I have to upgrade to 4 first then 5.

Thanks,

SynsorIT

ASKER

Upgrading to 4.1 made the problem is worse now...

I have opened a new question with the same title.