Re: [PATCH v2 0/3] newidle_balance() PREEMPT_RT latency mitigations

[Mike Galbraith]

On Mon, 2021-05-03 at 11:33 -0500, Scott Wood wrote:
> On Sun, 2021-05-02 at 05:25 +0200, Mike Galbraith wrote:
> > On Sat, 2021-05-01 at 17:03 -0500, Scott Wood wrote:
> > > On Thu, 2021-04-29 at 09:12 +0200, Vincent Guittot wrote:
> > > > Hi Scott,
> > > >
> > > > On Thu, 29 Apr 2021 at 01:28, Scott Wood <swood@xxxxxxxxxx> wrote:
> > > > > These patches mitigate latency caused by newidle_balance() on large
> > > > > systems when PREEMPT_RT is enabled, by enabling interrupts when the
> > > > > lock
> > > > > is dropped, and exiting early at various points if an RT task is
> > > > > runnable
> > > > > on the current CPU.
> > > > >
> > > > > On a system with 128 CPUs, these patches dropped latency (as
> > > > > measured by
> > > > > a 12 hour rteval run) from 1045us to 317us (when applied to
> > > > > 5.12.0-rc3-rt3).
> > > >
> > > > The patch below has been queued for v5.13 and removed the update of
> > > > blocked load what seemed to be the major reason for long preempt/irq
> > > > off during newly idle balance:
> > > > https://lore.kernel.org/lkml/20210224133007.28644-1-vincent.guittot@xxxxxxxxxx/
> > > >
> > > > I would be curious to see how it impacts your cases
> > >
> > > I still get 1000+ ms latencies with those patches applied.
> >
> > If NEWIDLE balancing migrates one task, how does that manage to consume
> > a full *millisecond*, and why would that only be a problem for RT?
> >
> > 	-Mike
> >
> > (rt tasks don't play !rt balancer here, if CPU goes idle, tough titty)
>
> Determining which task to pull is apparently taking that long (again, this
> is on a 128-cpu system).  RT is singled out because that is the config that
> makes significant tradeoffs to keep latencies down (I expect this would be
> far from the only possible 1ms+ latency on a non-RT kernel), and there was
> concern about the overhead of a double context switch when pulling a task to
> a newidle cpu.

What I think has be going on is that you're running a synchronized RT
load, many CPUs go idle as a thundering herd, and meet at focal point
busiest.  What I was alluding to was that preventing such size scale
pile-ups would be way better than poking holes in it for RT to try to
sneak through.  If pile-up it is, while not particularly likely, the
same should happen with normal tasks, wasting cycles generating heat.

The main issue I see with these patches is that the resulting number is
still so gawd awful as to mean "nope, not rt ready", making the whole
exercise look a bit like a noop.

	-Mike