CPU maxed out whilst reading a SATA drive

[MAM59]

This is a puzzling thing to me. Every time reading from a simple SATA Drive in the Array starts, CPU usage goes up (and stays) at almost 100% on all cores!

(This does NOT happen for NVMe drives!)

This would normally start troubling me, but there are strange side effects, that calm me down.

 

First of all, the machine is still very responsive. It does not lag or block, you can start many transfers on many drives, they all work as expected and rather fast too.

Then, take a look at the CPU Temp! 56,1°C is slightly below idle temp (which is around 52° here in summer), and far away from the 82° that it gets if a few cores get really under stress.

 

I see no errors, no warning and everything seems to work. still this 100% usage ...

 

(BTW, the machine has 4 different SATA controllers, each drive is attached to one of them, it happens on ALL, either onboard or inslot)

 

should  I be worried or is this simply a bad reading in 6.10.3???

 

f-diagnostics-20220707-2002.zip

[JorgeB]

top - 20:02:41 up 2 days, 11:58,  0 users,  load average: 91.94, 86.80, 80.17
Tasks: 493 total,   3 running, 490 sleeping,   0 stopped,   0 zombie
%Cpu(s):  3.8 us, 47.7 sy,  0.0 ni,  0.8 id, 46.2 wa,  0.0 hi,  1.5 si,  0.0 st
MiB Mem : 128561.5 total,    789.1 free,  26788.8 used, 100983.6 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.  99438.7 avail Mem

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
23186 imagema+  20   0 2695560  65296  29028 S 605.9   0.0 104:44.36 smbd

 

It's not a bad reading, though keep in mind that the dashboard includes IO wait, that doesn't really load the CPU, still load average is very high, as is the CPU usage by smb, what are you using to read from the disks?

[MAM59]

an initial full backup is running from a different machine over the lan. A simple "robocopy" from windows. It did not hurt whist copying the NVMe drives, but now it came to the first "normal" SATA disks and the load exploded instantly.

The transfer is rather slow, I guess it will take 3 days or so at this level.

 

[MAM59]

Tnx for your explanation, I came to the conclusion, that the robocopy was overrunning UNRAID with too many threads at the same time (default is using 8 threads concurrently).

Now I limited it to 1 thread only, you see the result and, as a positive side effect, the backup is now running much faster.

 

 

[MAM59]

some more improvements led to this result:

Turned off SMB Multichannel support and AIO

(transfer speed was doubled again, now it uses ~280Mb/s, the maximum the target drive can handle)

 

(all Screens show the same usage, permanent copy of huge files over lan with robocopy as the receiver)