`zpool create' fails on geli ...

Fabian Keil freebsd-listen at fabiankeil.de
Thu Aug 23 09:47:24 UTC 2012


Mark Martinec <Mark.Martinec+freebsd at ijs.si> wrote:

> > Fabian Keil wrote:
> > > Zeus Panchenko <zeus at ibs.dn.ua> wrote:
> > > > geli init -K /path/key -s 4096 -e aes-xts /dev/adaX
> > > 
> > > Does your disk actually use 4k sectors? Otherwise it's not clear
> > > to me that "-s 4096" makes sense when using ZFS.
> > > 
> > > I'm not claiming that it's obviously wrong, but I'm not aware of
> > > any benchmarks that show that it's better than the default in
> > > any way.

I probably should have clarified that I don't deny that
workloads exist where using 4k sectors indeed improves the
performance even if the disk is using smaller sectors.

> It benefits geli performance (tried it, it does):

You probably didn't test with random read or write operations of less
than 4k, for which a smaller sector size should result in better
performance.

> $ man geli
>   -s sectorsize   Change decrypted provider's sector size.
>                   Increasing sector size allows to increase per-
>                   formance, because we need to generate an IV
>                   and do encrypt/decrypt for every single sector
>                   - less number of sectors means less work to

Provided you actually need the content of the whole sector ...
And if you always do, why not increase the sector size even further?

If -s 4096 would provide the best results for all work loads
it probably would be the default already.

> > It is my understanding that creating a 4K setup will prepare you
> > for the day when your replacement drive is a 4K one.

That's true and I didn't consider this (I don't usually replace
drives in single-drive pools).

> > No benefit today, but also no real performance hit. And we avoid
> > a real performance hit later.  

As I said previously, I'm not aware of any benchmarks that show
how much impact the geli sector size has on the ZFS performance.

Fabian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20120823/1ce25556/signature.pgp


More information about the freebsd-fs mailing list