NixOS Takes Action After 1.2GB/s ZFS Encryption Speed Drops To 200MB/s With Linux 5.0+

[phoronix]

Phoronix: NixOS Takes Action After 1.2GB/s ZFS Encryption Speed Drops To 200MB/s With Linux 5.0+

The change in Linux 5.0 that initially broke ZFS On Linux compatibility ends up being pretty nasty for the ZFS encryption performance... A NixOS developer reports that the functions no longer exported by Linux 5.0+ and previously used by ZoL for AVX/AES-NI support end up dropping the ZFS data-set encryption performance to 200MB/s where as pre-5.0 kernels ran around 1.2GB/s...

http://www.phoronix.com/scan.php?page=news_item&px=NixOS-Linux-5.0-ZFS-FPU-Drop

[Veerappan]

Dear Oracle:

Please relicense or dual license ZFS so we can be done with this once and for all.

Yours,
A disgruntled developer who's forced to use your software at work


To be clear, I use ZFS on a BSD based NAS box and would like to use it on my other machines, but the license makes it too big a pain since every time I upgrade kernels, something breaks the dkms build.

[skeevy420]

I just knew this would happen; that someone out there would just work around that commit. Time to apply this patch and recompile my kernel

Thanks, NixOS & Mic92.

[skeevy420]

but the license makes it too big a pain since every time I upgrade kernels, something breaks the dkms build.

That sometimes happens on kernel point releases. Not really a big deal since it's usually fixed within a week if you don't mind switching to a git master build until the next ZFS rc or point release. It's what I've been doing for around three years now.

IMHO, it really isn't all that different than when the Nvidia kernel module breaks on point releases or when Catalyst used to do that AND require specific X.org versions. You run out-of-kernel stuff, you deal with out-of-kernel problems.

[numacross]

Dear Oracle:

Please relicense or dual license ZFS so we can be done with this once and for all.

I'm not sure that this would help. OpenZFS diverged too much from the Oracle version, which is closed source now anyway. Their implementations are incompatible.

Even if Oracle agreed, would re-licensing OpenZFS require all contributors to agree? Is that even practical?

To be clear, I use ZFS on a BSD based NAS box and would like to use it on my other machines, but the license makes it too big a pain since every time I upgrade kernels, something breaks the dkms build.

Can't you use a distribution that has support for ZFS like Ubuntu or Promox VE (there's probably more, but I haven't had the need to look for them)?

[ThoreauHD]

There is a fundamental problem here. If the license is disallowing it from being maintained in the kernel, it's a novelty. It's nothing other than a hobby filesystem. And that's all it ever will be until Oracle's lawyers get off their ass. If they can buy multi-million dollar companies and rip them apart to their bare atoms, they can figure out licensing on one fking driver.

No enterprise is going to put their faith in a dkms module for petabytes of data. They really need to sort this horse shit out before ZFS dies from apathy and neglect.

[numacross]

There is a fundamental problem here. If the license is disallowing it from being maintained in the kernel, it's a novelty. It's nothing other than a hobby filesystem. And that's all it ever will be until Oracle's lawyers get off their ass. If they can buy multi-million dollar companies and rip them apart to their bare atoms, they can figure out licensing on one fking driver.

No enterprise is going to put their faith in a dkms module for petabytes of data. They really need to sort this horse shit out before ZFS dies from apathy and neglect.

Oracle could open ZFS, but they won't since they are selling billions of dollars worth of solutions using it. Support contracts for those are very expensive as well.

There are a lot of commercial products using OpenZFS so it's not going to die any time soon...

[skeevy420]

There is a fundamental problem here. If the license is disallowing it from being maintained in the kernel, it's a novelty. It's nothing other than a hobby filesystem. And that's all it ever will be until Oracle's lawyers get off their ass. If they can buy multi-million dollar companies and rip them apart to their bare atoms, they can figure out licensing on one fking driver.

No enterprise is going to put their faith in a dkms module for petabytes of data. They really need to sort this horse shit out before ZFS dies from apathy and neglect.

It really isn't any different than Nvidia being an out of tree module due to the license keeping it out of the tree and it being closed source.

Is the Nvidia driver a novelty? Nope.

Is the Nvidia driver a hobby graphics driver? Not at all.

Does that stop enterprises from using Nvidia? Hell fucking no.

The only thing that needs to be sorted out are the personal feeling of the Linux kernel developers. Reading the mailing lists about this, they don't have any good reasons other than "we don't care about ZFS so we'll be assholes and intentionally cripple your FS even though y'all constantly try to work with us to try to come up with solutions that will make us, the Linux Kernel Devs, and the ZoL devs happy".

[lumks]

Would all that afford go into a new in-tree zfs-like-FS + a tool to convert ZFS to that new filesystem. We would have it in 1-2 years from now on in a stable version. But since it's the linux community, let's cry and rant over licenses, try to workaround things and don't actually do something in the right direction.

bravo.

[aht0]

There is a fundamental problem here. If the license is disallowing it from being maintained in the kernel, it's a novelty. It's nothing other than a hobby filesystem. And that's all it ever will be until Oracle's lawyers get off their ass. If they can buy multi-million dollar companies and rip them apart to their bare atoms, they can figure out licensing on one fking driver.

No enterprise is going to put their faith in a dkms module for petabytes of data. They really need to sort this horse shit out before ZFS dies from apathy and neglect.

Who do you think is backing much of it's development, despite it being out-of-tree? Enterprise.

Be nice if further ZFS/ZoL/ZoF comparisons between Linux and FreeBSD included encryption speed testing.