CFT/CFR: NUMA policy branch

[Adrian Chadd]

On 8 July 2015 at 13:43, Andriy Gapon <avg at freebsd.org> wrote:
> On 08/07/2015 22:18, Alfred Perlstein wrote:
>>
>>
>> On 7/7/15 11:38 PM, Adrian Chadd wrote:
>>> There's a phabricator review. It's not up to date, because:
>>>
>>> * it broke for a while, and
>>> * kib requested he be sent patches, not a phabricator review.
>>>
>>
>>
>> So Kib is complaining that his feedback is getting lost, but refuses to use a
>> review tracker?
>
> How about phabricator losing diffs? Would that be a valid complaint?

Hi,

Let's not get side tracked. I've invited a variety of people to review
and comment on this stuff. Some people want it in reviews.freebsd.org,
some want it via diffs. Different people want work done in different
units of work.

My plan is to get this into "good enough" state to throw into -HEAD. I
don't even care if in 12 months it's completely replaced with an
alternate implementation and/or API. What i care about right now is
getting the basic pieces in place so further work and experimentation
can be done. Right now the entry limit to evaluating any NUMA things
on FreeBSD is "you don't, without numa.diff", and that's unacceptable.
I completely expect that it'll change over the course of a few years.
But the fact we still don't have even the most basic userland exposed
API for controlling things is IMHO unacceptable and reflects poorly on
us as a community.

So, I'm looking for less nit-picking and more "this is wrong, you
should do this." A lot of kibs responses have been errors on my part
that I hadn't picked up on and weren't exposed during testing. I'm
looking for more of those. I haven't yet gone over Garrett's comments
in too much depth; I'll look at that tonight if I don't fall asleep
first.

The important thing here is to try and finally move the default
available functionality along a little bit so people can get
interested and start using this and contribute their own work.

Thanks,



-adrian