Is ZFS production ready?

[Matthias Gamsjager]

On Thu, Jun 21, 2012 at 3:43 PM, Wojciech Puchar <
wojtek at wojtek.tensor.gdynia.pl> wrote:

> I really want to see your face when you fsck 48TB w/o ffs+j (since that is
>> so young must be immature :S ) of data with the phone ring non stop with
>>
>
> Even if ZFS would be the only filesystem in existence i would make one per
> 2 disks (single mirror).
>
> No matter what's going on, what do you prefer in case say - double disk
> failure from one mirror on 48 disk systems?
>
> losing completely data of 1/24 of users (and then restoring that amount
> from backups), or losing randomly chosen 1/24 of files from whole system?
>
> answer yourself.
>

Sorry but I don;t follow you right there. with 48 disks you would not
mirror 24vs24. I will perform very well but there is too much risk in that.
you would rather go with a raidz2 stripe sets.


>
> With UFS of  course i would have single disk fsck time - less than a hour.
> which CAN be done out of work hours with soft updates.
>
> i normally turn off automatic fsck for large data filesystems, and if
> crash happened i run it after/before work hours.
>
>
> raid is not a backup. You can loose data with any configuration or fs. so
like in the compiler discussion. There is no perfect something in this
world. It's always a tradeoff.
with ZFS you have access to most advanced techniques and I believe that
data is most safe with raidz3 as it can be. UFS cant match that and you
have to rely on a raidcontroller which can screw up your data as well.