0

I have two servers of identical spec (bought at the same time), except for the drive configuration:

Server A:

  • C: and D: drives are partitions on the same 2 disk RAID 1
  • E: is a 2 disk RAID 0

Server B:

  • C: is a 2 disk RAID 1
  • D: is a 2 disk RAID 1
  • E: is a 4 disk RAID 10

Server A was set up for testing and so only had limited number of drives. Server B is to be used as a live system (and incidentally has more memory and an extra processor).

Compared to Server A, Server B has double the read performance on the E: drive. Other than that though, all the drives are performing worse than the drives on Server A (as tested using SQLIO). For write speed, all of the drives are writing at less than half the rate of Server A.

I'm not certain that the drives are from the same manufacturer, but they are of equivalent spec, so whilst I might expect a certain difference in speed due to manufacturer differences, I wouldn't expect the performance differences that I am seeing, particularly in write speed.

What could be causing this?

paulH
  • 183
  • 2
  • 3
  • 14
  • 1
    How is RAID configured? What RAID controller is used? Does it have write cache enabled? What disks are used? SATA/SAS?!? RPMs?!? – dsmsk80 Sep 24 '13 at 11:03
  • I'm not sure which RAID controller is used. I know the same one was used on both servers and the configuration was left with default values on both servers. All disks are 300GB SATA 15k. Write cache I have no idea. If it is part of the RAID configuration then I'm sure it's the same on both servers. If it's an operating system setting then I should be able to check that somehow? – paulH Sep 24 '13 at 11:13
  • I should add that the 'slow' Server B is currently not being used by anything at all - I'm still in the process of setting it up - so I don't expect the performance is being affected by anything running on the server. – paulH Sep 24 '13 at 11:15
  • Please find out what RAID controller is used in both servers. You should be able to identify it in device manager or during servers boot up. Btw, how did you configured RAID10 on those disks?!? Did you use some tool? Perhaps, server model would help to identify it as well ... Just to be sure, are we talking about physical servers? – dsmsk80 Sep 24 '13 at 11:24
  • The controller is a PERC H700 integrated. The servers are Dell PowerEdge R510's and yes they are physical. I don't know specifically how the configuration was done. The building of the server is not something I am directly involved in, but I'm assured that they were built in exactly the same way other than the differences described above. If it helps to limit the discussion I'd be happy for now to ignore the E: drive and concentrate on how the D: drive of the servers could exhibit such differences in performance given that they are both 2-disk RAID 1. – paulH Sep 24 '13 at 14:56
  • 1
    This hba is equipped with battery-backed write cache by default. I would verify the current state of the battery as the common policy is to disable the cache when the battery is charging or when it is faulty. Do you have Dell OMSA installed? It's a management system for Dell servers. Either try to connect to this system with web browser, there should be available an web interface running on the server and listening on the port 1311 or there should be installed CLI tools in c:\Program Files\SysMgt\oma\bin\. Use the command "omreport storage battery controller=0 battery=0" to check the battery. – dsmsk80 Sep 24 '13 at 16:53
  • That turned out to be a *very* useful command. The battery state was 'charging' when I first checked, but five minutes later it had changed to 'ready'. That didn't affect the write speed though. But what I did find, by looking at what other bits of information I could glean from the omreport command, is that the slow server had 3Gbps drives, and the fast server has 6Gbps drives! Case solved and I can go back to the server guys and ask them to re-think their definition of "equal specification" :-) – paulH Sep 25 '13 at 09:39
  • @dsmsk80 - if you care to add your comments as an answer then I'll give you the credit for the answer. – paulH Oct 22 '13 at 16:49

0 Answers0