I want to benchmark something that's not been done before. ZFS/ZoL related. Would like input

I just saw a benchmark over on phoronix that compares ZoL vs ZFS on FreeBSD. I want to expand on this a little. Linux KVM is superior (most would say) to any other free hypervisor, so I currently run ZoL on my Ubuntu server with 4 KVM guests 24/7. It got me wondering what the results would be (on a proper Raidz-2) of FreeBSD with ZFS vs Ubuntu ZoL vs Ubuntu KVM with passed through SATA and FreeBSD with ZFS?

Thoughts? What benchmarking software should I use?

sounds like you are simply looking for KVM and ZFS at once. Have you taken a look at some Illumos distributions?
Illumos can run both ZFS (native) and KVM.

information links:
some distros here: https://wiki.illumos.org/display/illumos/Distributions
SmartOS by Joyent has a “why SmartOS” which you might find informative: https://wiki.smartos.org/display/DOC/Why+SmartOS+-+ZFS%2C+KVM%2C+DTrace%2C+Zones+and+More

None of the BSD based distros do hardware passthrough very well, which I must have. So I want to answer the question: Is it better to have a dedicated box running native ZFS or is virtualized/ZoL good enough.

ZFS wants direct access to disks. If you are feeding it virtual disks, then all bets are off.

1 Like

Have you heard of proxmox?

FreeBSD virtualized with passed through storage. The pool would be bare metal.

Yes I have

Are you using SSDs? There’s still no TRIM in ZoL which (correct me if I’m wrong) caused @nx2l some performance problems a while back.

It’s also still absent in Illumos according to the OpenZFS feature page.

http://open-zfs.org/wiki/Features#TRIM_Support

No my issue was a bug with zvols

1 Like

I think Nexenta had TRIM and the WIP ZoL TRIM is somewhat inspired by that IIRC, but otherwise it hasn’t made it out of Nexenta AFAIK. FreeBSD does have TRIM but we are working on pivoting to ZoL’s ZFS repo to reduce divergence, once they have that new TRIM implementation figured out.

Regarding the Phoronix benchmarks, something to keep in mind is that they benched a full debug build of TrueOS/ZoF meant for early testers against a bunch of non-debug releases. The results are going to be heavily skewed by that.

1 Like

Ah, my mistake. Do you know if it was resolved?

no idea… it’s stable and not trying to fix it if it isnt broken.

1 Like