-5

HP's $1,000 Smart Array P822 controller is outclassed by an on-board raid controller. I'm having a hell of a time trying to figure this one out. My problem is that performance on the HP controller is significantly below what the drive is capable of, as demonstrated on the Dell computer and other benchmarks performed by others online. I realize that these are 3rd party drives being used in an HP server, however, I still expect performance to be comparable and not be different by a 50x factor. If I went out and purchased a HP SSD, what performance should I expect to see? Please advise on anything that you believe would be likely to cause this problem, general recommendations or any reasons why my expectations should be adjusted.

Please see the image here: Image Link

enter image description here

I have enabled "Physical Drive Write Cache State" on the controller and that helped a bit. The 4k reads and writes are still lacking. Any ideas? Here is a picture of the performance after the change. enter image description here

ewwhite
  • 194,921
  • 91
  • 434
  • 799
DrWhy
  • 11
  • 1

1 Answers1

9

You're not one for the scientific method, are you?

  • Different operating systems.
  • Different platform (server versus workstation).
  • Different CPU architecture.
  • Unsupported SSD on the RAID controller.
  • Raw SATA versus a disk connected to a RAID controller backplane.
  • The assumption that a RAID controller is supposed to be "faster".
  • Misconfigured settings on the server's storage array.
  • What were the controller and logical drive cache settings?
  • Did you try without SmartPath enabled?

What type of result were you expecting?

What were you trying to compare?

Community
  • 1
ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • I reject that hypotheses. Of course I am. :). To answer your questions, I'm expecting the performance of the drive on the HP to be comparable to the performance of the same drive on the Dell. Thanks for listing out some general causes. My biggest concern is that this is caused because I'm using an unsupported SSD. Should I expect that the SSD performance when using an HP SSD would be be comparable to the current performance of the Dell? – DrWhy Apr 15 '15 at 19:02
  • @DrWhy The performance of the SSD shouldn't be impacted by the use of the RAID controller. Did you try without SmartPath? What were the controller and logical drive cache settings? If you only care about using a single disk, use a dedicated HBA and skip the RAID controller. – ewwhite Apr 15 '15 at 19:11
  • I've tried smart path both enabled and disabled. There was not any notable difference. When I disabled smart path, I left caching disabled. – DrWhy Apr 15 '15 at 19:16
  • when smartpath was disabled and caching was enabled at 100% write, I got much better results, however, I determined that this is just the cache talking here and is not indicative of sustained and actual performance of the drive. – DrWhy Apr 15 '15 at 19:18
  • I want to avoid this becoming an X-Y question. What is your intended goal? If it's to fill an HP server with consumer SSDs, I'd say it was not a good idea. If it's to run a single SSD, the configuration you have isn't quite right. Use an HBA. If it's academic, and you're just curious about the disparity, there are a lot of reasons this may be the case. – ewwhite Apr 15 '15 at 19:19
  • Ultimately, I plan to have this drive this in a RAID1, but am just testing it as a single drive to eliminate that as a cause. I am open to moving to an LSI controller - do you know if they will work with the HP server? It's a ML350p Gen8 with 18 LFF drive bays. 14 of those drives are 6TB HP SAS drives. – DrWhy Apr 15 '15 at 19:21
  • My end goal is to use this server as a Virtual Host. These SSD disks will serve as the storage for OS virtual disks. The other 14x6TB HP SAS drives I mentioned will be for a large file server storage. I was planning to use RAID 6 and put an SSD in front it and use HP's smart cache. All the while, replicating this host to a standby host. – DrWhy Apr 15 '15 at 19:23
  • 99% it depend on the controller disabling the disk private DRAM cache. SSD (especially MLC ones) need a fast, local DRAM cache for efficient program/erase cycles. Note that I am _not_ referring to the RAID cache, but to the DRAM cache _inside_ your SSD. If you can, instruct the controller to left disk cache enabled and repeat the benchmarks, At the same time, take note that enabling disk cache on SSDs without power loss protection _can_ lead to data loss (it really depend on of the RAID controller and the SSD works together). – shodanshok Apr 15 '15 at 19:25
  • @shodanshok I enabled disk caching and got these results. [link](http://content.screencast.com/users/CASEIT/folders/Default/media/d46045d7-685c-4227-a90d-dd9712ab09d5/enabled%20disk%20cache.png) It certainly seemed to make a difference. It mainly helped the Seq and 512k writes. The 4k reads and writes are still lacking. – DrWhy Apr 15 '15 at 19:45
  • @DrWhy Remember to edit your question to incorporate the additional information. Comments are meant to be temporary, and they are not always presented to site users anyway depending on how they read the site. – Michael Hampton Apr 15 '15 at 20:29
  • @ewwhite from reading your other posts, it appears that you have a lot of experience with hp smart arrays. Can you tell me what results I should expect to see with a supported HP SSD drive? – DrWhy Apr 16 '15 at 16:39