Hi all,

 

Wondering if anyone can advise me. I've had a long enough standing issue, persistent through multiple Unraid releases, thereby I can't saturate write speeds to my cache drive. 95% of the time, anything I transfer across the LAN here will fluctuate between 50-70MB/s when transferring from my PC (Windows 10) to my Unraid Server (using x2 SSD cache drives in a BTRFS pool.)

 

If I bypass the SSD Cache Pool, and transfer directly to the array, I'll saturate the LAN speed (~112MB/s), which rules out any type of hardware/network issue. So it's actually slower for me to use my SSD write cache, which is frustrating. The Cache pool is made up of a Samsung 870 EVO 1TB, and a Samsung 860 EVO 1TB, so not exactly bargain bin drives.

 

I've read a fair bit on this issue, it's not uncommon, but seemingly quite individual at times. The most recent thing I tried was adding...

 

server min protocol = SMB3_11
client min protocol = SMB3_11

 

...to my Samba Extra Configuration settings in Unraid, but to no avail.

 

When I originally set up Unraid, I think one of the 6.7 or 6.8 releases, I had no such issues with transfer speeds, and could saturate the write cache without problem. I'm at my wits end now and unsure how to diagnose the issue. This happens for large single files, and not only for large amounts of small files where you'd expect to see this behavior.

 

Speeds are erratic, and hugely inconsistent. The very odd day though, it might not happen and I'll see ~114MB/s across the lan to the write cache, but it never lasts, and the next day I'll be right back to fluctuating/slower speeds.

1 hour ago, Unrayed said:

If I bypass the SSD Cache Pool, and transfer directly to the array, I'll saturate the LAN speed (~112MB/s)

This suggests a problem with the pool devices or the pool itself, is the pool old? Some time ago partitions on SSD devices were changed to be 1MiB aligned for better performance, also are you regularly trimming the pool?

2 hours ago, JorgeB said:

This suggests a problem with the pool devices or the pool itself, is the pool old? Some time ago partitions on SSD devices were changed to be 1MiB aligned for better performance, also are you regularly trimming the pool?

Thanks for the reply Jorge,

 

No the cache pool isn't old - the 870 EVO was bought this year and has ~500 hours logged (total lbas written = 1152436200), and the 860 EVO was bought maybe 2 years ago and has ~6000 hours on it (total lbas written = 22846605052). Neither have any reallocated sectors, and appear in perfect health. Both drives also report as 1MiB alligned.

 

Trimming is scheduled for a daily run @ 16:30, using Unraid's own Trim tool (settings, scheduler, trim). 

Apologies, I forgot to attach the server-diagnostics.zip file in the opening post....here it is now....

server-diagnostics-20221016-1307.zip

I would suggest removing one of the devices from the pool to test, if the same add it back to the pool, remove the other and test again, just to check if it's one of them slowing down the pool.

38 minutes ago, JorgeB said:

I would suggest removing one of the devices from the pool to test, if the same add it back to the pool, remove the other and test again, just to check if it's one of them slowing down the pool.

Thanks for the suggestion, I'll definitely try this. Will the pool function as normal for the purposes of testing with one of its drives removed yes?

Ok tried disabling each of the drives and performing an SMB transfer of a 5GB file from W10 to Unraid. With the 870 disabled, the file transferred at an inconsistent ~50/70MB/s. With the 860 disabled, the exact same file began transferring at full 113MB/s, but after about 2GB of it, the speed dropped down into the same low/fluctuating range as before. 

That still suggests a device problem, do you have a different SSD or even a fast HDD you could use to test?

9 hours ago, JorgeB said:

That still suggests a device problem, do you have a different SSD or even a fast HDD you could use to test?

I'll have to find something. How would I go about this, in terms of my current cache being a dual drive btrfs pool and a theoretical replacement/test drive likely being much smaller in capacity and only being a single drive? 

You can create a new pool, just for testing, then set a coupe of shares to use the new pool as cache=yes.

4 minutes ago, JorgeB said:

You can create a new pool, just for testing, then set a coupe of shares to use the new pool as cache=yes.

Perfect, leave it with me for a while and I'll try grab a small ssd from somewhere. Thanks for all your help with this