Hi All,

 

I don't even know how to properly categorize this issue, so I'll try the General

 

Anyway, my current is setup is 2x1TB M.2 for Cache (1TB usable) and then I have 2x20TB for Parity and 6 drives for array (3 are 16TB and 3 are 8 TB) making it 72TB usable.

 

SABNZBd is setup to download first to Cache, and then mover will move the data to array. This works excellent. When downloading I hit my max speed of just under 100MB/s. All is well. 

The issue ONLY arises when and if I am downloading something that's over 1TB. I'll get Cache low space alerts and all that. so if I want to download something that's 1TB I need to go into the data and set it up to use array instead of cache. Well then the download speed fluctuates from a few KB/s to 50-60-MB/s and it constantly goes up and down. For example:

 

 

 I am thinking it's something to do with the I/O since I am downloading directly to Array. Not sure why that would matter since I have 6 disks in array and even before Unraid I'd download stuff directly to HDDs without any issues whatsoever. I am positive the I/O can handle the speed of my max download, like I mentioned 100MB/s.

 

Again, if I am downloading much smaller files to Cache, it flies. The post-processing also happens super fast and all is well. But with 1TB files I cannot download them into the cache (I tried and halfway thru I'd pause the SAB, then kick off the Mover, wait but since half data is downloaded it already started the extraction and because it's not done the mover doesn't move those to the array, until it's fully processed.

 

What I am confused about is why downloading directly to the array causes soooo much D/L variance....

 

Thanks in advance for all your help. 

You have found the exact reason that a cache drive was added to Unraid in the first place.  There is no way around this slow-writing problem when writing directly to the array.   (I did a quick calculation and you are getting a 1TB of data every 3 hours...)  

 

You can speed things up a bit by enabling "Turbo-Write".  See here:

 

https://forums.unraid.net/topic/50397-turbo-write/

 

I recall reading the other day, that there many be some additional relief for delays caused by FUSE when we get to Linux kernel 6.20 but there will still be the physical limitations imposed by head movements and rotation delays of hard drives.

Thank you for that. It was driving me insane :)). I guess the only way around it is either to break the cache and have 2x1TB but not mirroring or to add larger capacity ie. 2x2TB and mirror.