2

I recently provisioned a new Windows Server 2012 R2 server at my hosting company. This is my first time using an SSD as my hard drive and the system is currently utilizing a single 800Gb drive (split into 2 partitions). I noticed immediately that Windows is not recognizing the drive as a "Solid State" drive and instead sees it listed as a standard drive.

When I view the disk drive in Device manager I see it listed as "Adaptec Array SCSI Disk Device". When I asked the hosting company about this, they said they connect the drive to a RAID controller for "performance reasons".

Is this true? I can't understand why there would be a performance boost from a RAID controller with only a single drive. Additionally, I feel like the consequences of Windows improperly performing defrag on my mislabeled SSD would cause more problems down the road.

I would be curious to hear your thoughts on this.

Update 10/2/2016

I ran benchmarks using CrystalDiskMark on the server while using the RAID controller and after removing the controller and instead having the same drive directly connected to the motherboard using one of the 6GB/s SATA ports. Here are the results:

With RAID Controller

Connected to Motherboard SATA

As you can see by having the drive connected via RAID controller (even though I'm not using a RAID configuration indeed does improve performance. My theory about this is that the controller must somehow alleviate some of the workload off of the CPU or the on-board SATA controller is just plain slow.

Dave L
  • 123
  • 5

1 Answers1

2

A single drive behind a hardware RAID controller can potentially benefit from a writeback cache, but it's very unlikely that this would actually benefit you.

If the RAID controller supports TRIM in a reasonable way, you should have nothing to be concerned about with your drive being behind an abstraction. TRIM also only becomes a mild concern when you're doing a LOT of changing writes and overwrites to the drive.

In general, a single RAID 0 to pass an SSD to an OS is a silly thing to do. HOWEVER, some RAID controllers do not provide JBOD "passthrough" of disks directly to the operating system. In these cases, a single disk RAID 0 is required to get the OS to see the disk underneath it. This is probably the case.

But for performance reasons? That's dubious at best. Perhaps you should ask for specifics and possible metrics on that for your workload if you're concerned with the statement itself. They should be able to provide you with such a thing, as its a hosting company that I'm assuming uses an at least semi-proven model for deployments.

EDIT:

On some server boards there is a real performance difference between the onboard SATA controllers and a dedicated board that works at a reasonable speed. If this is in question and performance is a large concern, performing tests on either is a reasonable course of action.

Spooler
  • 7,046
  • 18
  • 29
  • Thanks so much for the clear explanation. So if you were configuring a new server with a single SSD hard drive would you bypass the RAID controller altogether and simply connect the drive to the motherboard? – Dave L Oct 01 '16 at 16:17
  • If I was managing the server myself, yes. If I were a hosting company, I would want a data center tech to be able to swap the drive without having to pop open the server, and also make sure there are no special snowflakes in my datacenter with drives hidden in their chassis. – Spooler Oct 01 '16 at 16:20
  • Ah... that makes sense. So it could be argued that the main reason they use RAID controllers is so they could easily swap hard drive(s) on the fly? – Dave L Oct 01 '16 at 16:23
  • Update: I just heard back from the hosting company. The rep says: "Typically the SATA controllers on the motherboard are unable to support the read/write speeds that SSDs are capable of, which is why we provision SSDs on a RAID controller by default. We can, by request, remove the RAID controller and put the SSD on the on board SATA controller but you may experience performance loss with the drive" I've never heard this before... is this true? – Dave L Oct 01 '16 at 17:10
  • I've never had this issue in any of the several thousand machines I've managed. We decided to put SSDs on the motherboard of several dozen machines for use as a bcache devices. This was fine, but we stopped because it was a pain to pop open the chassis when the SSDs died. I can see this being a problem possibly if the motherboard SATA controller is gen 2 rather than gen 3, but that would be an older server - and maybe it is. Only other argument I can see them making is that the RAID controller performs offload, but the CPU is more than capable of handling far more than one SSD without a hitch. – Spooler Oct 01 '16 at 18:22
  • Thanks for all the helpful info. I updated my original post in case you're curious with benchmark results with the RAID controller and without. It looks like the drive performs the benchmark tests better when connected to the RAID controller on this particular server. – Dave L Oct 02 '16 at 17:13
  • How interesting. What server is it, exactly? – Spooler Oct 02 '16 at 17:15
  • I was surprised as well! Here's the system configuration: http://i.imgur.com/A0RA09f.png I have to now decide whether I should have them put the RAID controller back and just accept that Windows won't be able to identify the drive properly. – Dave L Oct 02 '16 at 17:18
  • You could just slap it into the RAID controller and either disable defragmentation on the drive or schedule it for only once a month. You won't get intelligent TRIM, but the performance difference is big enough between the controllers that this isn't remotely a concern. – Spooler Oct 02 '16 at 17:29