I set up a 2x 1Tb 2.5" SSD cache to use as a "work" space. The 2 drives are connected to a single port from my 6GB/s HBA card, with the 2nd port feeding my drive shelf. This cache is the write cache for the data array, plus the storage space for downloads/processing from Sabnzbd. I was also using this space as a transcode location for Tdarr to use. 

Everything was working fantastic until a couple of days ago when I started having issues with the download speed from Sab, with it being a roller-coaster ride, from sub-1MB/s to the normal of 24MB/s, which is what I have it set to use as the max speed. The peaks and valleys were very consistent. I initially thought that it was my internet connection, of course, especially since it's cellular (T-Mobile Home Internet). Multiple reboots of the modem with no change. Through a lot of troubleshooting, everything came down to the write speeds of the Work cache, with the folder speed-tests in Sab being in the same speeds as the download speeds, which means that the bottleneck is the Work cache itself. 
 

I noticed that when I paused the Tdarr transcodes, the Work speeds were back to normal. Ok, easy fix - change Tdarr to use the ramdisk for transcoding. I verified that Tdarr is in fact using the ramdisk, but when Tdarr is running, the Work cache is still running slow. I ran the hdparm utility on the 2 drives that make up the Work cache and this is the results:
 

root@Neo:/tmp/TdarrCache# hdparm -tT /dev/sdc

/dev/sdc:
 Timing cached reads:   13192 MB in  1.99 seconds = 6627.77 MB/sec
 Timing buffered disk reads: 480 MB in  3.01 seconds = 159.52 MB/sec
root@Neo:/tmp/TdarrCache# hdparm -tT /dev/sdd

/dev/sdd:
 Timing cached reads:   17142 MB in  1.99 seconds = 8619.70 MB/sec
 Timing buffered disk reads: 736 MB in  3.01 seconds = 244.60 MB/sec

 

This is driving me nuts, especially since I don't know what's causing the bottleneck now. Neither of the 2 drives have any Smart or Scrub errors whatsoever. All I know is that a 2x Raid-0 should be infinitely faster than a sub-Gbit download.

 

I'm still not sure on the WHY, but I *think* I fixed the issue. I changed the Work pools filesystem from a BTRFS Raid 0 to ZFS Raid 0. So far, the speeds are looking much more consistent. I haven't restarted Tdarr back up yet, so I don't know with absolute certainty that the problem has been solved or not.

Nope - Tdarr destroys the IO when it's transcoding.... I was planning on adding 2 more drives to the Work pool. What I might do is create a separate 2x Transcode pool, instead.