Vm whit Windows SVR2022 Lags

[PierGiorgio]

Good morning,I need help regarding a slowdown problem (probably caused by video problems) that I'm experiencing on a Windows SVR2022 VM. The server that hosts it is a few years old but is still a valid machine, it is an HP DL380 Gen10 dual Xeon Silver 4110 32 core processor, with 64GB of Ram and 4 SDD + 1TB for BK. Pharmacy management software is running on the VM, which is used in RDP by X users. In some parts of the software written in VB6 language there are colored grids and it seems that these colors create numerous lags, in fact if the colors are deactivated in the software the lags disappear. The vm is configured entirely with virtio including the video part, I also tried to passthrough with an Nvidia GT730 (it was just a test) but the result does not change. Any idea what the problem might be? Attached is the configuration of the vm.
Thank you

 

[cwt]

Is this a new VM? Build from scratch or imported?
The „grid“ occurs only in RDP sessions? Does it also show up when using NoVNC connections?

Some possible reasons:

- imported VM has driver problems with non-existent former vGPUs or VMWare tools are still installed
- buggy VirtIO drivers for 2022 (especially the V262)

 

[leesteken]

dual Xeon Silver 4110 32 core processor

It looks like a Xeon Silver 4110 has only 8 cores, which gives 16 threads and with two socket that's only 32 threads total. You are giving the VM 32 virtual cores but that leaves no slack for Proxmox, device emulator etc. Your lag is because you give the VM 100% of a resource, which is bad for latency and the opposite of virtualization best practices.
Best not the give the VM more than 8 cores and enable NUMA, and it will run much smoother. See this thread for a similar situation: https://forum.proxmox.com/threads/vm-problem-freezes-randomdly.158084/#post-724186

 

[PierGiorgio]

thank you both so much, I'll answer you both here:
cwt: they are new vm created on proxmox, no import, the problem occurs mostly in RDP but, when there is a load of users it also occurs locally with noVNC. I also suppose that it is a driver problem, if they were the problem I hope that everything will be solved.

leesteken: yes yes I know it has 8 cores, what you see is a screen of the many tests I'm doing, the only test I haven't done is to further lower the resources and do a test.
Otherwise I'm really trying everything