which file system to use

[David Abrahams]

ubuntu at rio.vg writes:

> David Abrahams wrote:
>>>
>>> "ZFS is currently not available as a root filesystem since there is no
>>> ZFS boot support, Sun plan to add this feature in late 2006."
>> 
>> In the linux world, doesn't that translate to "ZFS is currently not
>> available as a /boot partition...?"
>
> Actually, given the license issue, you can't even go that far.  

Isn't that "given the fact that the FUSE implementation isn't finished
yet?"

> The more important part of the quote though, was the "late 2006"
> estimation on adding it to the boot possibilities of Solaris.  This
> suggests that it is still not a mature system on Solaris.

I guess it depends how you measure "mature."  Is LVM mature yet?
'cause if I'm not mistaken, you still can't use an LVM logical volume
for /boot on Linux.

>>> "Sun has said it is investigating the porting of ZFS to Linux but this
>>> is complicated by the incompatibility between the OpenSolaris license
>>> CDDL and the GPL which governs the Linux kernel."
>>>
>>> So don't expect the linux kernel to support it.  Unless Sun changes the
>>> license, it will never enter the kernel.  The current "Google Summer of
>>> Code" project is to get ZFS running under FUSE, so it will only run in
>>> userspace...
>> 
>> Is that a problem?
>
> Yes.  This means the filesystem will be slower, 

...than it would be otherwise, yes.  How it will fare against other
filesystems remains to be measured.

> no boot possibility ever, not just no /boot, but you'd have to wait
> until the full user system comes up before mounting it.  

Hm; what does that mean in practice?

> As a general rule, things that run inside the kernel are faster than
> things in userspace.

Understood.

>>> In terms of ZFS, the operative phrase is "Don't hold your breath."
>>> Maybe, someday, it'll be ready, but not any time soon... there's a long
>>> way from "We've jury-rigged it to run through FUSE", and "I'll trust it
>>> with my critical data."
>> 
>> What makes that a "jury-rig?"  Seems to my naive eye that a
>> kernel-supported implementation, being more invasive and harder to
>> test, is also likely to be more prone to instability.
>
> FUSE is noted for doing things like sshfs.  Things that are very cool,
> but not exactly what I'd depend on for production machines, nevermind
> entrusting it to critical data.

Wow, educate me.  What's wrong with sshfs?

>> Anyway, I guess I'm liking the idea of Nextenta OS for my server
>> more and more.  As long as I can run virtualization software like
>> VMWare on it, I'm all set.
>
> We have, shall we say, somewhat different requirements in our server
> environments.  

Maybe not.  I'm still setting up my first server, and am the first to
admit I have lots to learn.

> Good luck with Nextenta.  I used to run Solaris servers, and
> needless to say, I'll be much happier never to run one ever again.

Maybe not needless.  At least, I'd like to know the details.  From
what I've read from "Linux people" who've been looking at the
OpenSolaris code, its standard of quality is far above what they've
been given to expect.

> I hope ZFS becomes a good mature Filesystem, but it's not their yet,
> and getting all excited about it now is just absurd.  

I think what to be excited about is really a personal choice :).

> An FS isn't like some beta piece of software you can just throw
> about and restart or debug if it breaks down.  When an FS breaks, it
> takes all your data with it.  

Sure.  My impression of ZFS is that resiliency been much more
carefully thought out than on any of the others, and that the bulk of
the code will have been very heavily exercised on Solaris in the field
before it ever makes it over to FUSE.

> I heartily recommend looking back on the early days of ReiserFS.
> Heck, even the early days of ext2 had major problems, as I recall.

It's definitely a consideration, but I think it'll become obvious
fairly soon if "major problems" with ZFS are going to be seen, and
then I can always steer clear.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com