Skip to contentSkip to conversation

sched/numa: Detect 'big' processes

tip:sched/numa tip-bot for Peter Zijlstra 1 participant 1 message Sep 28, 2012

Subsystems (2)
lore
Share & tools
lore
#0 lore raw
Commit-ID:  1b9fc936e56d62dfbb12ab4651c0432e91c10e2a
Gitweb:     http://git.kernel.org/tip/1b9fc936e56d62dfbb12ab4651c0432e91c10e2a
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Mon, 16 Jul 2012 13:08:54 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 27 Sep 2012 17:04:48 +0200

sched/numa: Detect 'big' processes

Detect 'big' processes for which the one home-node per process isn't going to work as desired.

The current policy for such tasks is to ignore them entirely and put the home-node back to -1 (no preference) so they'll behave as if none of this NUMA home node awareness is there.

The current heuristic for determining if a task is 'big' is if its consuming more than 1/2 a node's worth of cputime. We might want to add a term here looking at the RSS of the process and compare this against the available memory per node.

Since we now do multiple things from the task_work thing, we need to extend the state to determine which of the things we're there for -- a change in numa node, or a periodic poll of 'big'-ness.

---
 include/linux/mm_types.h |    1 +
 include/linux/sched.h    |    4 +-
 kernel/sched/core.c      |   12 ++++--
 kernel/sched/fair.c      |   96 ++++++++++++++++++++++++++++++++++++++-------
 4 files changed, 93 insertions(+), 20 deletions(-)
include/linux/mm_types.h +1 -0
index f407966..930c006 100644
@@ -405,6 +405,7 @@ struct mm_struct {
 	struct cpumask cpumask_allocation;
 #endif
 #ifdef CONFIG_SCHED_NUMA
+	unsigned int  numa_big;
 	unsigned long numa_next_scan;
 #endif
 	struct uprobes_state uprobes_state;
include/linux/sched.h +3 -1
index 63dde76..b8f7461 100644
@@ -1521,8 +1521,10 @@ struct task_struct {
 #endif
 #ifdef CONFIG_SCHED_NUMA
 	int node;			/* task home node   */
-	int node_last;			/* home node filter */
+	int node_curr, node_last;	/* home node filter */
 	u64 node_stamp;			/* migration stamp  */
+	u64 numa_runtime_stamp;
+	u64 numa_walltime_stamp;
 	unsigned long numa_contrib;
 #endif /* CONFIG_SCHED_NUMA */
 
kernel/sched/core.c +8 -4
index fda809d..6668b0d 100644
@@ -1535,13 +1535,18 @@ static void __sched_fork(struct task_struct *p)
 #endif
 
 #ifdef CONFIG_SCHED_NUMA
-	if (p->mm && atomic_read(&p->mm->mm_users) == 1)
+	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
+		p->mm->numa_big = 0;
 		p->mm->numa_next_scan = jiffies;
+	}
 
 	p->node = -1;
+	p->node_curr = -1;
 	p->node_last = -1;
 	p->node_stamp = 0ULL;
-#endif /* CONFIG_NUMA */
+	p->numa_runtime_stamp = 0;
+	p->numa_walltime_stamp = local_clock();
+#endif /* CONFIG_SCHED_NUMA */
 }
 
 /*
@@ -6024,8 +6029,7 @@ void sched_setnode(struct task_struct *p, int node)
 	if (running)
 		p->sched_class->put_prev_task(rq, p);
 
-	p->node = node;
-	p->node_last = node;
+	p->node = p->node_curr = p->node_last = node;
 
 	if (running)
 		p->sched_class->set_curr_task(rq);
kernel/sched/fair.c +81 -15

137 lines not on this page · raw message