5

As Microsoft's ReFS draw my attention because of data corruption occurred to my storage I'm wondering if it's recommend to use ReFS in a virtualized environment.

To be precise my stack would be a bunch of physical discs on a RAID Controller on a physical Esxi host, which virtualizes a Windows Server 2016 as a file server.

The datastore is created on top of a RAID Array. Within the datastore e.g. two virtual drives are created (vmdk's) which are connected to the file server VM.

Both of these virtual drives are used to create a mirrored ReFS file system with data integrity enabled.

Overview:

  Physical host                --            physical discs
       |                                            |
       |                                     RAID Controller
       |                                            | 
       |                                       RAID 0 or 1
       |                                            | 
     Esxi Host                                   Datastore
       |                                            | 
 VM (file server)                      multiple virt. drives (vmdk)
       |                                            | 
Windows Server 2016                           Storage Spaces
                                                    | 
                                            Virtual Drive (ReFS)
                                                    | 
                                         populated network share
                                          ---------------------

Would you recommend this stack or what elements could be problematic? Major goal is to have protection against bit rot (silent data corruption). The constraints are one physical machine that needs to hold different VMs (one could be the file server) and a RAID controller not supporting JBOD (but controller could be removed).

Manu Kast
  • 53
  • 5

1 Answers1

5
  • Make sure you understand ReFS with checksumming enabled turns itself into log-structured file system, and this thing alone dramatically changes I/O pattern. TL;DR: Not all the workloads are OK with "safe" ReFS.

https://www.starwindsoftware.com/blog/log-structured-file-systems-microsoft-refs-v2-investigation-part-1

  • It's bad do layer one log on top of the other.

https://www.usenix.org/node/187064

BaronSamedi1958
  • 12,510
  • 1
  • 20
  • 46