Re: pinning cores

From: Jason Tubnor <jason_at_tubnor.net>
Date: Tue, 25 Nov 2025 21:23:28 UTC
On 26/11/25 00:53, void wrote:
>
>       -p vcpu:hostcpu
>
> but there's nothing for package. On my system, there's 2x "packages" 
> if I'm reading correctly - CPU0 & CPU1.
>
> The system has CPU E5-2690 v2 @ 3.00GHz x2 (hw.ncpu=20 in total - 10 
> on each CPU) and because SHT is off hw.vmm.maxcpu: 20
>
> If i set a vm to use -p 0:0 -p 1:10, then will it use the first core on
> the first physical package CPU0 and the first core on the second 
> physical package CPU1, if I understand correctly?
>
That is my understanding. You should see corresponding physical core 
load with your preferred monitoring tooling when you wire the guest up 
like that and push some load on it. When not pinned, the scheduler will 
move the virtual core workload to different cores (not ideal for some 
workloads).

However, be careful stretching across your NUMA like that, it is highly 
likely that you'll introduce significant latency into your guest as the 
guest OS will think the total workload is on the same package and your 
caching/memory access will be less than optimal for multi-core aware 
software.


Cheers