CPU Limit with Hosting Practices

[Shadow Sysop]

I have a a couple of dozen clients utilizing VMs on my Proxmox cluster. I've made a practice of always setting resource limits for VMs, at half the core count. So if it is an 8 core VM, I will set the CPU Limit to 4. I have a client using 4 Windows VMs, maxing out the CPUs with something Twitch & Discord related (according to NTOPNG traffic). This puts allot of stress on the nodes with his VMs, so I'm considering setting his CPU limit to 3. When doing this, my load averages became a bit more reasonable. My question is, how negatively would the performance of apps like Discord and Twitch be affected by this change? I want to spin up a VM to test it myself, but the load averages on all my nodes is high enough that I really can't unless I can tighten those CPU limits. Any thoughts or suggestions are appreciated.

 

[Stefan_R]

The question isn't "how can you get the load average lower", but rather "what problems do you see with a high load average". In general, provisioning a machine to the max should not cause any issues - VM processes are time-sliced, so every VM should get it's share of performance according to it's "CPU units" setting.

The benefit of doing it with "CPU units" instead of "CPU limit", is that under low-load situations, a single VM can access all of the performance of it's assigned cores. With "CPU limit", you're essentially limiting your cores anyway.

For desktop use-cases I believe using a "CPU limit" may actually be detrimental to performance, as programs on the guest *believe* they have access to a certain number of fast cores, but can never actually use them fully.

Edit: See also our documentation: https://pve.proxmox.com/pve-docs/chapter-qm.html#qm_cpu