CPU Cores have no performance effect?

[Raymond Burns]

I was under the impression, through several forum posts, that assigning Virtual Cores to KVM guests have zero (0) impact on KVM performance.
However, after a quick test, I show that there is a big difference when assigning the proper cores.
What is the general rule of thumb when assigning cores?

Below is the test:
Host is (16 x Intel Xeon CPU E5620 @ 2.40GHz (2 Sockets))
Centos 6.6_x64 KVM using the KVM64 chip
Zoneminder server with 8GB of memory and 45 cameras configured

Test one is using a single socket and single core (Machine was very laggy):



Test two is using a single socket and 8 cores (Machine load and host load was considerably lower):


Host load:
Test 1 = 1.6
Test 2 = .92

 

[adamb]

If that was the case then why would they allow assigning of more than 1 core. More cores most definitely provides more performance.

 

[Raymond Burns]

http://forum.proxmox.com/threads/4661-CPU-Sockets-VS-Cores-Socket
http://forum.proxmox.com/threads/8053-CPU-Sockets-vs-CPU-cores?p=45836#post45836

Threads by developers themselves say otherwise.
That's two different Proxmox developers saying that virtual cores assigned to KVM makes no difference. Only for licensing

 

[adamb]

http://forum.proxmox.com/threads/4661-CPU-Sockets-VS-Cores-Socket
http://forum.proxmox.com/threads/8053-CPU-Sockets-vs-CPU-cores?p=45836#post45836

Threads by developers themselves say otherwise.
That's two different Proxmox developers saying that virtual cores assigned to KVM makes no difference. Only for licensing

Those are only referring to sockets vs cores. For example 2 sockets and 1 core is the same as 1 socket and 2 cores. Giving the VM more cores definitely increases cpu performance.

 

[mir]

In wmware 1 vcpu equals 1/4 of a physical cpu.