Re: [PATCH 1/3] sched, balancing: Update rq->max_idle_balance_cost whenever newidle balance is attempted

[Jason Low <jason.low2@hp.com>]

On Fri, 2014-04-25 at 10:42 +0530, Preeti U Murthy wrote:
> I agree with this. However I am concerned with an additional point that
> I have mentioned in my reply to Peter's mail on this thread.
> 
> Should we verify if rq->next_balance update is independent of
> pulled_tasks? sd->balance_interval is changed during load_balance() and
> rq->next_balance should perhaps consider that?

Hi Preeti,

I agree that we may want to consider having rq->next balance update be
independent of pulled_task. As you mentioned, load_balance() can modify
the balance_interval.

There are a few things I'm wondering if we would need to also add then:

1. In the case that this_rq->avg_idle < sysctl_sched_migration_cost, we
   would need to also traverse the domains to properly compute
   next_balance (without the sd->busy_factor) as we would be going idle.
   Otherwise, next_balance could get set to jiffies + HZ while the
   CPU goes idle.

2. In the domain traversal, when we pulled_task, we might want to
   multiply interval by sd->busy_factor because the rq will remain busy.

3. If this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost, then we
   may still want to compute next_balance, rather than simply break out
   of the sched domain traversal loop. This is also to avoid having
   the next_balance = jiffies + HZ when a domain should rebalance
   less than 1 second later.