[Devel] Re: [ckrm-tech] containers development plans

[Paul Menage]

On 7/9/07, Balbir Singh <balbir at linux.vnet.ibm.com> wrote:
> > - splitting the memory and cpu isolation parts of cpusets into two
> > separate subsystems (still backwards-compatible)
>
> I see memory isolation using cpusets as very topology dependent
> and I am not sure if the model would work for memory controllers.

I wasn't suggesting making any changes to the page-based memory
controllers as part of this.

Currently in the mainline kernel, the cpumask and nodemask portions of
cpusets are essentially two mostly-independent modules that happen to
be coupled together in the same file and use the same process tracking
system (cpusets). Once we have generic process containers, splitting
this into a "cpusets" subsystem that handles all the cpumask portions
of the existing cpusets, and a "memsets" subsystem that handles all
the nodemask and memory migration portions would remove that coupling
and give more flexibility.

Paul