ZFS woes (was Re: Backing up sparse files ... VM's and TrueCrypt ... etc)

[richard.pieri-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org (Richard Pieri)]

On Feb 21, 2010, at 12:21 AM, Edward Ned Harvey wrote:
> 
> The terms of the Linux, solaris, and windows kernels are all mutually
> exclusive.  You can't take any code from one and build it into any other.
> So how can you get the portability across OSes?

Not code portability.  Filesystem portability.  We don't have that and because of the CDDL we never will.

> For one, they could build a loadable standalone module, which loads into
> kernel but isn't distributed with the kernel, and isn't built as part of the
> kernel.  It's got an interface that the kernel can use ... and it can run in
> kernel space instead of user space ... and eliminates the conflict of
> license agreements.
> 
> As long as it loads into the kernel, what's the problem?  

>From my perspective, both as a user and a sysmonster, ZFS not being in the kernel is a losing proposition.  Never mind not having cross-platform ZFS, I'll eventually run into problems with ZFS not being cross-distribution.  There are issues with CDDL code in Debian, for example, so conversion from CentOS (say) to Debian may be a problem, and maintaining the converted system will end up being a hassle of kernel and module dependencies.  ZFS's features aren't worth it.

That's not even beginning to touch on the performance and security implications of user-space filesystems.

--Rich P.