effect of differing spindle speeds on prospective zfs vdevs

[Paul Mather]

On Fri, 4 Dec 2020 23:43:15 +0000, tech-lists <tech-lists at zyxst.net> wrote:

> Normally when making an array, I'd like to use all disks all same speed, 
> interface, make and model but from different batches. In this case, I've no 
> choice, so we have multiple 1TB disks some 7.2k some 5.4k. I've not mixed
> them like this before.
> 
> What effect would this have on the final array? Slower than if all one or the other?
> No effect? I'm expecting the fastest access will be that of the slowest vdev.


I believe you are correct in intuiting that the performance of the pool will be influenced by the slowest devices.

ZFS supports a variety of pool organisations, each with differing I/O characteristics, so "making an array" could cover a multiplicity of possibilities.  I.e., a "JBOD" pool would have different I/O characteristics than a RAIDZ pool.  Read access would also be different than write access, and so the use case of the pool (read-intensive or write-intensive) would I/O speeds.  (And, furthermore, small random vs. large sequential I/O will have an impact.)

IIRC, write IOPS of RAIDZ pools are limited to the IOPS of the slowest device.


> Similarly some disks block size is 512b logical/512b physical, others are 
> 512b logical/4096 physical, still others are 4096/4096. Any effect of
> mixing hardware? I understand sfs sets its own blocksize.


IIRC, ZFS pools have a single ashift for the entire pool, so you should set it to accommodate the 4096/4096 devices to avoid performance degradation.  I believe it defaults to that now, and should auto-detect anyway.  But, in a mixed setup of vdevs like you have, you should be using ashift=12.

I believe having an ashift=9 on your mixed-drive setup would have the biggest performance impact in terms of reducing performance.

Cheers,

Paul.