Nvme cache with Nvme Pool?

[Jlarimore]

If I, knowing the risks, go ahead and make a 4 x 8tb nvme data pool, is there a reason to still use an nvme cache drive? As I understand it, the write speed bottleneck is going to come from computing/writing parity.

 

I only have 6 m.2 slots to work with. So, the more of those that go towards storage the better. Can't RAM act as the cache?

[JonathanM]

The use of the word "cache" in Unraid isn't typical.

 

Storage pools is a much better description.

 

The parity array is the primary pool, and it's the only pool that works with the Unraid special sauce of independent file systems on each drive with the ability to mix and match sizes, with a dedicated drive to add parity protection for 1 or 2 device failures.

 

The other pools can be either single volume XFS, or multi member BTRFS. The mover is configured to transfer files between the primary parity pool and the share defined pool.

 

Each user share can be configured to write files initially to the main array or one of the pools, and to move the files or leave them. It's a little more complex than that, but not much.

 

The traditional definition of "caching" only loosely applies, if at all.

 

Originally when the main parity array was the only option and writes were slower than typical network speeds, a temporary fast write location was added, and the mover cleared it out overnight. The files never existed on both places, and subsequent access was in place, so writes to an already existing file were always slow. That was the "cache" drive concept, and it has been modified and added to ever since, but true caching was never a thing.

[Jlarimore]

Thank you for the detailed explanation. I think I get it. I guess my question is, how much do parity calculations slow down the writing to my parity protected storage pool? Are they going to slow down writes from 6600MB/s to 6300MB/s or from 6600MB/s to 300MB/s. If it's the latter, I can see how I might still want to use a mover. If it's minimal, why waste a drive on it, right?

[trurl]

Most of the write performance hit for HDD array is due to rotation so wouldn't apply to all SSD array. It still has to read before it can write to figure out parity update, but doesn't have to wait for the sector to come back around before it can write.

[trurl]

https://wiki.unraid.net/Manual/Storage_Management#Array_Write_Modes