[OpenIndiana-discuss] Dedup - warning use or futher work?

[Ray Arachelian]

On 12/22/2011 07:41 AM, Milan Jurik wrote:
> Do you want to have big warning to every setup which can be problematic
> in some cases and in some not? In such case you will have warning in too
> many places.
>
> Maybe manpage can be improved to note that dedup makes sense only in
> specific setups and good configurations?
>

The only useful warning I can think of is to check the memory
requirements for the pool you're about to turn it on for and check if
there's an ARC partition/drive for that pool vs the total memory of the
host and warn about the estimated requirement for it.

I really wish there was an offline option to do deduplication, something
that could be run out of cron instead, say once a month or whatever and
wouldn't require a ton of memory.

My impression is that you can turn it on, copy your data to the zpool,
then turn it off and not have it un-deduplicate.  Maybe I'm wrong, but
if that's the case, they'd be a much smaller memory requirement at the
end, and you'd still get the benefits.  

For example, have two sets of disks.  Once in a while, you'd wipe one
and enable dedup on it, then copy all the data off the other one to it,
and then disable dedup on it.  Then, next time around, wipe the other
set, and repeat.  Painful, but would it work?