Re: [RFC v3] mm: add page preemption

[Hillf Danton <hdanton@sina.com>]

On Sun, 3 Nov 2019 09:06:49 -0800 Matthew Wilcox wrote:
> On Sun, Nov 03, 2019 at 07:57:27PM +0800, Hillf Danton wrote:
> > The cpu preemption feature makes a task able to preempt other tasks
> > of lower priorities for cpu. It has been around for a while.
> > 
> > This work introduces task prio into page reclaiming in order to add
> > the page preemption feature that makes a task able to preempt other
> > tasks of lower priorities for page.
> > 
> > No page will be reclaimed on behalf of tasks of lower priorities
> > under pp, a two-edge feature that functions only under memory
> > pressure, laying a barrier to pages flowing to lower prio, and the
> > nice syscall is what users need to fiddle with it for instance if
> > they have a bunch of workloads to run in datacenter, and some
> > difficulty predicting the runtime working set size for every
> > individual workload which is sensitive to jitters in lru pages.
> > 
> > Currently pages are reclaimed without prio taken into account; pages
> > can be reclaimed from tasks of lower priorities on behalf of
> > higher-prio tasks and vice versa.
> > 
> > s/and vice versa/only/ is what we need to make pp by definition, but
> > it could not make a sense without prio introduced; otherwise we can
> > simply skip deactivating the lru pages based on prio comprison, and
> > work is done.
> > 
> > The introduction consists of two parts. On the page side, we have to
> > store the page owner task's prio in page, which needs an extra room the
> > size of the int type in the page struct. That room sounds impossible
> > without inflating the page struct size, and is walked around by making
> > pp depend on CONFIG_64BIT.
> 
> ... and !MEMCG.  Which means that your work is uninteresting because all
> the distros turn on CONFIG_MEMCG.

I have no idea which one they turn on by default, ext4, btrfs or xfs,
and why, but I think they feel free to do the right, or the left.
So do users to configure and build the kernel they need to power their
machines, a 4-socket server or a TV-top box.

> You still haven't given us any numbers.  Or a workload which actually
> benefits from this patch.

Though I do not know why it is turned on by distros and for what
workloads, I would like to try to run a couple of the workloads you
may have interest in.

Thanks
Hillf