This happening for a long time, since Windows 10, updated virtIO several times. Current running version 100.90.104.21700 (2/23/2022).

 

The strange thing is, it's not related to the SSD cache. But entirely when mover is running VM virtio network goes down to 4 MB/S. IN/OUT the unraid.

 

Per example, if, I mount an external share from another computer in Unraid GUI, and start copying to unraid when mover is running I got 110 MB/S probably maxing the network. So there is no bottleneck on the SSD cache (2XSATA SSD in RAID-0).

 

Tweaked iothreadpin on the VM, with no improvements. I'm kinda baffled, where the bottleneck is, Running Ryzen 1700. CPU usage when mover is running is negligible. 

 

Of course I tested it without any load. And the issue remains.

 

Any clues, where the bottleneck exists?

 

 

Edited September 22, 20223 yr by mpiva

I partially fix the issue, following Matias Hueber's "Performance optimizations for gaming on virtual machines (mathiashueber.com)"

 

on his site, mentioned the topology of a Ryzen Processor .

 

 

If you see his configuration it uses the second CCX for the guest

 

The strange thing, is the ordering is the same in the web UI. Which means cpuset = CPU number in UI. With the above all the HTs will be light up.

 

But notice the CPU pairing is CPU 0 and CPU 8, CPU 1 and CPU 9.

 

When in the above graphics the pairing is [CPU 0, CPU 1] [CPU 2, CPU 3]

 

 

So my question is, which one of this are wrong?

 

It seems the UI is showing the wrong pairing. I used the UI to create the cpuset, and seems that created the bottleneck.

 

IDK: if this also is replicated on CPU Pinning but since the UI is the same I guess so.  

 

 

Edited September 23, 20223 yr by mpiva