[conspire] The practice of making ext4 a default needs to die an excruciating and gruesome death

[Peter Knaggs]

On Mon, Aug 13, 2012 at 11:03 AM, Rick Moen <rick at linuxmafia.com> wrote:
> Quoting Josef Grosch (jgrosch at gmail.com):
>
>> It's my understanding that Lawrence Livermore National Lab has a
>> contract from DOE to port ZFS to Linux. The project page is here
>>
>> http://zfsonlinux.org/
>
> Hi, Josef!
>
> Yes, they completed the job, and it works.  It remains an out-of-tree
> patch for annoying licensing reasons.
>
> Personally, if I wanted to do ZFS and leaned towards GNU-POSIX-ish
> userspace, I'd use Nexenta rather than a system based on the Linux
> kernel, so as to have fully integrated ZFS rather than a third-party
> patched subsystem.



I also noticed "OpenIndiana" has a nice installer, and it's based
on the Illumos kernel, so it has all the ZFS stuff fully integrated,
along with all the other goodies like DTrace.

I haven't installed OpenIndiana on "real hardware", but they have instructions
for installing it into a VirtualBox, and from there I was able to play with all
the ZFS commands (usually via filesystem files, rather than raw disk
partitions).

It's a fun way to experiment with ZFS, to get a sense of how it copes when
a disk gets corrupted, dies or becomes unresponsive. Using VirtualBox is
also a safe place to try out ZFS commands before running them on
your real ZFS pool, because sometimes what you expect isn't
what ZFS actually does, and recovering from a "user error" with ZFS
is sometimes a matter of "Hey, just restore your backup. You do have a
backup, right?" (that's all well and good, but I found that a 6TB restore
takes me about four days at an average speed of 20MB/s).

One "user gotcha" with ZFS, for example, is if you have one "raidz" pool,
and you want to add more more free space to it, you'd think you could add
one disk at a time, like for ordinary ZFS pools, but it turns out, you need to
add an entire second "raidz" pool, otherwise the single drive you added
just degrades the redundancy of the entire pool to that of a non-redundant
single drive, and there's no way to "undo".

They claim the reason they never implemented adding one-disk-at-a-time
to raidz pools is that "enterprise customers" never need to make their raidz
pools bigger, and if they do it's usually by huge increments.