Lab VM NFS Storage - Sanity check

[HoneyBadger]

Testing with Windows there was a huge difference between those two.

No surprise here. Windows might be bloated but even the worst of 10/11 would be able to fit the vast majority of itself into the cache area of an SSHD. The nature of single-user access and the persistence of the cache also means hit rates will be much higher there vs. a multi-user access scenario like TrueNAS.

Can TrueNAS handle those SSD Cache from those disk? How to enable it?
System has it's own UPS.

Much like DM-SMR the MLC cache on these drives is handled entirely by firmware voodoo and there's no way for ZFS or any other software solution (outside of something vendor-controlled and probably under NDA) to adjust its behaviours.

The read side will behave as an "L3ARC" victim cache, in that frequent reads from the same (unchanging) disk LBAs will end up getting served from the MLC cache, but if that data is ever updated, the copy-on-write nature of ZFS will mean the disk will consider it as "new data" because it'll be written to a new LBA, invalidating any kind of internal MFU/MRU listing the firmware has.

Random writes might be accelerated by the ability of the drive to potentially soak larger amounts of writes into that NAND buffer, and it could then spin the writes off to the disk later; but again, this is all supposition inside the black-box of a vendor-supplied solution.

TL;DR the caching is all done on-disk via firmware, we can't really see what it's doing, and it's likely not optimized for a multi-user copy-on-write filesystem.

 

[Herr_Merlin]

Intresting..
Need to find a way to expose those SAS drive to another computer. For testing.
Found two Seagate 2.5" SATA laptop HDDs with SSD Cache. Those two behave totally different. One is exposed to windows as a single drive the other as SSD and HDD as two different drives.
Now I am curious if you could expose those 1.8TB SAS as SSD and HDD to freenas. If there ever is a firmware for this.
Means I could utilize 18x 32GB MLC for whatever kind of usecase.. thats just below 600GB raw..

 

[HoneyBadger]

Throw an LSI HBA into a tower and you've got an easy way to get SAS on a desktop.

What are the model numbers of the Seagates? Those drives typically/historically just left the caching entirely to firmware level and didn't let you "pin" anything to the NAND or treat it separately. Presenting two distinct devices also would imply SATA multiplication which introduces challenges for hardware support.

If it's like the WD Black^2 drives, what it's doing is simply partitioning the drive. From the OS perspective, it's one long stretch of LBAs, with the first 128GB or so (however big the NAND piece is) of LBAs being mapped to the "SSD" and the rest going to the spinny rust. So if you manually created partitions in TrueNAS from the command line aligning with those boundaries (maybe allowing a big piece of "dead space" in the middle to be safe) then it would let you treat them separately.

But based on the whitepapers from Seagate, they aren't doing anything like that; it's all in-firmware control.