What are they claiming?
So, the performance and latency numbers are on the pretty damn slow side. Probably still bottlenecks associated with using Data OnTap which is famously slow. Azure has consistently shown far better storage performance number than this in the Storage Spaces Direct configuration.
I have seen far better numbers on MariaDB using storage spaces direct in the lab as well. With a properly configured RDMA solution for SMB3 in the back end, there is generally between 80 and 320gb/s back end performance. This is substantially better than any NVMe configuration mainly because NVMe channels are so small in comparison. Of course, the obscene amount of waste in the NVMe protocol adds to that as well. NVMe is only well suited for direct to device attachment. By routing it through a fabric, it severely hurts storage latency and increases chances for errors which aren’t present when using PCIe as designed.
Overall, it’s almost always better to use MariaDB scaled on Hyper-V with paravirtualized storage drivers then to do silly things like running it virtualized over NFS. In fact, you will see far better numbers on proper Windows technologies than by using legacy storage systems like this.
I think the main issue here is that Microsoft didn’t want to deal with customers who absolutely insist on doing thing wrong. So they bought a SAN and just said... “Let NetApp deal with these guys. We’ll manage customers who have actual technical skills, NetApp can have the customers who think virtual servers are smart”.