CPU Hotplug Bug

C

cshill

Member
May 8, 2024
83
11
8
Good Morning/Evening,

I was testing out CPU Hotplug as an option and added the rule based on this documentation, Proxmox Hotplug

Vcpu's were not working correctly when I tested it so I removed the rule and thought everything will be fine. I now have setup a lot of windows 11 users and multiple different softwares are having challenges performing. I tested these softwares on a vm on a different server and it runs MUCH faster.

From what others have stated in forum posts is that the CPU hotplug capability can break the cpu optimization under the hood. Is there anything I can do to correct this other than reformat the system?

Edit: My plan is to only reformat the Proxmox OS and then remount the old datastores.
 
Last edited:
Reddit Post about CPU Hotplug
Reddit Post 2

1. I was trying to use the x86-64-v2 versions in the past but it was causing issues, can't remember exact ones but I didn't have any problems with "host"

2. There has been discussion that Hotplug has been working fine for people, some people it is not responding correctly, and that top reddit link states it breaks certain optimizations underneath the hood. I was in the middle category for a while where it didn't matter what I did, The windows 11 OS was not receiving an increase of more vcpu's. When that happened I just said, "eh it would have been nice but I have to move on." I removed the rule and it has been rebooted since the rule change. I had for weeks people saying the system is very slow but I thought it was something they were doing in the background.

3. These people that use the VM's are programmers utilizing Windows 11 with WSL Version 2 so it was a pain in the butt to find these CPU Flags that work for them. These flags were put on both VM's, the 'bad' server VM's and 'good' server VM's. The good server ran a downloader tool in probably a 1/10th the time with half the threads allocated to the VM. I know it's not a network issue about pulling the files, and it's not a disk issue. It was a thought that maybe 5 or so VM's on the same disk were maxing it out but the good server still has about 5 VMs on it so nothing crazy.

I sat down with one of them and tried to test the CPU and it said it was performing fine. The way I tested it was with CPU-Z and I would bench test it and compare among the same models of servers. CPU-Z tells me the bad server performs at a 40-50 point loss compared to that of the VM's on the good server.
 
You must log in or register to reply here.
Top Bottom