Can FreeNAS allow for data to be written to a volume comprised of SSDs 1st and then moved to HDDs

[Thomas Blackwell]

On some of the higher end storage facilities the system has the ability to move data that reaches a defined point or policy expiration over from the faster SSD volume over to a secondary HDD-based volume. Can this be configured with a FreeNAS setup? If so, how would one do this configuration?

 

[depasseg]

FreeNAS doesn't support the concept of tiered storage in that manner. And I haven't heard of any hacks to provide it, but they might be out there. LMK if you find one, I'd love to use it.

 

[joeschmuck]

That would be a nice feature. Not sure I'd use it but still I'd at least test it out to see how it worked. Think we could ask for a Feature Request on this one?

 

[depasseg]

Joe - this video gives an idea of how automated storage tiering/migration works (around the 1:50 mark). https://www.youtube.com/watch?v=yvK9o5NTFVk

I would love to see it, but would imagine the work to implement it properly would be enormous.

 

[Ericloewe]

There are several points to consider here, related to ZFS:

• The vast majority of writes are cached in RAM only. If the pool isn't massively fragmented and full, these can be written very quickly even by mechanical drives.
• ARC/L2ARC hit rates in the 90% range are not uncommon. That means that 90% of reads would already be served from SSD or even from RAM.
• This kind of thing is massively complex. In practice, you need a filesystem designed for it. ZFS would be nearly impossible to adapt, as this would require massive changes to pool logic and block pointer rewrite. Hell will freeze over before BPR sees the light of day (if you want to say that hell has already frozen over, pigs will fly before BPR exists)
• If you were to do it as an additional layer between ZFS pools and FreeBSD, it'd be only slightly easier. An additional layer of pseudo-filesystem would destroy performance in many scenarios that are latency-sensitive. Moving stuff between pools would also be sub-optimal. And it's nearly guaranteed to break. Microsoft tried something similar, with a different scope, in WHS vanilla. It was buggy as all hell to the point where they just gave up for WHS2011.

 

[Stux]

An option would be a non deleting rsync and some script magic

 

[Ericloewe]

Yeah, part of this can be done with script magic, but presenting a single unified pool of storage is not going to happen.

 

[danb35]

pigs will fly before BPR exists

Given adequate thrust, pigs fly just fine. (-: