From: Valentin Schneider <valentin.schneider@arm.com>
To: Srikar Dronamraju <srikar@linux.vnet.ibm.com>,
Vincent Guittot <vincent.guittot@linaro.org>
Cc: linux-kernel@vger.kernel.org, mingo@redhat.com,
peterz@infradead.org, quentin.perret@arm.com,
dietmar.eggemann@arm.com, Morten.Rasmussen@arm.com,
pauld@redhat.com
Subject: Re: [PATCH 3/5] sched/fair: rework load_balance
Date: Fri, 26 Jul 2019 15:09:24 +0100 [thread overview]
Message-ID: <b98ae93b-80f7-a4ca-0c4d-9d6c166055a7@arm.com> (raw)
In-Reply-To: <20190726135852.GB7168@linux.vnet.ibm.com>
On 26/07/2019 14:58, Srikar Dronamraju wrote:
[...]
>> @@ -8357,72 +8318,115 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
>> if (busiest->group_type == group_imbalanced) {
>> /*
>> * In the group_imb case we cannot rely on group-wide averages
>> - * to ensure CPU-load equilibrium, look at wider averages. XXX
>> + * to ensure CPU-load equilibrium, try to move any task to fix
>> + * the imbalance. The next load balance will take care of
>> + * balancing back the system.
>> */
>> - busiest->load_per_task =
>> - min(busiest->load_per_task, sds->avg_load);
>> + env->src_grp_type = migrate_task;
>> + env->imbalance = 1;
>> + return;
>> }
>>
>> - /*
>> - * Avg load of busiest sg can be less and avg load of local sg can
>> - * be greater than avg load across all sgs of sd because avg load
>> - * factors in sg capacity and sgs with smaller group_type are
>> - * skipped when updating the busiest sg:
>> - */
>> - if (busiest->group_type != group_misfit_task &&
>> - (busiest->avg_load <= sds->avg_load ||
>> - local->avg_load >= sds->avg_load)) {
>> - env->imbalance = 0;
>> - return fix_small_imbalance(env, sds);
>> + if (busiest->group_type == group_misfit_task) {
>> + /* Set imbalance to allow misfit task to be balanced. */
>> + env->src_grp_type = migrate_misfit;
>> + env->imbalance = busiest->group_misfit_task_load;
>> + return;
>> }
>>
>> /*
>> - * If there aren't any idle CPUs, avoid creating some.
>> + * Try to use spare capacity of local group without overloading it or
>> + * emptying busiest
>> */
>> - if (busiest->group_type == group_overloaded &&
>> - local->group_type == group_overloaded) {
>> - load_above_capacity = busiest->sum_h_nr_running * SCHED_CAPACITY_SCALE;
>> - if (load_above_capacity > busiest->group_capacity) {
>> - load_above_capacity -= busiest->group_capacity;
>> - load_above_capacity *= scale_load_down(NICE_0_LOAD);
>> - load_above_capacity /= busiest->group_capacity;
>> - } else
>> - load_above_capacity = ~0UL;
>> + if (local->group_type == group_has_spare) {
>> + long imbalance;
>> +
>> + /*
>> + * If there is no overload, we just want to even the number of
>> + * idle cpus.
>> + */
>> + env->src_grp_type = migrate_task;
>> + imbalance = max_t(long, 0, (local->idle_cpus - busiest->idle_cpus) >> 1);
>
> Shouldnt this be?
> imbalance = max_t(long, 0, (busiest->idle_cpus - local->idle_cpus) >> 1);
I think it's the right way around - if busiest has more idle CPUs than local,
then we shouldn't balance (local is busier than busiest)
However, doesn't that lead to a imbalance of 0 when e.g. local has 1 idle
CPU and busiest has 0 ?. If busiest has more than one task we should try
to pull at least one.
next prev parent reply other threads:[~2019-07-26 14:09 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-07-19 7:58 [PATCH 0/5] sched/fair: rework the CFS load balance Vincent Guittot
2019-07-19 7:58 ` [PATCH 1/5] sched/fair: clean up asym packing Vincent Guittot
2019-07-19 7:58 ` [PATCH 2/5] sched/fair: rename sum_nr_running to sum_h_nr_running Vincent Guittot
2019-07-19 12:51 ` Peter Zijlstra
2019-07-19 13:44 ` Vincent Guittot
2019-07-26 2:17 ` Srikar Dronamraju
2019-07-26 8:41 ` Vincent Guittot
2019-07-19 7:58 ` [PATCH 3/5] sched/fair: rework load_balance Vincent Guittot
2019-07-19 12:52 ` Peter Zijlstra
2019-07-19 13:46 ` Vincent Guittot
2019-07-19 12:54 ` Peter Zijlstra
2019-07-19 14:02 ` Vincent Guittot
2019-07-20 11:31 ` Peter Zijlstra
2019-07-19 13:06 ` Peter Zijlstra
2019-07-19 13:57 ` Vincent Guittot
2019-07-19 13:12 ` Peter Zijlstra
2019-07-19 14:13 ` Vincent Guittot
2019-07-19 13:22 ` Peter Zijlstra
2019-07-19 13:55 ` Vincent Guittot
2019-07-25 17:17 ` Valentin Schneider
2019-07-26 9:01 ` Vincent Guittot
2019-07-26 10:41 ` Valentin Schneider
2019-07-26 12:30 ` Vincent Guittot
2019-07-26 14:01 ` Valentin Schneider
2019-07-26 14:47 ` Vincent Guittot
2019-07-29 14:28 ` Valentin Schneider
2019-07-26 13:58 ` Srikar Dronamraju
2019-07-26 14:09 ` Valentin Schneider [this message]
2019-07-26 14:42 ` Vincent Guittot
2019-07-31 13:43 ` Srikar Dronamraju
2019-07-31 15:37 ` Vincent Guittot
2019-07-19 7:58 ` [PATCH 4/5] sched/fair: use load instead of runnable load Vincent Guittot
2019-07-19 7:58 ` [PATCH 5/5] sched/fair: evenly spread tasks when not overloaded Vincent Guittot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=b98ae93b-80f7-a4ca-0c4d-9d6c166055a7@arm.com \
--to=valentin.schneider@arm.com \
--cc=Morten.Rasmussen@arm.com \
--cc=dietmar.eggemann@arm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=pauld@redhat.com \
--cc=peterz@infradead.org \
--cc=quentin.perret@arm.com \
--cc=srikar@linux.vnet.ibm.com \
--cc=vincent.guittot@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®