Performance Test: Create and Measure Performance of Storage Spaces Pool 3 Virtual HDDs in WSE2012R2P

[Joe_Miner]

Here's a video I created to show the creation of a Storage Spaces Pool and then the measuring of the performance of three Virtual HDDs created in that Pool: 

 

 

 

It's also at link: youtube.com/watch?v=IF-bCzsZe0A

[ikon]

Cool JM. And, who knows, if Storage Spaces winds up on the tech trash heap, maybe the video will become a collectors' item

[pcdoc]

Nice video and nice work on the topic.  I sure do not like SS but you did a great job on the how to.  I assume you did this on Server 2012 not essentials since you were using virtual drives.

[Joe_Miner]

Nice video and nice work on the topic.  I sure do not like SS but you did a great job on the how to.  I assume you did this on Server 2012 not essentials since you were using virtual drives.

 

Thanks PCDoc!  It was a learning experience doing the Video.  I think SS has a lot of potential down the road but if you're looking at Parity Resiliency on a Virtual HDD created out of the SSPool the Write performance is going to be "rather" dismal.

 

I actually did this in Server 2012 Essentials R2 Preview -- I suppose the confusion lies in my attempt to emphasize that the HDD's created out of the SSPool are really Virtual HDD's which was why I did the chart:

 

 

to illustrate that with Thin Provisioning the Virtual HDD's created out of the physical HDD pool could have substantially greater "potential" capacity -- very similar to what we see with VHD's in Hyper-V

 

In the test system I did this on I do have Hyper-V up and running and am running a VM of Win8.1 on it.

[timekills]

Nice video. If I might add some constructive criticism, I found myself going back and forth in the video to compare the benchmarks. If you could post a screen with all three drive types using one of the benchmarks (and maybe even with the bare drive's benchmark) it would help in the comparison.

[Joe_Miner]

That's an excellent idea.  Thanks TimeKills!

 

 

 

 

 

 

Edited August 20, 2013 by Joe_Miner

[Joe_Miner]

This is also blogged at Creating and Measuring the Performance of a Storage Space Pool in Windows Server 2012 Essentials R2 Preview

[timekills]

Joe:

Great comparo diagram. Minus the lower than one would expect write speeds for the parity RAID (well, lower than one would expect if they hadn't seen earlier Storeage Space speeds...) it falls in line fairly well. The mirror read speeds with three drives means you're going to get read benefits from using two drives simultaneously at best, while they RAID5-type (parity) means you're typically getting at least read access from two drives. The max read therefore is the same, but the parity will pull ahead in the average sequential reads - as shown in your tests. Of course, getting to read and write to all three drives at all times (simple/striped/RAID0) is going to give the best speeds and it seems to bear out as the read and write speeds are fairly linear and just about 3x the speeds from using one drive.

 

Looks like the biggest challenge right now is the parity calculations on the writes. If MS can improve on that it will be a hit. I wonder how much of it has to do with calculations affected by the ability to overprovision the drives.

[pcdoc]

Joe:

Create comparo diagram. Minus the lower than one would expect write speeds for the parity RAID (well, lower than one would expect if they hadn't seen earlier Storeage Space speeds...) it falls in line fairly well. The mirror read speeds with three drives means you're going to get read benefits from using two drives simultaneously at best, while they RAID5-type (parity) means you're typically getting at least read access from two drives. The max read therefore is the same, but the parity will pull ahead in the average sequential reads - as shown in your tests. Of course, getting to read and write to all three drives at all times (simple/striped/RAID0) is going to give the best speeds and it seems to bear out as the read and write speeds are fairly linear and just about 3x the speeds from using one drive.

 

Looks like the biggest challenge right now is the parity calculations on the writes. If MS can improve on that it will be a hit. I wonder how much of it has to do with calculations affected by the ability to overprovision the drives.

 

One addtion to your analysis is that the numbers that he saw on the parity writes were a bit skewed by the use of VHDX.  In my last round of testing, I found the performance of parity on physical drives to be slower than VHDx especially on benchmarks.  I assume due to some memory caching though I am guessing at that statement.  I still hold to the real life usage of the parity writes in storage spaces (16 megs/s) does not work for most people and certainly not for business.  Even the 44 megs/s for 3 10K drives is pretty bad.  It tells me that the efficientcy of SS is no where near what is should be.

[Joe_Miner]

One addtion to your analysis is that the numbers that he saw on the parity writes were a bit skewed by the use of VHDX.  In my last round of testing, I found the performance of parity on physical drives to be slower than VHDx especially on benchmarks.  I assume due to some memory caching though I am guessing at that statement.  I still hold to the real life usage of the parity writes in storage spaces (16 megs/s) does not work for most people and certainly not for business.  Even the 44 megs/s for 3 10K drives is pretty bad.  It tells me that the efficientcy of SS is no where near what is should be.

 

I'm not sure I'm following your point.  

 

I created a Parity drive in Storage Spaces based on creating a volume (virtual disk) out of the Storage Spaces Pool made of Physical HDD's (which can be expanded/added to in the future).  I'm unclear how else it should be done.  I don't recall it being called a VHD or VHDX but rather Virtual Disks.

 

While they are 5th generation 10K drives in this test -- the sequential read/writes are more what I'm seeing with a WD Black's & Red's (w/64MB caches) and less than a Seagate Baracuda ST3000DM001 (w/64MB OptiCache).  The 10K's do have a distinct advantage with much better random access which may help.  If you check my Table 1 summary results in Storage Spaces Performance in Windows Server 2012 Essentials on a HP N40L ProLiant MicroServer Drive K: (a 1TB, thin, Parity, NTFS) gave me ATTO read/writes of 463/37 [interesting: Parity Drive J: (1TB, Thin, Parity, ReFS) gave me ATTO read/writes of 467/46 ] -- and that was with a physical pool underlying SS of 5 ST3000DM001's which typically have transfer rates of ~190-200.

Edited August 21, 2013 by Joe_Miner