OpenZFS 2.0-RC2 Released With Dozens Of Fixes

[starshipeleven]

Yes and no. The OpenZFS parts that interface with the Linux kernel (Solaris Porting Layer, SPL) are GPL. Those could be upstreamed but pretty sure the kernel devs don't want to.

afaik it's not just using SPL but also interfacing directly with kernel.

[starshipeleven]

Independent bench tests here on Phoronix indicate that BTRFS is not as fast as the simpler partition types: EXT4 & NTFS. Whether ZFS or openZFS can do better than BTRFS probably needs testing.

From my testing, ZFS has better performance than btrfs, especially in RAID modes.

Would be nice to have some benchmarks on phoronix, especially now that Ubuntu supports ZFS by default

[smitty3268]

I want to ask & confirm something:

Am I correct?

CDDL license can not. If it was BSD or MIT (or gpl v2), it could.

Though I suspect it still may not be allowed even if the license was ok, as the design of ZFS doesn't mesh well with what the linux devs prefer. It manages everything itself internally rather than using the various existing kernel systems that all the other filesystems share with each other.

[rlkoekie]

The 2.0 release should also bring persistent l2arc cache. This is a huge plus for larger HDD based arrays with SSD caches on machines that might need to reboot from time to time

[ehansin]

It manages everything itself internally rather than using the various existing kernel systems that all the other filesystems share with each other.

Just a curious question. In the case of ZFS and what it does, I assume this is a good thing and how it should be. Would that sound correct? I mean, my understanding is that it is supposed to manage all the "layers" that separate pieces currently handle in other Linux setups.

[smitty3268]

Just a curious question. In the case of ZFS and what it does, I assume this is a good thing and how it should be. Would that sound correct? I mean, my understanding is that it is supposed to manage all the "layers" that separate pieces currently handle in other Linux setups.

That's a matter of debate. People who like ZFS say it's good, people who don't like it say it's bad.

[ehansin]

That's a matter of debate. People who like ZFS say it's good, people who don't like it say it's bad.

Okay, got it. Seems like the same for many things

[mbod]

I want to ask & confirm something:
Since "Open ZFS" is not GPL 2. It can never come in the mainline Linux kernel.
Am I correct?

That is correct. But this is only relevant for kernel developers resp. distributors. Basically this is an issue if you are shipping the source code of Linux.

The ZFS source can not be included into the kernel due to the license issue. But it can be distributed as a module like Ubuntu or Manjaro are doing it.

[pavlerson]

That's a matter of debate. People who like ZFS say it's good, people who don't like it say it's bad.

No, that is not a matter of debate. ZFS handling all layers itself, is what gives the data safety. In other filesystems, there are different layers. One raid layer, one filesystem layer, etc. This is problematic because it is difficult for each layer to pass read errors and repair the errors. Some layers can not receive read errors. So this multilayered approach is the reason other filesystems are unsafe.
OTOH, ZFS is monolithic and ZFS handles everything itself, so it is aware of all the different errors and can repair all errors. This monolithic approach, which Linux kernel devs famously called "a rampant layering violation", is the reason ZFS is superior. And even though Linux kernel devs mocked and laughed ZFS "bad design doesnt make sense" soon after BTRFS emerged - which, you guessed it, is a monolithic design just like ZFS. But BTRFS is not done right, so it is unsafe.

The conclusion is that a storage system must handle all errors itself to be able to be aware and repair all errors. Otherwise, you get bad safety.

[smitty3268]

No, that is not a matter of debate. ZFS handling all layers itself, is what gives the data safety. In other filesystems, there are different layers. One raid layer, one filesystem layer, etc. This is problematic because it is difficult for each layer to pass read errors and repair the errors. Some layers can not receive read errors. So this multilayered approach is the reason other filesystems are unsafe.
OTOH, ZFS is monolithic and ZFS handles everything itself, so it is aware of all the different errors and can repair all errors. This monolithic approach, which Linux kernel devs famously called "a rampant layering violation", is the reason ZFS is superior. And even though Linux kernel devs mocked and laughed ZFS "bad design doesnt make sense" soon after BTRFS emerged - which, you guessed it, is a monolithic design just like ZFS. But BTRFS is not done right, so it is unsafe.

The conclusion is that a storage system must handle all errors itself to be able to be aware and repair all errors. Otherwise, you get bad safety.

I think you know that there are people who disagree with this, and therefore by definition there is debate.

Which is why you felt the need to try and argue so much for it here.

Like I said, people who like ZFS claim it's superior. People who don't like it think it's a bad design.