I have a Core i5 3570K and 32GB ram with 2 SSD cache and 21TB array with Parity. I notice that I can max out my CPU and get really bad performance when I move files (manually or mover) and I have sonarr/radarr/sabnzbd.  I also get interuptsion when using Jellyfin on my FireStick4k.   I feel like something is kinda messed up in my config.  I keep finding small things and fixing them (like moving libvert/dockers to cache from the array, turbo mode app) but I haven't been able to fix this.  Feels like I'm doing it wrong.  Might try moving the cache to xfs from btfrs. 

Anyone more knowledgeable take a look and see if something obvious is going on?   I see the iowait goes 75% often.

nas4x12-diagnostics-20200104-1355.zip

I believe there have been reports of Adata SSDs performing badly with btrfs raid, or just in general, you can try xfs cache to see if there's any difference.

Whats the best way to transition from a btrfs cache pool to a single xfs cache drive?  I have a Samsung SSD unused if that expedites things.

You can use this:

https://forums.unraid.net/topic/46802-faq-for-unraid-v6/?do=findComment&comment=511923

 

This was happening to me too and it's infuriating. Because of the way the Unraid works and the way sabnzbd works, it was bringing my computer down to its knees whenever a file decided to repair. And this is with a brand new 9600k, 32GB RAM, 6x2TB WD Blue (1 parity), and 1 evo 500GB for cache.

I tried forcing downloads only to cache, the docker image only, etc. The only thing that actually fixed it was to move downloads to their own drive outside the array / cache. I bought a 256GB nvme when I built the computer that was not in the raid / cache yet. Here is what ended up being fairly stable (at least so far):

• Installed Unassigned Devices Plugin. Formatted NVME as XFS then mounted as unassigned device.
• Sonarr / Radarr
  • /data -> /mnt/disks/{nvme}/downloads
  • /media -> /mnt/user/media/{tv/movies} (media is a share, setup as 'Cache Yes')
• SABnzbd
  • /data -> /mnt/disks/{nvme}/downloads. Both completed and in-progress mapped below that.

The nvme gets thrashed with the download, verification, repair (if needed), and extraction without taking IO from any of my other docker containers / etc. Just to make sure SABnzbd doesn't ever lock everything, I pegged that VM to 3 of my 6 cores too. Once a download completes sonarr/radarr moves it to the media share, which because it's going from SSD to SSD transfers it over to the cache at a fairly fast speed. Even downloading at the same time as mover is running doesn't seem to kill it anymore. I hope it works well this way.