Is there any performance gain or loss when using DrivePool for gaming?

[Opifex]

Hi all,

I've got an old DrivePool license on a mini server that is never used so I was thinking of transferring to my Win11 machine.

I have 5 SSDs (ranging from 500GB to 2TB in size, all same speed 500MB read/write) and basically all of them store games among other rubbish. I figured how nice it would be to have one massive drive on "This PC".

I'm wondering does DrivePool have cause any issues if I have games on the pool? I can't tell if it could be beneficial like a RAID0 setup where performance is actually better in a pool, or if theres losses.

To be picky; im not really interested in "you wont notice a difference" etc. Like if loading Final Fantasy 14 on a DrivePool incurs a 50 millisecond increase in load time, thats still a performance loss.
I'd love to hear the numbers or metrics if there are any!

 

Thanks

[Christopher (Drashna)]

Shouldn't be. 

I actively game off of a pool, with no issues.  

[Jonibhoni]

Well, from my personal history: When I switched from 2xSSD in a Windows Storage Space to the same SSDs in a DrivePool I actually perceived a *really* noticable degradation in some edge cases. I had my development folder on that pool and suddenly some disk-intensive self-written Python programs took multiple times longer than before to initially read their data (hundreds of files with some GB total) from the disk. Unfortunately I don't have any numbers on that specific thing.

But I just did some new checks for you (conclusion/tldr at the bottom):

Test 1: Everything is done with

Quote

------------------------------------------------------------------------------
CrystalDiskMark 8.0.4 x64 (C) 2007-2021 hiyohiyo
                                  Crystal Dew World: https://crystalmark.info/
------------------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

Profile: Real
   Test: 8 GiB (x2)
   Mode: [Admin]
   Time: Measure 5 sec / Interval 5 sec
   Date: 2022/06/14 9:58:34
     OS: Windows 10  [10.0 Build 19044] (x64)

In theory, if DrivePool had no performance impact, there should be no difference between all test cases.

Test 1a: Drive R: = 10 GB test partition on a ADATA SX900 SSD

Quote

[Read]
  SEQ    1MiB (Q=  1, T= 1):   415.371 MB/s [    396.1 IOPS] <  2522.73 us>
  RND    4KiB (Q=  1, T= 1):    19.408 MB/s [   4738.3 IOPS] <   210.56 us>

[Write]
  SEQ    1MiB (Q=  1, T= 1):   176.845 MB/s [    168.7 IOPS] <  5923.53 us>
  RND    4KiB (Q=  1, T= 1):    83.046 MB/s [  20274.9 IOPS] <    49.02 us>

Test 1b: Drive G: = DrivePool consisting of just the partition R.

Quote

[Read]
  SEQ    1MiB (Q=  1, T= 1):   412.134 MB/s [    393.0 IOPS] <  2540.07 us>
  RND    4KiB (Q=  1, T= 1):    19.058 MB/s [   4652.8 IOPS] <   214.38 us>

[Write]
  SEQ    1MiB (Q=  1, T= 1):   205.610 MB/s [    196.1 IOPS] <  5092.97 us>
  RND    4KiB (Q=  1, T= 1):    74.665 MB/s [  18228.8 IOPS] <    54.53 us>

Test 1c: Drive G: = DrivePool consisting of the partition R AND an empty 300 GB partition on a much faster Kingston SKC1000 NVM SSD, but with the latter being disabled for use in pool by the drive usage limiter balancer.

Quote

[Read]
  SEQ    1MiB (Q=  1, T= 1):   415.326 MB/s [    396.1 IOPS] <  2521.62 us>
  RND    4KiB (Q=  1, T= 1):    18.949 MB/s [   4626.2 IOPS] <   215.62 us>

[Write]
  SEQ    1MiB (Q=  1, T= 1):   188.073 MB/s [    179.4 IOPS] <  5565.96 us>
  RND    4KiB (Q=  1, T= 1):    73.516 MB/s [  17948.2 IOPS] <    55.38 us>

(Why test 1c? - To see if the managment overhead of having two SSDs in the same pool has an impact, with the data actually still written on the same drive as in test 1b.)

While there is a degradation noticable in random access reads and writes when switching from the bare SSD to a pool with that SSD (and optionally other SSDs), the performance drop is almost non-existent.

I was a bit puzzled how I experienced the performance drop on my development folder when I originally switched to DrivePool. So I did another test with an additional HDD in my test pool, but again, which was disabled via balancing settings. Just to see if it makes an impact when having SSD and HDD mixed in a pool. The results were about the same.

I then discovered that CrystalDiskMark does all the benchmarking via a single large file, so the overhead of managing files over various disks might not really be tested there. So I decided to just do another real-world test with data I know.

Test 2: CRC32 calculation via 7zip of 12231 files in 1701 folders with an overall size of 6.38 GB, with Windows read cache flushed before each test.

Test 2a: Drive R: = 10 GB test partition on a ADATA SX900 SSD:
Total time: 25 seconds

Test 2b: DrivePool consisting of just the partition R:
Total time: 42 seconds

Test 2c: DrivePool consisting of the partition R AND an empty 300 GB partition on a much faster Kingston SKC1000 NVM SSD, but with the latter being disabled for use in pool by the drive usage limiter balancer:
Total time: 42 seconds

Test 2d: DrivePool consisting of the partition R AND an empty 300 GB partition on a much faster Kingston SKC1000 NVM SSD AND an empty 5 GB partition of a Seagate ST2000DM005 HDD, but with the latter two being disabled for use in pool by the drive usage limiter balancer:
Total time: 43 seconds

Test 2e: DrivePool consisting of the partition R AND my 80% filled Samsung SSD 850 Pro 500 GB software partition , but with the latter being disabled for use in pool by the drive usage limiter balancer:
Total time: 43 seconds

Test 2f: The same files on another partition on the same disk (ADATA SX900) as part of my real in-use DrivePool consisting of 2 SSDs and few HDDs with a total of ~2 TB of mixed size data:
Total time: 46 seconds

---------------------------------------------------

Conclusion:

With my little test setup, the time to process (read) many small files is roughly doubled if they are accessed via a drive pool (Test 2).
For larger files the performance was roughly comparable with and without drive pool, regardless if sequential or random access (Test 1).

Edited June 14, 2022 by Jonibhoni
Added Test 2f

[bobloadmire]

16 hours ago, Christopher (Drashna) said:

Shouldn't be. 

I actively game off of a pool, with no issues.  

Would there be any benefit to duplicating game data so you can do striped reads? optimal amount of duplications?