Server 2019, NVMe only storage spaces, parity, dirt slow

Question

I have single servers that were bought with NVMe storage. They have 8 1.5tb drives configured for about 8tb of total storage with parity. While we expected parity to be slower than mirrored, we expected the NVMe to keep them adequately fast. The OS drive is on separate mirroed NVMe in the Dell BOSS controllers and runs nicely fast.

The storage space pool drive is horrible, slower than I believe equivalent HDD's might be, certainly nothing like what we expected. We cannot manage more than about 50MB/s in a mixed workload.

I have been reading and reading and am trying to understand if this is some kind of misguided use of writecache, where it is staging writes on the same media it is writing too, then flushing them quickly due to the tiny size (1gb). But that's speculation, all I know for sure is it is dirt slow.

In searching it seems about 90% are aimed at storage spaces direct (which this isn't) or actual tiered storage (which this also isn't).

The spaces and drives were created with defaults.

We have this situation on two similar servers (not in a cluster), both dirt slow, I do not think this is a hardware issue. These are Dell Poweredge R740xd's, the subsystem is simply called a "PCIe SSD Subsystem and configured for NVMe 1.2. There's no apparent settings for the subsystem, and no caching (I presume since all access is PCIe and fast).

Dell doesn't support NVMe drives on raid controllers so storage spaces were the natural alternative. I guess software raid was a possibility but did not try that.

Am I missing something basic? Should I be trying to turn off default write cache on the pool so the created volume does not do any on-disk caching, just direct writes?

I'd rather not recreate the virtual drives since they are quite full with data and no good place to stage them during recreation, but worst case we could.

Is there some way to encourage in-memory caching? (These are all UPS powered).

This just feels like some gross misconfiguration we have done, any advice welcomed.

If it matters, these are destined for a near term upgrade to Windows 2022 server.

Linwood

Answer 1

Hi. Thank you for your question and reaching out. My name is John and I’d be more than happy to help you with your query.

According to this article https://learn.microsoft.com/en-us/windows-server/storage/storage-spaces/troubleshooting-storage-spaces
To debug your Storage Spaces Direct deployment, use the information below.

Start generally with the actions listed below:

1. Using the Windows Server Catalog, verify whether the SSD's brand and model are approved for Windows Server 2016 and Windows Server 2019. Verify with the seller that Storage Spaces Direct is compatible with the drives.
2. Look through the storage for any problematic drives. Utilize storage management software to examine the drives' state. Work with your seller if any of the drives are malfunctioning.
3. If required, update the drive and storage firmware. Make that all nodes have the most recent Windows updates installed. The most recent updates for Windows Server 2016 and 2019 are avai
4. Update the firmware and drivers for network adapters.
5. Run the cluster validation process, go over the Storage Space Direct part, and check to see if any faults are indicated for the drives that will be utilized for the cache.lable from the Windows 10 and Windows Server 2016 update history and the Windows 10 and Windows Server 2019 update history, respectively.

If the reply was helpful, please don’t forget to upvote or accept as answer, thank you.