we usually pull in upstream minor releases quickly, and that patch set is slated for inclusion in the next one or the one after it.
we usually pull in upstream minor releases quickly, and that patch set is slated for inclusion in the next one or the one after it.
ZFS 0.8.3 has been released, with SIMD fixes for Linux 5.0: https://github.com/zfsonlinux/zfs/releases/tag/zfs-0.8.3
FYI: We backported SIMD for our ZFS Kernel module with pve-kernel in version 5.0.21-1 (released end of August 2019) so there should be no change regarding that - if you use recent kernels at least 
A updated kernel and updated userland packages for ZFS 0.8.3 is currently getting tested, may get to the public testing repository today or start of next week, if all goes well.
A updated kernel and updated userland packages for ZFS 0.8.3 is currently getting tested, may get to the public testing repository today or start of next week, if all goes well.
updated kernel and updated userland packages for ZFS 0.8.3 are now available through the pvetest repositories (
See https://pve.proxmox.com/wiki/Package_Repositories#sysadmin_test_repo )
See https://pve.proxmox.com/wiki/Package_Repositories#sysadmin_test_repo )
ZFS is still awefully slow with me. (
Trying to run some low load vm's with a few Containers inside from a encrypted ZFS kills my Host for minutes. I have even thrown a Intel Optane on it. There's something seriously flawed with ZFS on Proxmox 6.1-5/9bf061 (Ryzen 2700x/64gb ECC/WD-Red + Optane 900p as SLOG)
Trying to run some low load vm's with a few Containers inside from a encrypted ZFS kills my Host for minutes. I have even thrown a Intel Optane on it. There's something seriously flawed with ZFS on Proxmox 6.1-5/9bf061 (Ryzen 2700x/64gb ECC/WD-Red + Optane 900p as SLOG)
What does your zfs config look like? Raidz? Mirror? Disk counts? Are you using the SSD for read cache or zil?
How does the output of "zpool iostat 1 10000000" look? Is it mostly reads or writes (post a snipit here is possible).
A 2 drive mechanical setup isn't going to provide much in regards to IOPS.
Is that setup the same? Are you running VM's on zvol's just like proxmox? Or are you simply using the zfs filesystem with no zvol's involved on the Debian setup. We use zfs within proxmox in many many setups with zero issues, but there are alot of options configuration wise to pin down for proper performance.
There are no zvol's involved on the other machine, just a ZFS that makes some exports. But on the other hand it's much weaker in terms of CPU (6y old HP Microserver with AMD Turion cpu only). I know it can't be compared 1:1, but I am pretty sure that what I see is not normal. It can't be. Still watiting for migrate to finnish to provide you with some solid data.
That is your issue then, they really can't be compared performance wise. I doubt the cpu aspect is that big unless its saturated. I had a feeling as we too have seen considerably better performance with a non zvol setup vs zvol. The only time we have found success with zvol's is having a large amount of vdev mirrors to get the IOPS up there.
However, we have found for our workloads, that moving to 128K volblocksize for the zvol's helps a ton and gets it slightly closer to non zvol performance. That would require you to re-write out all the data though.
Doesn't surprise me, imo 128k block would help here and not make the disks work so hard (As you really only have the power of 1 disk IOPS wise). However its never going to provide that much performance. Maybe a couple hundred IOPS.
Bummer, sounds like you need more vdev's to get the required IOPS to run zvol's for your VM's.