tower-diagnostics-20260327-2203.zip

I'm having some strange behaviour that I'm having trouble explaining.

For a couple years now my Frigate docker container was writing to a 4 TB SMR drive as an unassigned device.

With the amount of cameras I have it's about 500 GB/day of footage that gets stored. This worked fine until recently the drive started getting written to VERY slowly, like 4 MB/s max throughput, which means I was dropping recordings on the floor.

So I decided to take a 500 GB SSD as a cache pool, and put that in front of the SMR. I reformatted the SMR. After a day the new SSD needed to run the mover to dump data onto the 4 TB drive, and was still getting max 4-6 MB/s.

Ok, so maybe this SMR drive is really bad or it's cache has gotten corrupted somehow, fine.

So I drop in an 8 TB Western Digital Ultrastar which has more than adequate performance. It is now the spinning disk pool for the SSD cache pool.

Exact same problem. Frigate is writing in total about 15 MB/s, SSD fills up in a day. Mover starts running. The mover seems to only be able to run at a speed of ~12-15 MB/s (coincidence probably).

Since the mover, and Frigate writing to the cache pool are seemingly around the same speed, 24 hours later, the mover is STILL running, and the SSD is staying at approximately 70% full.

Running performance tests on each drive individually and they perform exactly as I'd expect. (within their advertised characteristics)

What's going on here?

What typical size are those files?

Also what do you get if you do

pv /mnt/frigate-ss/file > /mnt/frigate-spinning/file

All the chunks are about 10 MB.

pv /mnt/frigate-ssd/frigate_video/recordings/2026-03-28/08/Frontdoor/03.09.mp4 > /mnt/frigate-spinning/frigate_video/03.09.mp4

7.57MiB 0:00:00 [ 667MiB/s] [================================================================================================================>] 100%

Doing things manually seems to get me the exact performance I expect. The mover, which is still running some 48 hours later, seems to be slow. Same behaviour after waking up. SSD drive is at 93% utilization, mover still moving from SSD to spinning disk at about 10 MB/s, slowly moving to the spinning disk.

10MB is very small for the mover, mover is not efficient with many small files because it checks if every file is open before the move, this adds a lot of overhead with small files, not very noticeable with large ones, but in this case I would recommend using a user script with rsync instead, for example; it should be much faster.

On 3/29/2026 at 5:20 AM, JorgeB said:

10MB is very small for the mover, mover is not efficient with many small files because it checks if every file is open before the move, this adds a lot of overhead with small files, not very noticeable with large ones, but in this case I would recommend using a user script with rsync instead, for example; it should be much faster.

User script using rsync seems to be working well, thanks!