[OpenIndiana-discuss] Puzzling behaviour in zpools

[Jan Owoc]

Hi Len,

On Mon, Mar 4, 2013 at 11:52 AM, Len Zaifman <leonardz at sickkids.ca> wrote:
> I have a system which I am configuring for maximum space to use as a low cost backup service. It has an Areca raid card , 24 2 TB drives (Format reports 1.82 TB)  and 12 4 TB drives (format reports 3.64 TB).

Yes, that is to be expected. Manufacturers say "4TB" meaning "4 000
000 000 000" bytes. Many operating systems say "3.64 TiB" meaning
"3.64 * 1024 * 1024 * 1024 * 1024". They are two different ways of
expressing the same capacity.

> using 2 2 TB drives for a mirrored rpool, I create a 2 TB pool as 2 x  11 disk raidz2 vdevs and a 4 TB pool as 1 x 12 disk raidz2 vdev

Ok, so for each set of raidz2 disks, 2 are for parity, so you have 9
for data. 2 vdevs * 9 disks * 2TB = 36.0 TB = 32.7 TiB

On the other pool, for each set of raidz2 disks, 2 are for parity, so
you have 4 for data. 2 vdevs * 4 disks * 4TB = 32.0 TB = 29.1 TiB

> df shows
>
> rpool/ROOT/openindiana
>                       1.8T  1.6G  1.8T   1% /
> rpool/smallbkup       1.6T   31K  1.6T   1% /rpool/smallbkup
> ccmbkup12TB            32T   68K   32T   1% /ccmbkup12TB
> fourTBpool             29T   56K   29T   1% /fourTBpool    <<< I think I am 7 TB short here

See calculation above :-). It matches up exactly.

As your pools fill up, people suggest using "zfs list" to get actual
usable capacity, as df gets confused by compression, deduplication,
and other things. Your pools are empty, so the numbers happen to
match.

Cheers,
Jan