Well, yeah, good point to make but it is becoming harder and harder to find CMR drives, especially in 2.5".
ZFS should really adapt to this. Perhaps using bigger block sizes or something. Because SMR is not going away.
Well, yeah, good point to make but it is becoming harder and harder to find CMR drives, especially in 2.5".
ZFS should really adapt to this. Perhaps using bigger block sizes or something. Because SMR is not going away.
Although frankly at this point if you care about performance you're on NVME anyway.
I use two 5TB 2.5 SMR drives in ZFS mirror:
https://vermaden.wordpress.com/2019/04/03/silent-fanless-fre...
These drives can slow down to 30-40 MB/s when filled to 80% or more but I use that storage over WiFi which is at most 11-12MB/s which means the SMR problem does not exist for me.
If I would be using that storage over LAN the 30-40 MB/s in 'WORST CASE' is also not bad considering that maximum real life LAN speed over gigabit network is about 80-90 MB/s.
Its also not possible to get non-SMR large 2.5 drives. I use 2.5 drives as they are silent and they need very small amount of power comparing to 3.5 drives.
Rebuilds certainly are a problem but beyond that I think it's simply "be ok with slow drives". I've been operating a decently sized SMR pool for 3 years now with no major issues, including surviving two drive failures.
If you try to put random write workloads onto SMR you're gonna have a bad time no matter what you do. In my use case it's great, since this is effectively WORM storage writing giant files to disk all at once so I have effectively 0% fragmentation and all subsequent reads of the file tend to be sequential.
That said, this is for my personal use and lab projects. Not sure I'd go ZFS+SMR for a production workload.