SOLVED

Storage performance when using parity

Copper Contributor

Hello, Storage Team!  I am a consultant (and Microsoft MVP) that sees Hyper-V Replica, Storage Spaces and Storages Spaces Direct as great solutions for my SMB (the protocol and small- to mid-sized businesses). :)

 

The major issue I've run into is storage performance when using parity.  Specifically, write performance can be extremely slow, and rebuild times have made some use cases impractical.  Is there any work being done to help with this?  I understand that storage tiering (hot/cold tiers) might be one approach, but costs are a major issue for these clients.  Any suggestions would be welcome!

4 Replies
I'll also add that I found this Storage Spaces documentation to be very helpful, but I'd like a way to estimate read/write speeds with certain disk combinations. Of course, performance is the major concern for relatively large arrays (spinning disks).

https://docs.microsoft.com/en-us/windows-server/storage/storage-spaces/choosing-drives

(P.S. Sorry for not startring a new conversation; I didn't see a way to do that). :)
best response confirmed by Anil Desai (Copper Contributor)
Solution

Hi Anil, 

 

We have worked to improve rebuild times for Spaces across mirror, parity, and mirror-accelerated parity volumes. These improvements will be available in the Windows Server, Version 1709 release, and some of these improvements have been backported to previous 2016 releases as well. Let us know if you continue to see unsatisfactory rebuild performance, and if so, we'd like to hear more and use your feedback and target scenarios to improve rebuild performance further.

 

For parity performance in general, we are focusing our efforts on improving the performance of mirror-accelerated parity rather than parity alone. Depending on the size of your working set and the drift of that working set, you can improve your performance dramatically by using SSDs as only 20% or even less of your volume's capacity, which we hope offers a cost-conscious way to improve performance for parity.

 

Thanks,

Garrett

I also recommend checking out this video which discusses design considerations for Storage Spaces Direct, and covers the scenarios when you should use Mirror vs. Miror-accelorated Parity vs. Parity:

https://channel9.msdn.com/Blogs/windowsserver/Design-Considerations-for-Storage-Spaces-Direct

 

Thanks!
Elden

 

 

 

Hello…

OS is Windows 10 pro

So I’m running storage spaces with 14 drives (56TB) it was working fine until I added the last 4tb drive. (So it was working with 13 / 52TB). I went to run optimization and a BSOD occurred with the following error: stop code PAGE FAULT in NONPAGED AREA (what failed?) spaceport.sys

I have tried many things but no matter whenever I try to bring all the drives online the same stop code error occurs..

I don’t know what information you need but I will give as much as I can think. All drives state OK in storage spaces when they are connected in groups but when I try to connect all at the same time, BSOD error!

I have tried

Looked for device issues

Looked for drive issues

Ran windows RAM test

DISM online cleanup

SFC scan now

Replaced the spaceport.sys fire

Edited virtual memory settings

Reinstalled windows (kept personal files and apps)

Reinstall windows (kept personal files Only is running as I’m typing this)

Can anyone help me, Please? If this can’t be fixed, what do I do? Delete the volume? Can a recovery be run to try and restore the data? I’m really just so upset I don’t know what to do…. Any help, advice or magic silver bullet is welcome!!
1 best response

Accepted Solutions
best response confirmed by Anil Desai (Copper Contributor)
Solution

Hi Anil, 

 

We have worked to improve rebuild times for Spaces across mirror, parity, and mirror-accelerated parity volumes. These improvements will be available in the Windows Server, Version 1709 release, and some of these improvements have been backported to previous 2016 releases as well. Let us know if you continue to see unsatisfactory rebuild performance, and if so, we'd like to hear more and use your feedback and target scenarios to improve rebuild performance further.

 

For parity performance in general, we are focusing our efforts on improving the performance of mirror-accelerated parity rather than parity alone. Depending on the size of your working set and the drift of that working set, you can improve your performance dramatically by using SSDs as only 20% or even less of your volume's capacity, which we hope offers a cost-conscious way to improve performance for parity.

 

Thanks,

Garrett

View solution in original post