Re: Re: Re: Re: Re: Re: [PATCH v1] sched/numa: add per-process numa_balancing

On Wed, Nov 17, 2021 at 03:07:43PM +0800, Gang Li wrote:

On 11/10/21 12:26 AM, Mel Gorman wrote:

quoted

Of those two, I agree with the second one, it would be tricky to implement
but the first one is less clear. This is based on an assumption. If prctl
exists to enable/disable NUMA baalancing, it's possible that someone
else would want to control NUMA balancing on a cgroup basis instead of
globally which would run into the same type of concerns -- different
semantics depending on the global tunable.

Hi!

You talk about the "semantics" of NUMA balancing between global, cgroup and
process. While I read the kernel doc "NUMA Memory Policy", it occur to me
that we may have a "NUMA Balancing Policy".

Since you are the reviewer of CONFIG_NUMA_BALANCING. I would like to discuss
the need for introducing "NUMA Balancing Policy" with you. Is this worth
doing?

It's a bit vague but if you wanted to put together the outline, I'd read
over it. Note that this was all in the context of trying to introduce an
API like

Disable/enable per-process numa balancing:
        prctl(PR_NUMA_BALANCING, PR_SET_NUMA_BALANCING, 0/1);

i.e. one that controlled both enabling and disabling. You also have
the option of introducing the NUMAB equivalent of PR_SET_THP_DISABLE --
an API that is explicitly about disabling *only*.

-- 
Mel Gorman
SUSE Labs