Btrfs Gets Talked Up, Googler Encourages You To Try Btrfs

[drSeehas]

Because you cannot distribute ZFS with the kernel with its current license?

I know, but what is the problem with not distributing ZFS with the kernel?

[drSeehas]

... ZFS license ...

It is not a "ZFS" license, it is the CDDL.

... is one huge pile of dung.

Why? It is an approved open source license.
OpenSolaris (itself CDDLd) and the BSDs have no problem with the CDDL.
As far as I understand, the GPL is not compatible with the CDDL.
So I think, Linux has the wrong license and should change it if possible.

[jeff_g]

The primary reasons given in the talk (he seemed to stress it more than the slides make it seem) was regarding snapshots and the ability to send those snapshots to remote systems. Slide 35 and 36 actually explain something I've done (2.5 ssd to msata though) as a fall back method for whenever I manage to really bork up my laptop and suddenly need it usable for something.

I was sure it was going to be a full on flame war by now. ZFS does have plenty of die hard fans.

It almost did at the presentation. Some guy in the front row started listing "factual errors" in Marc's statements regarding ZFS legal details and just refused to drop it until after Chris spoke up.

I'm using btrfs for my home directory now, but I have noticed that sometimes there are some odd bugs or hangs. For example, on my laptop running linux 3.16 the moment the partition is full the system becomes so unresponsive that I have to SysRq reboot it. Then afterwards some files are missing, and as usual the chrome preferences are corrupted...

Marc and Chris were discussing this after the talk actually - are you using Laptop Mode Tools and/or aggressive power saving tweaks? I think Marc said he put in a bug report for it or was going to. I've encountered similar issues, but I thought I had done something to break it and so didn't dig very deep into it.

[smitty3268]

I know, but what is the problem with not distributing ZFS with the kernel?

Maintaining such an important piece of the kernel externally away from upstream code is certainly possible, but to date I'm not sure it's ever been done very well and it comes with a lot of well known problems you have to overcome. I'm not saying it's impossible, but it seems unlikely ZFS will ever become too popular on linux systems while it is developed separately. If that weren't true, btrfs probably would have been dropped by now.

As far as I understand, the GPL is not compatible with the CDDL.

That's semantics, you can say it either way. The reality is that the GPL existed first, and CDDL was clearly written with the intent to be incompatible with it for business reasons.

So I think, Linux has the wrong license and should change it if possible.

You're free to think whatever you want, but at this point changing the license of linux is practically speaking impossible. They'd basically have to start a new kernel from scratch in order to do so, and at that point there'd be little point in retaining the "linux" name for it.

[Tired_]

Can btrfs be used for archival, or does it have the same problem as ZFS, where if you don't have the fs actively running, it gets corrupted? I'm looking for a new filesystem for a home server, but I don't want to lose all my files again if I need to power down the server for a long while.

[drSeehas]

Maintaining such an important piece of the kernel externally away from upstream code is certainly possible, but to date I'm not sure it's ever been done very well and it comes with a lot of well known problems you have to overcome. I'm not saying it's impossible, but it seems unlikely ZFS will ever become too popular on linux systems while it is developed separately. If that weren't true, btrfs probably would have been dropped by now.

First I want to thank you for your factual answers.

You are right. Maintaining such an important piece of the kernel externally away from upstream code will come with problems. Going with Btrfs will be easyer.

That's semantics, you can say it either way. The reality is that the GPL existed first, and CDDL was clearly written with the intent to be incompatible with it for business reasons.

I agree that the GPL existed first, but disagree that the CDDL was clearly written with the intent to be incompatible with the GPL. This is a myth and was vetoed by SUN officials in public.
SUN had the right to chose any license they want. Fortunately they chose an open source license.

... changing the license of linux is practically speaking impossible.

Thank you for clarifying.

They'd basically have to start a new kernel from scratch ...

And this would be a much larger task than reimplementing ZFS from scratch.

[jeff_g]

Can btrfs be used for archival, or does it have the same problem as ZFS, where if you don't have the fs actively running, it gets corrupted? I'm looking for a new filesystem for a home server, but I don't want to lose all my files again if I need to power down the server for a long while.

Are you talking about a purely flash based server? Either way, you could use the method I do for more important data - btrfs raid 1 with weekly or even nightly scrubs. If it finds any blocks that fail a checksum check, it rebuilds from the second disk.
(or, if you don't run often, kick it off as part of startup or shutdown and have it email you upon anything abnormal)

[ryao]

Can btrfs be used for archival, or does it have the same problem as ZFS, where if you don't have the fs actively running, it gets corrupted? I'm looking for a new filesystem for a home server, but I don't want to lose all my files again if I need to power down the server for a long while.

ZFS does not have that problem. If you had an issue, you likely should have gotten in touch with the developers. If this is a hardware problem, then we cannot help you. A stab in the dark would be that you tried using a hardware RAID controller like LSI MegaRAID. Prolonged power loss on those can and often will severely damage any file system once the BBU is drained whenever data is left in its write cache.

[ryao]

Sun's ZFS code is CDDL. What i meant was that linux devs can't reimplement ZFS from scratch using the specs (or reverse engineer it) as GPL code, because then they'd be sued for patent infringement.

The same likely applies to btrfs because many of the ZFS patents should apply to btrfs as well. There is no patent grant in the GPLv2 and I am not aware of any patent grant by Oracle for the ZFS patent portfolio. Unless we get it, we have a tivoization-like issue where you can look, but not touch. I suspect that Oracle Linux is the only distribution that is able to use btrfs without risking a lawsuit. If that is not the case, it would certainly be nice for Oracle to tell us that, but somehow I doubt they will.

That being said, I am not a lawyer, but I think the idea that Marc put forward during his talk that placing the code under the GPLv2 gave people a patent grant needs legal review. There is nothing about patent grants in the GPLv2. I ran this question by gregkh who teaches a class on the GPL to people studying law and he told me to consult an attorney.

[ryao]

Try:

noatime nobarrier nodatacow nodatasum ssd ssd_spread compression=lz4

LZ4 patch to compression... this will do a fast fs, but corruptions everywhere. For enthusiasts, is interesting try it some day...

It is not really possible to use nodatacow in a sane way with checksums in general, which is why ZFS does not do it. It could be that the LZ4 patch for btrfs has issues, but I am severely disinclined to think using nodatacow is a good idea. It would only be useful for things like files storing raw VM disks that have their own filesystem protection. That being said LZ4 works fine in ZFS.