nano_cpu and cpu_shares

In the doc, there’s (for container.run):

cpu_shares (int) – CPU shares (relative weight).
nano_cpus (int) – CPU quota in units of 10-9 CPUs.

It looks that if I want 50% CPU, nano_cpus should be 0.5 * 1e9. However, that doesn’t seem to be what I see in production (if I specify 0.9 * 1e9, I will get maybe 30% CPU utilization rate, on a single-core machine). Also, if I use a 4-core CPU, and I want to use half of it (say 2 cores), what should I specify?

Similarly, cpu_shares is an int. What does relative weight mean? Is that on a scale of 100, so if I use 50% of the CPU I should put 50? Again, what if I have a 4-core CPU and want to use half of it?

I would love to get more clarity on these, and ideally be added to the docs.

Thanks!

[shin-]

Here is a more comprehensive source: https://docs.docker.com/engine/admin/resource_constraints/#cpu which better describes what happens in a multi CPU scenario.

I don’t know that there is a way to say “use any 2 CPUs”, but you may wanna ask on the forums as it’s a bit outside my area of expertise.

[danqing]

What does “counterpart” mean? And cpus (cpu-period, cpu-quota) only mentions 1 CPU scenario. What happens when there are more CPUs?

And is there an option to assign CPU numbers, not specific CPUs? Say I have two containers, each wanting 2 cores. Do I need to assign specific cores (1,2; 3,4) to them, or is there a way to tell them to use 2 cores each?

Thanks!