Skip to contentSkip to conversation

sched,numa: ensure task_numa_migrate checks the preferred node

PATCH In review Rik van Riel 3 participants 3 messages Jun 4, 2014 – Jun 19, 2014 Scheduler

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched,numa: ensure task_numa_migrate checks the preferred node

The first thing task_numa_migrate does is check to see if there is CPU capacity available on the preferred node, in order to move the task there.

However, if the preferred node is all busy, we would skip considering that node for tasks swaps in the subsequent loop. This prevents NUMA convergence of tasks on busy systems.

However, swapping locations with a task on our preferred nid, when the preferred nid is busy, is perfectly fine.

The fix is to also look for a CPU on our preferred nid when it is totally busy.

This changes "perf bench numa mem -p 4 -t 20 -m -0 -P 1000" from not converging in 15 minutes on my 4 node system, to converging in 10-20 seconds.

---
This is a safe, simple patch to fix NUMA convergence, which is why it should go to -stable, IMHO.

 kernel/sched/fair.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
kernel/sched/fair.c +2 -3
index e5884d8..824e241 100644
@@ -1425,9 +1425,8 @@ static int task_numa_migrate(struct task_struct *p)
 	groupimp = group_weight(p, env.dst_nid, false) - groupweight;
 	update_numa_stats(&env.dst_stats, env.dst_nid);
 
-	/* If the preferred nid has capacity, try to use it. */
-	if (env.dst_stats.has_capacity)
-		task_numa_find_cpu(&env, taskimp, groupimp);
+	/* Try to find a spot on the preferred nid. */
+	task_numa_find_cpu(&env, taskimp, groupimp);
 
 	/* No space available on the preferred nid. Look elsewhere. */
 	if (env.best_cpu == -1) {
maintainer
#1 lore raw
Quoted 11 lines · Rik van Riel

The first thing task_numa_migrate does is check to see if there is CPU capacity available on the preferred node, in order to move the task there.

However, if the preferred node is all busy, we would skip considering that node for tasks swaps in the subsequent loop. This prevents NUMA convergence of tasks on busy systems.

However, swapping locations with a task on our preferred nid, when the preferred nid is busy, is perfectly fine.

Quoted 8 lines

The fix is to also look for a CPU on our preferred nid when it is totally busy.

This changes "perf bench numa mem -p 4 -t 20 -m -0 -P 1000" from not converging in 15 minutes on my 4 node system, to converging in 10-20 seconds.

Indeed so. Thanks Rik!

#2 lore raw

[tip:sched/core] sched/numa: Ensure task_numa_migrate() checks the preferred node

In reply to Rik van Riel
sched/numa: Ensure task_numa_migrate() checks the preferred node
Commit-ID:  a43455a1d572daf7b730fe12eb747d1e17411365
Gitweb:     http://git.kernel.org/tip/a43455a1d572daf7b730fe12eb747d1e17411365
Author:     Rik van Riel <[email protected]>
AuthorDate: Wed, 4 Jun 2014 16:09:42 -0400
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 18 Jun 2014 18:29:57 +0200

sched/numa: Ensure task_numa_migrate() checks the preferred node

The first thing task_numa_migrate() does is check to see if there is CPU capacity available on the preferred node, in order to move the task there.

However, if the preferred node is all busy, we would skip considering that node for tasks swaps in the subsequent loop. This prevents NUMA convergence of tasks on busy systems.

However, swapping locations with a task on our preferred nid, when the preferred nid is busy, is perfectly fine.

The fix is to also look for a CPU on our preferred nid when it is totally busy.

This changes "perf bench numa mem -p 4 -t 20 -m -0 -P 1000" from not converging in 15 minutes on my 4 node system, to converging in 10-20 seconds.

---
 kernel/sched/fair.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
kernel/sched/fair.c +2 -3
index fea7d33..8fbb011 100644
@@ -1302,9 +1302,8 @@ static int task_numa_migrate(struct task_struct *p)
 	groupimp = group_weight(p, env.dst_nid) - groupweight;
 	update_numa_stats(&env.dst_stats, env.dst_nid);
 
-	/* If the preferred nid has free capacity, try to use it. */
-	if (env.dst_stats.has_free_capacity)
-		task_numa_find_cpu(&env, taskimp, groupimp);
+	/* Try to find a spot on the preferred nid. */
+	task_numa_find_cpu(&env, taskimp, groupimp);
 
 	/* No space available on the preferred nid. Look elsewhere. */
 	if (env.best_cpu == -1) {