toastmatic

Looks like it was the 870 QVO drive. Switched back to an 850 EVO and haven't had a problem. Solved.

I'm on 6.8.3 but it looks like my testing wasn't thorough enough. After another round of sequential write testing it looks like the QVO drive's write buffer is the limitation. I'll be swapping it for an 850 later to see if performance improves. Thanks!

Hello, First time post, I've been looking around but haven't found an answer to this issue: I am currently running a 1TB Samsung 860 EVO and 1TB Samsung 870 QVO as mirrored btrfs cache drives. When testing each drive independently they will write 50GB+ at 112MB/s over a gigabit network, but when rebuilt as a mirrored array they will regularly (every ~6GB) drop to nearly 0 KB/s before returning to their normal speed. Some additional details: Each drive has current firmware and writes consistently as a single cache drive Files are confirmed as definitely writing to the cache drives Cache FS is BTRFS (non-encrypted) confirmed issue also exists while transferring from storage array to cache in Krusader (limited to ~150MB\s from platter read, still intermittently drops) tested with all VMS and dockers disabled E5-2690 V2, 32gb ECC DDR3, B75 chipset, 2x1TB SSD cache array on SATA3, 2x8TB XFS storage array on SATA 2, 650W PSU no errors, various scans are all clean, environment runs flawlessly otherwise Most of the incidents I have found online were due to reconstructive write on the storage array, but I could not find any examples of this on the cache. Any thoughts? Thanks.