Re: [PATCH RESEND v8 1/2] sched/numa: introduce per-cgroup NUMA locality info

[çè]

On 2020/2/21 äå11:47, Peter Zijlstra wrote:
> On Fri, Feb 21, 2020 at 02:20:10PM +0000, Mel Gorman wrote:
>> I fully acknowledge that this may have value for sysadmins and may be a
>> good enough reason to merge it for environments that typically build and
>> configure their own kernels. I doubt that general distributions would
>> enable it but that's a guess.
> 
> OTOH, many sysadmins seem to 'rely' on BPF scripts and other such fancy
> things these days.
> 
>  ( of course, we have the open question on what happens when we break
>    one of those BPF 'important' scripts ... )
> 
> My main reservation with this patch is that it exposes, to userspace, an
> ABI that is very hard to interpret and subject to implementation
> details.
> 
> So while it can be disabled; people who have it enabled might suddenly
> complain when we change the meaning/interpretation/whatever of these
> magic numbers.
> 
> Michael; you seem to have ignored the tracepoint / BPF angle earlier in
> this discussion; that is not something that could/would work for you?

At very beginning I think these fancy stuff may consume too much resources
them selves, so just as you said, ignored the possibility :-P

But now I understand there is a big gap here, which require a much more general
way to evaluate the NUMA platform, I'll try to follow this way see if there
are any practical approach instead~

Regards,
Michael Wang

>