Skip to contentSkip to conversation

sched/fair: add comments for group_type and balancing at SD_NUMA level

PATCH Applied Vincent Guittot 4 participants 5 messages Nov 12, 2019 – Nov 18, 2019 Scheduler

lore
Share & tools
lore

Vincent Guittot

author maintainer
#0 lore raw

PATCH 1/1 sched/fair: add comments for group_type and balancing at SD_NUMA level

Add comments to describe each state of goup_type and to add some details about the load balance at NUMA level.

---
 kernel/sched/fair.c | 29 ++++++++++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
kernel/sched/fair.c +28 -1
index c93d534..268e441 100644
@@ -6986,11 +6986,34 @@ enum fbq_type { regular, remote, all };
  * group. see update_sd_pick_busiest().
  */
 enum group_type {
+	/*
+	 * The group has spare capacity that can be used to process more work.
+	 */
 	group_has_spare = 0,
+	/*
+	 * The group is fully used and the tasks don't compete for more CPU
+	 * cycles. Nevetheless, some tasks might wait before running.
+	 */
 	group_fully_busy,
+	/*
+	 * One task doesn't fit with CPU's capacity and must be migrated on a
+	 * more powerful CPU.
+	 */
 	group_misfit_task,
+	/*
+	 * One local CPU with higher capacity is available and task should be
+	 * migrated on it instead on current CPU.
+	 */
 	group_asym_packing,
+	/*
+	 * The tasks affinity prevents the scheduler to balance the load across
+	 * the system.
+	 */
 	group_imbalanced,
+	/*
+	 * The CPU is overloaded and can't provide expected CPU cycles to all
+	 * tasks.
+	 */
 	group_overloaded
 };
 
@@ -8608,7 +8631,11 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
 
 	/*
 	 * Try to use spare capacity of local group without overloading it or
-	 * emptying busiest
+	 * emptying busiest.
+	 * XXX Spreading tasks across numa nodes is not always the best policy
+	 * and special cares should be taken for SD_NUMA domain level before
+	 * spreading the tasks. For now, load_balance() fully relies on
+	 * NUMA_BALANCING and fbq_classify_group/rq to overide the decision.
 	 */
 	if (local->group_type == group_has_spare) {
 		if (busiest->group_type > group_fully_busy) {
Signature · 1 line
2.7.4
#1 lore raw

Hi Vincent,

Quoted 4 lines · Vincent Guittot

Add comments to describe each state of goup_type and to add some details about the load balance at NUMA level.

Suggestions/nits below. There's a bit of duplication with existing comments (e.g. the nice blob atop sg_imbalanced()), but I think it can't hurt to have the few extra lines you're introducing.

---

kernel/sched/fair.c +8 -10
index bfdcaf91b325..ec93ebd02352 100644
@@ -6955,28 +6955,26 @@ enum fbq_type { regular, remote, all };
  * group. see update_sd_pick_busiest().
  */
 enum group_type {
-	/*
-	 * The group has spare capacity that can be used to process more work.
-	 */
+	/* The group isn't significantly pressured and can be used to process more work */
 	group_has_spare = 0,
 	/*
 	 * The group is fully used and the tasks don't compete for more CPU
-	 * cycles. Nevetheless, some tasks might wait before running.
+	 * cycles. Nevertheless, some tasks might wait before running.
 	 */
 	group_fully_busy,
 	/*
-	 * One task doesn't fit with CPU's capacity and must be migrated on a
-	 * more powerful CPU.
+	 * (SD_ASYM_CPUCAPACITY only) One task doesn't fit on its CPU's
+	 * capacity and must be migrated to a CPU of higher capacity.
 	 */
 	group_misfit_task,
 	/*
-	 * One local CPU with higher capacity is available and task should be
-	 * migrated on it instead on current CPU.
+	 * (SD_ASYM_PACKING only) One local CPU with higher capacity is
+	 * available and task should be migrated to it.
 	 */
 	group_asym_packing,
 	/*
-	 * The tasks affinity prevents the scheduler to balance the load across
-	 * the system.
+	 * The tasks affinity previously prevented the scheduler from balancing
+	 * load across the system.
 	 */
 	group_imbalanced,
 	/*
#2 lore raw

PATCH v2 1/1 sched/fair: add comments for group_type and balancing at SD_NUMA level

Quoted 43 lines · Valentin Schneider

Hi Vincent,

Quoted 4 lines · Vincent Guittot

Add comments to describe each state of goup_type and to add some details about the load balance at NUMA level.

Suggestions/nits below. There's a bit of duplication with existing comments (e.g. the nice blob atop sg_imbalanced()), but I think it can't hurt to have the few extra lines you're introducing.

---

kernel/sched/fair.c +5 -8
index bfdcaf91b325..ec93ebd02352 100644
@@ -6955,28 +6955,26 @@ enum fbq_type { regular, remote, all };
  * group. see update_sd_pick_busiest().
  */
 enum group_type {
-	/*
-	 * The group has spare capacity that can be used to process more work.
-	 */
+	/* The group isn't significantly pressured and can be used to process more work */
 	group_has_spare = 0,
 	/*
 	 * The group is fully used and the tasks don't compete for more CPU
-	 * cycles. Nevetheless, some tasks might wait before running.
+	 * cycles. Nevertheless, some tasks might wait before running.
 	 */
 	group_fully_busy,
 	/*
-	 * One task doesn't fit with CPU's capacity and must be migrated on a
-	 * more powerful CPU.
+	 * (SD_ASYM_CPUCAPACITY only) One task doesn't fit on its CPU's
+	 * capacity and must be migrated to a CPU of higher capacity.
 	 */
 	group_misfit_task,
 	/*
-	 * One local CPU with higher capacity is available and task should be
-	 * migrated on it instead on current CPU.
+	 * (SD_ASYM_PACKING only) One local CPU with higher capacity is

kernel/sched/fair.c enum fbq_type { regular, remote, all }; · line 6979

Quoted 10 lines
kernel/sched/fair.c +3 -2
+	 * available and task should be migrated to it.
 	 */
 	group_asym_packing,
 	/*
-	 * The tasks affinity prevents the scheduler to balance the load across
-	 * the system.
+	 * The tasks affinity previously prevented the scheduler from balancing
+	 * load across the system.
 	 */
 	group_imbalanced,

Thanks - I did a few more fixes and updates to the comments, this is how it ended up looking like (full patch below):

/*
* 'group_type' describes the group of CPUs at the moment of load balancing. *
* The enum is ordered by pulling priority, with the group with lowest priority
* first so the group_type can simply be compared when selecting the busiest
* group. See update_sd_pick_busiest().
*/
enum group_type {
/* The group has spare capacity that can be used to run more tasks. */ group_has_spare = 0,
/*
* The group is fully used and the tasks don't compete for more CPU
* cycles. Nevertheless, some tasks might wait before running. */
group_fully_busy,
/*
* SD_ASYM_CPUCAPACITY only: One task doesn't fit with CPU's capacity
* and must be migrated to a more powerful CPU.
*/
group_misfit_task,
/*
* SD_ASYM_PACKING only: One local CPU with higher capacity is available,
* and the task should be migrated to it instead of running on the
* current CPU.
*/
group_asym_packing,
/*
* The tasks' affinity constraints previously prevented the scheduler
* from balancing the load across the system.
*/
group_imbalanced,
/*
* The CPU is overloaded and can't provide expected CPU cycles to all
* tasks.
*/
group_overloaded
};

I also added your Acked-by, which I think was implicit? :)

Thanks,

Ingo

=====>
From: Vincent Guittot <[email protected]>
Date: Tue, 12 Nov 2019 15:50:43 +0100
Subject: [PATCH] sched/fair: Add comments for group_type and balancing at SD_NUMA level

Add comments to describe each state of goup_type and to add some details about the load balance at NUMA level.

[ Valentin Schneider: Updates to the comments. ]
[ mingo: Other updates to the comments. ]

---
 kernel/sched/fair.c | 35 +++++++++++++++++++++++++++++++----
 1 file changed, 31 insertions(+), 4 deletions(-)
kernel/sched/fair.c +31 -4
index 2fc08e7d9cd6..1f93d96dd06b 100644
@@ -6980,17 +6980,40 @@ static unsigned long __read_mostly max_load_balance_interval = HZ/10;
 enum fbq_type { regular, remote, all };
 
 /*
- * group_type describes the group of CPUs at the moment of the load balance.
+ * 'group_type' describes the group of CPUs at the moment of load balancing.
+ *
  * The enum is ordered by pulling priority, with the group with lowest priority
- * first so the groupe_type can be simply compared when selecting the busiest
- * group. see update_sd_pick_busiest().
+ * first so the group_type can simply be compared when selecting the busiest
+ * group. See update_sd_pick_busiest().
  */
 enum group_type {
+	/* The group has spare capacity that can be used to run more tasks.  */
 	group_has_spare = 0,
+	/*
+	 * The group is fully used and the tasks don't compete for more CPU
+	 * cycles. Nevertheless, some tasks might wait before running.
+	 */
 	group_fully_busy,
+	/*
+	 * SD_ASYM_CPUCAPACITY only: One task doesn't fit with CPU's capacity
+	 * and must be migrated to a more powerful CPU.
+	 */
 	group_misfit_task,
+	/*
+	 * SD_ASYM_PACKING only: One local CPU with higher capacity is available,
+	 * and the task should be migrated to it instead of running on the
+	 * current CPU.
+	 */
 	group_asym_packing,
+	/*
+	 * The tasks' affinity constraints previously prevented the scheduler
+	 * from balancing the load across the system.
+	 */
 	group_imbalanced,
+	/*
+	 * The CPU is overloaded and can't provide expected CPU cycles to all
+	 * tasks.
+	 */
 	group_overloaded
 };
 
@@ -8589,7 +8612,11 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
 
 	/*
 	 * Try to use spare capacity of local group without overloading it or
-	 * emptying busiest
+	 * emptying busiest.
+	 * XXX Spreading tasks across NUMA nodes is not always the best policy
+	 * and special care should be taken for SD_NUMA domain level before
+	 * spreading the tasks. For now, load_balance() fully relies on
+	 * NUMA_BALANCING and fbq_classify_group/rq to override the decision.
 	 */
 	if (local->group_type == group_has_spare) {
 		if (busiest->group_type > group_fully_busy) {
#3 lore raw
Quoted 3 lines · Ingo Molnar

Thanks - I did a few more fixes and updates to the comments, this is how it ended up looking like (full patch below):

[...]

LGTM, thanks!

Quoted 2 lines

I also added your Acked-by, which I think was implicit? :)

Hah, I'm not used to handing those out, but sure!

#4 lore raw

[tip: sched/core] sched/fair: Add comments for group_type and balancing at SD_NUMA level

In reply to Vincent Guittot
The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     a9723389cc759c891d481de271ac73eeaa123bcb
Gitweb:        https://git.kernel.org/tip/a9723389cc759c891d481de271ac73eeaa123bcb
Author:        Vincent Guittot <[email protected]>
AuthorDate:    Tue, 12 Nov 2019 15:50:43 +01:00
Committer:     Ingo Molnar <[email protected]>
CommitterDate: Mon, 18 Nov 2019 14:33:12 +01:00

sched/fair: Add comments for group_type and balancing at SD_NUMA level

Add comments to describe each state of goup_type and to add some details about the load balance at NUMA level.

[ Valentin Schneider: Updates to the comments. ]
[ mingo: Other updates to the comments. ]

---
 kernel/sched/fair.c | 35 +++++++++++++++++++++++++++++++----
 1 file changed, 31 insertions(+), 4 deletions(-)
kernel/sched/fair.c +31 -4
index 2fc08e7..1f93d96 100644
@@ -6980,17 +6980,40 @@ static unsigned long __read_mostly max_load_balance_interval = HZ/10;
 enum fbq_type { regular, remote, all };
 
 /*
- * group_type describes the group of CPUs at the moment of the load balance.
+ * 'group_type' describes the group of CPUs at the moment of load balancing.
+ *
  * The enum is ordered by pulling priority, with the group with lowest priority
- * first so the groupe_type can be simply compared when selecting the busiest
- * group. see update_sd_pick_busiest().
+ * first so the group_type can simply be compared when selecting the busiest
+ * group. See update_sd_pick_busiest().
  */
 enum group_type {
+	/* The group has spare capacity that can be used to run more tasks.  */
 	group_has_spare = 0,
+	/*
+	 * The group is fully used and the tasks don't compete for more CPU
+	 * cycles. Nevertheless, some tasks might wait before running.
+	 */
 	group_fully_busy,
+	/*
+	 * SD_ASYM_CPUCAPACITY only: One task doesn't fit with CPU's capacity
+	 * and must be migrated to a more powerful CPU.
+	 */
 	group_misfit_task,
+	/*
+	 * SD_ASYM_PACKING only: One local CPU with higher capacity is available,
+	 * and the task should be migrated to it instead of running on the
+	 * current CPU.
+	 */
 	group_asym_packing,
+	/*
+	 * The tasks' affinity constraints previously prevented the scheduler
+	 * from balancing the load across the system.
+	 */
 	group_imbalanced,
+	/*
+	 * The CPU is overloaded and can't provide expected CPU cycles to all
+	 * tasks.
+	 */
 	group_overloaded
 };
 
@@ -8589,7 +8612,11 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
 
 	/*
 	 * Try to use spare capacity of local group without overloading it or
-	 * emptying busiest
+	 * emptying busiest.
+	 * XXX Spreading tasks across NUMA nodes is not always the best policy
+	 * and special care should be taken for SD_NUMA domain level before
+	 * spreading the tasks. For now, load_balance() fully relies on
+	 * NUMA_BALANCING and fbq_classify_group/rq to override the decision.
 	 */
 	if (local->group_type == group_has_spare) {
 		if (busiest->group_type > group_fully_busy) {