What about replacing failed HDD by SSD ?

[Youri Andropov]

Hello there,

High capacity SSD (2+ TB) are still expensive today, but their price is decreasing.
My main freenas installation is using 6x Seagate Barracuda 3.0 TB HDD in a raid-Z2 ZFS pool. So I was wandering if it would be possible (and usefull) to progressively replace my failing HDD with 4.0 TB SSD (Samsung 850 Evo by instance).
Will SSD Trim command work ? When all disks will be replaced by SSD, will I be able to reclaim the unused 1.0 TB space on the disks ?

 

[HoneyBadger]

"Possible" is yes. As long as it's presented as a storage device with equal or greater space than the one it's replacing, it will work. It may not get TRIM commands passed to it in a mixed-media vdev though - I admit I haven't tried that use case.

"Useful" depends if you need the higher read/write performance of the SSD - and you'll still be limited by the speed of the slowest device in the pool.

And finally, you'll have to replace all six drives with SSDs in order to expand the pool to the 4TB/drive size. Until you do, that extra space you paid for is unusable.

 

[Ericloewe]

For generic file storage, SSDs are way overkill outside of high-performance scenarios. ZFS' caches make the HDD experience quite bearable. Improvements to compression that extend it to ARC will only improve this.

Will SSD Trim command work ?

It should.

 

[Youri Andropov]

Well I'm not thinking about performance when replacing HDD with SSD, because I'm already saturating the Gbit ethernet link when doing file transfer (~110 MB/s). Performance is not the point.
I rather believe that SSD may be much more robust over time, 850 Pro series have a 10-years warranty, and huge MTBF. A retail HDD typically dies in 2 to 4 years, this is not acceptable.
I've built my NAS in 2011 with 5 disks, none of them is still working as of today. My plan is to use SSD as a much more robust and reliable technology.

My only reservation is : will the SSD work as expected when used in a RAID array ? Won't it be penalized by unaligned writes and things like that ? Won't the wear leveling logic be bypassed, shortening its lifespan ?

 

[Ericloewe]

I rather believe that SSD may be much more robust over time,

That is to be seen.

A retail HDD typically dies in 2 to 4 years

That's on the very low end of things.

I've built my NAS in 2011 with 5 disks, none of them is still working as of today.

Big deal, SSD is much more expensive and might not be any better.

My only reservation is : will the SSD work as expected when used in a RAID array ? Won't it be penalized by unaligned writes and things like that ? Won't the wear leveling logic be bypassed, shortening its lifespan ?

There's no specific reason to not use SSDs other than cost.

 

[horizonbrave]

I'm also wondering the same, if a pool of 1 Vdev (4-6 HDD wide) could have its drive replaced one by one for SSDs.
I wonder if anyone tried it before.
Cheers :)

 

[Ericloewe]

Of course it could.