As I understand it now, with 2019/Windows 10 1903, Hyper-V has gotten more advanced, such that we now have PCie passthru in the form of Discrete Device Assignment (DDA) for NVMe devices as well partial PMEM availability. DDA NVMe passthru seems to require SR-IOV, and PMEM is cheating a bit as it uses a VHD file on a DAX enabled NTFS volume on configured PMEM physical storage on the Hyper-V host.
So, with the NVMe passthru, one could do optane and generic NVMe (M.2 and U.2 flavor) direct to a FreeNAS VM. The NVMe optane passthru might be attractive for SLOG/ZIL, while the PMEM (DC Optane) might be attractive for SLOG/ZIL as well.
Has anyone tried this yet, and is it better or worse than previous attempts at FreeNAS VM's on Hyper-V?
And for added fun, if someone attempted to do this with a PLX chip downstream of the CPU to increase their NVMe device count (Amfeltec Squid or similar cards, backplanes, etc.), would that make things even worse, even with the SR-IOV in theory protecting the data path?
I ask this because Windows Storage Spaces, even with the mirror-parity multi-resilent volume with journal on ReFS with checksum setups now available, appears to still have efficiency issues (3 journal devices, 3 devices for triple mirror cache tier, and 7 devices for dual parity tier would be a minimum config it seems). Well, that and some people still feel Storage Spaces is a raging tire fire...
So, with the NVMe passthru, one could do optane and generic NVMe (M.2 and U.2 flavor) direct to a FreeNAS VM. The NVMe optane passthru might be attractive for SLOG/ZIL, while the PMEM (DC Optane) might be attractive for SLOG/ZIL as well.
Has anyone tried this yet, and is it better or worse than previous attempts at FreeNAS VM's on Hyper-V?
And for added fun, if someone attempted to do this with a PLX chip downstream of the CPU to increase their NVMe device count (Amfeltec Squid or similar cards, backplanes, etc.), would that make things even worse, even with the SR-IOV in theory protecting the data path?
I ask this because Windows Storage Spaces, even with the mirror-parity multi-resilent volume with journal on ReFS with checksum setups now available, appears to still have efficiency issues (3 journal devices, 3 devices for triple mirror cache tier, and 7 devices for dual parity tier would be a minimum config it seems). Well, that and some people still feel Storage Spaces is a raging tire fire...
