Is it worth dividing service vm by task during virtualization?

Question

VMware

Asked a person to look at my ESXI 6.5 installation with vm, this is what he saw

1.linux-ubuntu-16 = nat, vpn, SIP (asterisk), NGINX, fmpeg (under stream encoding) with Pass-through VGA
2. win2016 = AD , RDGate, DNS
3. workStation based on Win10 6 pieces each with PCI-Passthrough video

He told me: the rules of good manners dictate that the very Linux that under paragraph 1 should be divided into 3 parts, according to the services they provide:
Nat/vpn
HTTP/ FFMPEG (with Pass-through VGA)
SIP (Asterisk)
The explanation is simple: each machine should be more specialized for greater reliability. If something happens on one it will collapse the network. Because there all routing hangs on one machine. From the point of view of logic, everything seems logical.
The question is related to the fact that I try to reserve cores for each machine and give the maximum number for machines from the workStation group, because they consume a lot, a reserve is needed. There are 56 cores in total and in order to release this I need to find them and remove them from the rest of the machines. But I wonder who does what, and in general opinions on this topic. In working condition, the same linux sits with 8 cores (division in soft 2 processors / 4 cores for each), its round-the-clock loading is 60%. The main load is created by ffmpeg. That is, in any case, I will have to keep one with 8 cores

Answer 1

Virtualization is designed to overcommit. The fact that you have 28 physical processor cores (56 logical processors, including HT) does not mean that you can give virtual machines only 56 vCPUs in total. You can select twice as much, and everything will work, in the general case, even normally. So you can safely separate the services, and leave 8 cores in the virtual machine with ffmpeg. Of course, there is an overhead from unnecessary operating systems, but in this case it will be negligible.