1

I had asked this question earlier and the question went missing so here it is again.

Bought a DELL Poweredge 2950 to use as in-house QA Server. Disk performance is beyond terrible, 1000-4000 ms response time on the drive with our SQL Server database .mdf. Sql Server disk queue upwards of 300 at times.

I'm a software guy, can anyone help me with steps to determine the issue?

I don't know what RAID controller it has, how can I determine that?

I'm speculating it could be BIOS issue. Perhaps the server used to have another kind of drive in it and when I added SATA the ??? buffer size is wrong???

Perhaps I chose wrong options (chose defaults) when setting up the RAID 1 arrays? I thought RAID 1 was a performance array?

OBSERVATION: Any disk activity on the SERVER causes the response time on another drive to drop, for instance, I was copying the DELL open manage ISO to the C: drive and I noticed the response time on the F: drive (RAID 1) different LUN, spiked up during the copy ??

Tom DeMille
  • 239
  • 3
  • 8
  • How much load are you putting through this build server? What type of RAID controller is in the box? – ITHedgeHog Feb 15 '11 at 15:20
  • You'll need to install DELL OpenManage, it should tell you what kind of RAID it has, if it is a software RAID (using windows server) then this is the problem. also the RAID Controller is very important. – miro23 Feb 15 '11 at 15:20
  • I can't find openmanage for 2008 x64, is it not supported? – Tom DeMille Feb 15 '11 at 18:37

2 Answers2

4

1)Install Dell OpenManage

As it was mentioned above install Dell OpenManage - its an administrative application that runs in Windows and allows you to look @ the hardware health/firmware levels. You will want to go to support.dell.com. Enter either the ST (service tag) of the server, which should be posted on the front bezel area or the back. Alternatively you can just enter PE 2950 and it will show you all the available firmware/software. Select the OS you are running in the drop-down and you will get the correct list.

2)Download the firmware updates that are out there, check the most current level vs what you have installed (view in OpenManage).

3)There has been an issue w/ some PERC card batteries (the raid controller) in the 2950s that we have had lately. This sounds like exactly what we were seeing. Major degradation in DB reads/writes. If you install Dell OpenManage you can also see this. Go to the logs in OpenMange and look @ the battery. You may see it doing its quarterly discharge more often. When this happens, in our case 1x a day, you raid controller goes to write through vs write back. This will wreck your DB performance and is due to a bad PERC card battery. This will be very easy to spot in OpenMange.

Chadddada
  • 1,680
  • 1
  • 19
  • 26
  • can't find openmanage for x64 server 2008, is it supported? – Tom DeMille Feb 15 '11 at 19:36
  • Battery if fine. The PERC5 i controller is ANCIENT, it's 2.0x and minimum required is 2.2 and current is like 5.5 so I'm trying to flash that now and see what happens – Tom DeMille Feb 15 '11 at 19:37
1

Well RAID1 is just a mirror. It doesn't really improve the performance over a single drive, maybe very slightly for reads. RAID0 is the only option for improving performance if you only have 2 drives. However I wouldn't recommend it as there's no redundancy there and is more prone to data loss than a single drive. If you have at least 4 drives you can try using RAID10 if your controller supports it. But in reality the number and type of drives are always going to dictate the number of IOPS that your storage system is capable of. Tuning your workloads etc can buy you a bit but you're normally talking a few percentage points at that point.

Without more information I'd say you're likely just seeing the limitations of your hardware.

EDIT: Also if your workload is heavy write then a battery backed up cache on the RAID controller and setting it's mode to Write-Back may improve the situation considerably.

3dinfluence
  • 12,449
  • 2
  • 28
  • 41
  • I don't believe we are hitting the limits of the hardware. One guy testing, from one web page, making one call. Or creating one index. The same operations on my development machine are 20 times faster. – Tom DeMille Feb 15 '11 at 15:53