Running n CPU-bound tasks on an n CPUs platform:
- with asymmetric CPU capacity
- not having SD_SHARE_PKG_RESOURCES flag set at the DIE
sched domain level (i.e. not DynamIQ systems)
might result in a task placement where two tasks run on a big CPU
and none on a little CPU. This placement could be more optimal by
using all CPUs.
Testing platform:
Juno-r2:
- 2 big CPUs (1-2), maximum capacity of 1024
- 4 little CPUs (0,3-5), maximum capacity of 383
Testing workload ([1]):
Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
is affine to a CPU, except for:
- one little CPU which is left idle.
- one big CPU which has 2 tasks affine.
After the 100ms (step 2), remove the cpumask affinity.
Before patch:
During step 2, the load balancer running from the idle CPU tags sched
domains as:
- little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
4 CPUs sched-domain, and the idle CPU provides enough spare
capacity.
- big CPUs: 'group_overloaded'. Indeed, 3 tasks run on a 2 CPUs
sched-domain, so the following path is used:
group_is_overloaded()
\-if (sgs->sum_nr_running <= sgs->group_weight) return true;
The following path which would change the migration type to
'migrate_task' is not taken:
calculate_imbalance()
\-if (env->idle != CPU_NOT_IDLE && env->imbalance == 0)
as the local group has some spare capacity, so the imbalance
is not 0.
The migration type requested is 'migrate_util' and the busiest
runqueue is the big CPU's runqueue having 2 tasks (each having a
utilization of 512). The idle little CPU cannot pull one of these
task as its capacity is too small for the task. The following path
is used:
detach_tasks()
\-case migrate_util:
\-if (util > env->imbalance) goto next;
After patch:
As the number of failed balancing attempts grows (with
'nr_balance_failed'), progressively make it easier to migrate
a big task to the idling little CPU. A similar mechanism is
used for the 'migrate_load' migration type.
Improvement:
Running the testing workload [1] with the step 2 representing
a ~10s load for a big CPU:
Before patch: ~19.3s
After patch: ~18s (-6.7%)
Similar issue reported at:
https://lore.kernel.org/lkml/[email protected]/
v1:
https://lore.kernel.org/all/[email protected]/
Suggested-by: Vincent Guittot <[email protected]>
Signed-off-by: Pierre Gondois <[email protected]>
---
kernel/sched/fair.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index df348aa55d3c..53c18fd23ae7 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8907,7 +8907,7 @@ static int detach_tasks(struct lb_env *env)
case migrate_util:
util = task_util_est(p);
- if (util > env->imbalance)
+ if (shr_bound(util, env->sd->nr_balance_failed) > env->imbalance)
goto next;
env->imbalance -= util;
--
2.25.1
On Fri, 24 Nov 2023 at 16:33, Pierre Gondois <[email protected]> wrote:
>
> Running n CPU-bound tasks on an n CPUs platform:
> - with asymmetric CPU capacity
> - not having SD_SHARE_PKG_RESOURCES flag set at the DIE
> sched domain level (i.e. not DynamIQ systems)
Nit: SD_SHARE_PKG_RESOURCES is never set at the DIE level. In case of
DynamIQ systems, all CPUs are in the same MC level which has
SD_SHARE_PKG_RESOURCES flag
> might result in a task placement where two tasks run on a big CPU
> and none on a little CPU. This placement could be more optimal by
> using all CPUs.
>
> Testing platform:
> Juno-r2:
> - 2 big CPUs (1-2), maximum capacity of 1024
> - 4 little CPUs (0,3-5), maximum capacity of 383
>
> Testing workload ([1]):
> Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
> is affine to a CPU, except for:
> - one little CPU which is left idle.
> - one big CPU which has 2 tasks affine.
> After the 100ms (step 2), remove the cpumask affinity.
>
> Before patch:
> During step 2, the load balancer running from the idle CPU tags sched
> domains as:
> - little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
> 4 CPUs sched-domain, and the idle CPU provides enough spare
> capacity.
> - big CPUs: 'group_overloaded'. Indeed, 3 tasks run on a 2 CPUs
> sched-domain, so the following path is used:
> group_is_overloaded()
> \-if (sgs->sum_nr_running <= sgs->group_weight) return true;
>
> The following path which would change the migration type to
> 'migrate_task' is not taken:
> calculate_imbalance()
> \-if (env->idle != CPU_NOT_IDLE && env->imbalance == 0)
> as the local group has some spare capacity, so the imbalance
> is not 0.
>
> The migration type requested is 'migrate_util' and the busiest
> runqueue is the big CPU's runqueue having 2 tasks (each having a
> utilization of 512). The idle little CPU cannot pull one of these
> task as its capacity is too small for the task. The following path
> is used:
> detach_tasks()
> \-case migrate_util:
> \-if (util > env->imbalance) goto next;
>
> After patch:
> As the number of failed balancing attempts grows (with
> 'nr_balance_failed'), progressively make it easier to migrate
> a big task to the idling little CPU. A similar mechanism is
> used for the 'migrate_load' migration type.
>
> Improvement:
> Running the testing workload [1] with the step 2 representing
> a ~10s load for a big CPU:
> Before patch: ~19.3s
> After patch: ~18s (-6.7%)
>
> Similar issue reported at:
> https://lore.kernel.org/lkml/[email protected]/
>
> v1:
> https://lore.kernel.org/all/[email protected]/
>
> Suggested-by: Vincent Guittot <[email protected]>
> Signed-off-by: Pierre Gondois <[email protected]>
Reviewed-by: Vincent Guittot <[email protected]>
> ---
> kernel/sched/fair.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index df348aa55d3c..53c18fd23ae7 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8907,7 +8907,7 @@ static int detach_tasks(struct lb_env *env)
> case migrate_util:
> util = task_util_est(p);
>
> - if (util > env->imbalance)
> + if (shr_bound(util, env->sd->nr_balance_failed) > env->imbalance)
> goto next;
>
> env->imbalance -= util;
> --
> 2.25.1
>
On 24/11/2023 16:33, Pierre Gondois wrote:
> Running n CPU-bound tasks on an n CPUs platform:
> - with asymmetric CPU capacity
> - not having SD_SHARE_PKG_RESOURCES flag set at the DIE
nit: DIE is now called PKG on tip sched/core.
f577cd57bfaa - sched/topology: Rename 'DIE' domain to 'PKG' (2023-10-12
Peter Zijlstra)
> sched domain level (i.e. not DynamIQ systems)
> might result in a task placement where two tasks run on a big CPU
> and none on a little CPU. This placement could be more optimal by
> using all CPUs.
>
> Testing platform:
> Juno-r2:
> - 2 big CPUs (1-2), maximum capacity of 1024
> - 4 little CPUs (0,3-5), maximum capacity of 383
>
> Testing workload ([1]):
> Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
> is affine to a CPU, except for:
> - one little CPU which is left idle.
> - one big CPU which has 2 tasks affine.
> After the 100ms (step 2), remove the cpumask affinity.
>
> Before patch:
> During step 2, the load balancer running from the idle CPU tags sched
> domains as:
> - little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
> 4 CPUs sched-domain, and the idle CPU provides enough spare
> capacity.
What is meant by 'idle CPU provides enough spare capacity? I thought the
task (util_avg ~ 512_ does not fit on the sched group [1,3-5] when we
consider util_avg/capacity (383)
The calculated imbalance of ~350 is too small for the task-size and
that's why we need the 'shr_bound(util, env->sd->nr_balance_failed)' to
let the task load-balance if nr_balance_failed = 2?
[...]
> Similar issue reported at:
> https://lore.kernel.org/lkml/[email protected]/
>
> v1:
> https://lore.kernel.org/all/[email protected]/
>
> Suggested-by: Vincent Guittot <[email protected]>
> Signed-off-by: Pierre Gondois <[email protected]>
Even though this cures only classical big.LITTLE it might have a
positive effect on today's Arm DynamIQ Android systems with Phantom SDs
when running benchmarks like Geekbench.
[...]
Reviewed-by: Dietmar Eggemann <[email protected]>
On 11/29/23 09:55, Dietmar Eggemann wrote:
> On 24/11/2023 16:33, Pierre Gondois wrote:
>> Running n CPU-bound tasks on an n CPUs platform:
>> - with asymmetric CPU capacity
>> - not having SD_SHARE_PKG_RESOURCES flag set at the DIE
>
> nit: DIE is now called PKG on tip sched/core.
>
> f577cd57bfaa - sched/topology: Rename 'DIE' domain to 'PKG' (2023-10-12
> Peter Zijlstra)
>
>> sched domain level (i.e. not DynamIQ systems)
>> might result in a task placement where two tasks run on a big CPU
>> and none on a little CPU. This placement could be more optimal by
>> using all CPUs.
>>
>> Testing platform:
>> Juno-r2:
>> - 2 big CPUs (1-2), maximum capacity of 1024
>> - 4 little CPUs (0,3-5), maximum capacity of 383
>>
>> Testing workload ([1]):
>> Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
>> is affine to a CPU, except for:
>> - one little CPU which is left idle.
>> - one big CPU which has 2 tasks affine.
>> After the 100ms (step 2), remove the cpumask affinity.
>>
>> Before patch:
>> During step 2, the load balancer running from the idle CPU tags sched
>> domains as:
>> - little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
>> 4 CPUs sched-domain, and the idle CPU provides enough spare
>> capacity.
>
> What is meant by 'idle CPU provides enough spare capacity? I thought the
> task (util_avg ~ 512_ does not fit on the sched group [1,3-5] when we
> consider util_avg/capacity (383)
Right, I meant that when evaluating the 'group_type', there is enough spare
capacity when summing the utilization of CPUs in the the MC sched domain:
---
group_has_capacity()
{
[...]
if ((sgs->group_capacity * 100) >
(sgs->group_util * imbalance_pct))
return true;
[...]
}
---
>
> The calculated imbalance of ~350 is too small for the task-size and
> that's why we need the 'shr_bound(util, env->sd->nr_balance_failed)' to
> let the task load-balance if nr_balance_failed = 2?
Yes exact, the tasks are too big and cannot fit this imbalance value
(representing the available spare capacity in the little CPUs in this case).
'shr_bound(...)' allows to progressively reduce the size of the tasks and
allow migrations after having tried to balance 'nr_balance_failed' times.
>
> [...]
>
>> Similar issue reported at:
>> https://lore.kernel.org/lkml/[email protected]/
>>
>> v1:
>> https://lore.kernel.org/all/[email protected]/
>>
>> Suggested-by: Vincent Guittot <[email protected]>
>> Signed-off-by: Pierre Gondois <[email protected]>
>
> Even though this cures only classical big.LITTLE it might have a
> positive effect on today's Arm DynamIQ Android systems with Phantom SDs
> when running benchmarks like Geekbench.
>
> [...]
>
> Reviewed-by: Dietmar Eggemann <[email protected]>
>
On 29/11/2023 11:48, Pierre Gondois wrote:
>
>
> On 11/29/23 09:55, Dietmar Eggemann wrote:
>> On 24/11/2023 16:33, Pierre Gondois wrote:
[...]
>>> Testing workload ([1]):
>>> Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
>>> is affine to a CPU, except for:
>>> - one little CPU which is left idle.
>>> - one big CPU which has 2 tasks affine.
>>> After the 100ms (step 2), remove the cpumask affinity.
>>>
>>> Before patch:
>>> During step 2, the load balancer running from the idle CPU tags sched
>>> domains as:
>>> - little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
>>> 4 CPUs sched-domain, and the idle CPU provides enough spare
>>> capacity.
>>
>> What is meant by 'idle CPU provides enough spare capacity? I thought the
>> task (util_avg ~ 512_ does not fit on the sched group [1,3-5] when we
>> consider util_avg/capacity (383)
>
> Right, I meant that when evaluating the 'group_type', there is enough spare
> capacity when summing the utilization of CPUs in the the MC sched domain:
>
> ---
> group_has_capacity()
> {
> [...]
> if ((sgs->group_capacity * 100) >
> (sgs->group_util * imbalance_pct))
> return true;
> [...]
> }
I see. But doesn't group_has_capacity() already return true because of
`if (sgs->sum_nr_running < sgs->group_weight)` in this case (3 < 4)?
Maybe you can make this clearer when sending the v3 with the Reviewed-By
tags?
> ---
>
>>
>> The calculated imbalance of ~350 is too small for the task-size and
>> that's why we need the 'shr_bound(util, env->sd->nr_balance_failed)' to
>> let the task load-balance if nr_balance_failed = 2?
>
> Yes exact, the tasks are too big and cannot fit this imbalance value
> (representing the available spare capacity in the little CPUs in this
> case).
>
> 'shr_bound(...)' allows to progressively reduce the size of the tasks and
> allow migrations after having tried to balance 'nr_balance_failed' times.
[...]
On 11/24/23 16:33, Pierre Gondois wrote:
> Running n CPU-bound tasks on an n CPUs platform:
> - with asymmetric CPU capacity
> - not having SD_SHARE_PKG_RESOURCES flag set at the DIE
> sched domain level (i.e. not DynamIQ systems)
> might result in a task placement where two tasks run on a big CPU
> and none on a little CPU. This placement could be more optimal by
> using all CPUs.
>
> Testing platform:
> Juno-r2:
> - 2 big CPUs (1-2), maximum capacity of 1024
> - 4 little CPUs (0,3-5), maximum capacity of 383
>
> Testing workload ([1]):
> Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
> is affine to a CPU, except for:
> - one little CPU which is left idle.
> - one big CPU which has 2 tasks affine.
> After the 100ms (step 2), remove the cpumask affinity.
>
> Before patch:
> During step 2, the load balancer running from the idle CPU tags sched
> domains as:
> - little CPUs: 'group_has_spare'. Indeed, 3 CPU-bound tasks run on a
> 4 CPUs sched-domain, and the idle CPU provides enough spare
> capacity.
> - big CPUs: 'group_overloaded'. Indeed, 3 tasks run on a 2 CPUs
> sched-domain, so the following path is used:
> group_is_overloaded()
> \-if (sgs->sum_nr_running <= sgs->group_weight) return true;
>
> The following path which would change the migration type to
> 'migrate_task' is not taken:
> calculate_imbalance()
> \-if (env->idle != CPU_NOT_IDLE && env->imbalance == 0)
> as the local group has some spare capacity, so the imbalance
> is not 0.
>
> The migration type requested is 'migrate_util' and the busiest
> runqueue is the big CPU's runqueue having 2 tasks (each having a
> utilization of 512). The idle little CPU cannot pull one of these
> task as its capacity is too small for the task. The following path
> is used:
> detach_tasks()
> \-case migrate_util:
> \-if (util > env->imbalance) goto next;
>
> After patch:
> As the number of failed balancing attempts grows (with
> 'nr_balance_failed'), progressively make it easier to migrate
> a big task to the idling little CPU. A similar mechanism is
> used for the 'migrate_load' migration type.
>
> Improvement:
> Running the testing workload [1] with the step 2 representing
> a ~10s load for a big CPU:
> Before patch: ~19.3s
> After patch: ~18s (-6.7%)
>
> Similar issue reported at:
> https://lore.kernel.org/lkml/[email protected]/
>
> v1:
> https://lore.kernel.org/all/[email protected]/
>
> Suggested-by: Vincent Guittot <[email protected]>
> Signed-off-by: Pierre Gondois <[email protected]>
> ---
> kernel/sched/fair.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index df348aa55d3c..53c18fd23ae7 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8907,7 +8907,7 @@ static int detach_tasks(struct lb_env *env)
> case migrate_util:
> util = task_util_est(p);
>
> - if (util > env->imbalance)
> + if (shr_bound(util, env->sd->nr_balance_failed) > env->imbalance)
FWIW, this did help but not entirely when I tried it in the past
https://lore.kernel.org/lkml/20230718161829.ws3vn3ufnod6kpxh@airbuntu/
And nit, the comment in that posted diff was useful.
No objections anyway but I suspect there's room for further improvement if
somebody still cares.
Cheers
--
Qais Yousef