Skip to contentSkip to conversation

sched: avg_overlap decay

PATCH In review Peter Zijlstra 2 participants 3 messages Mar 10, 2009 – Mar 11, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched: avg_overlap decay

Mike, are you good with this patch as it stands?

---
Subject: sched: avg_overlap decay
From: Mike Galbraith <[email protected]>
Date: Tue Mar 10 19:08:11 CET 2009

avg_overlap is used to measure the runtime overlap of the waker and wakee.

However, when a process changes behaviour, eg a pipe becomes un-congested and we don't need to go to sleep after a wakeup for a while, the avg_overlap value grows stale.

When running we use the avg runtime between preemption as a measure for avg_overlap since the amount of runtime can be correlated to cache footprint.

The longer we run, the less likely we'll be wanting to be migrated to another CPU.

---
 kernel/sched.c |   24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)
kernel/sched.c +23 -1
Index: linux-2.6/kernel/sched.c
===================================================================
@@ -4692,6 +4692,28 @@ static inline void schedule_debug(struct
 #endif
 }
 
+static void put_prev_task(struct rq *rq, struct task_struct *prev)
+{
+	if (prev->state == TASK_RUNNING) {
+		u64 runtime = prev->se.sum_exec_runtime;
+
+		runtime -= prev->se.prev_sum_exec_runtime;
+		runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
+
+		/*
+		 * In order to avoid avg_overlap growing stale when we are
+		 * indeed overlapping and hence not getting put to sleep, grow
+		 * the avg_overlap on preemption.
+		 *
+		 * We use the average preemption runtime because that
+		 * correlates to the amount of cache footprint a task can
+		 * build up.
+		 */
+		update_avg(&prev->se.avg_overlap, runtime);
+	}
+	prev->sched_class->put_prev_task(rq, prev);
+}
+
 /*
  * Pick up the highest-prio task:
  */
@@ -4768,7 +4790,7 @@ need_resched_nonpreemptible:
 	if (unlikely(!rq->nr_running))
 		idle_balance(cpu, rq);
 
-	prev->sched_class->put_prev_task(rq, prev);
+	put_prev_task(rq, prev);
 	next = pick_next_task(rq);
 
 	if (likely(prev != next)) {
#1 lore raw
Quoted 1 line · Peter Zijlstra

Mike, are you good with this patch as it stands?

Yes, works for me.

-Mike

Quoted 66 lines

---
Subject: sched: avg_overlap decay
From: Mike Galbraith <[email protected]>
Date: Tue Mar 10 19:08:11 CET 2009

avg_overlap is used to measure the runtime overlap of the waker and wakee.

However, when a process changes behaviour, eg a pipe becomes un-congested and we don't need to go to sleep after a wakeup for a while, the avg_overlap value grows stale.

When running we use the avg runtime between preemption as a measure for avg_overlap since the amount of runtime can be correlated to cache footprint.

The longer we run, the less likely we'll be wanting to be migrated to another CPU.

---
 kernel/sched.c |   24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

Index: linux-2.6/kernel/sched.c
===================================================================

kernel/sched.c +23 -1
@@ -4692,6 +4692,28 @@ static inline void schedule_debug(struct
 #endif
 }
 
+static void put_prev_task(struct rq *rq, struct task_struct *prev)
+{
+	if (prev->state == TASK_RUNNING) {
+		u64 runtime = prev->se.sum_exec_runtime;
+
+		runtime -= prev->se.prev_sum_exec_runtime;
+		runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
+
+		/*
+		 * In order to avoid avg_overlap growing stale when we are
+		 * indeed overlapping and hence not getting put to sleep, grow
+		 * the avg_overlap on preemption.
+		 *
+		 * We use the average preemption runtime because that
+		 * correlates to the amount of cache footprint a task can
+		 * build up.
+		 */
+		update_avg(&prev->se.avg_overlap, runtime);
+	}
+	prev->sched_class->put_prev_task(rq, prev);
+}
+
 /*
  * Pick up the highest-prio task:
  */
@@ -4768,7 +4790,7 @@ need_resched_nonpreemptible:
 	if (unlikely(!rq->nr_running))
 		idle_balance(cpu, rq);
 
-	prev->sched_class->put_prev_task(rq, prev);
+	put_prev_task(rq, prev);
 	next = pick_next_task(rq);
 
 	if (likely(prev != next)) {
#2 lore raw

[tip:sched/core] sched: add avg_overlap decay

In reply to Peter Zijlstra
Commit-ID:  df1c99d416500da8d26a4d78777467c53ee7689e
Gitweb:     http://git.kernel.org/tip/df1c99d416500da8d26a4d78777467c53ee7689e
Author:     "Mike Galbraith" <[email protected]>
AuthorDate: Tue, 10 Mar 2009 19:08:11 +0100
Commit:     Ingo Molnar <[email protected]>
CommitDate: Wed, 11 Mar 2009 11:31:50 +0100

sched: add avg_overlap decay

Impact: more precise avg_overlap metric - better load-balancing

avg_overlap is used to measure the runtime overlap of the waker and wakee.

However, when a process changes behaviour, eg a pipe becomes un-congested and we don't need to go to sleep after a wakeup for a while, the avg_overlap value grows stale.

When running we use the avg runtime between preemption as a measure for avg_overlap since the amount of runtime can be correlated to cache footprint.

The longer we run, the less likely we'll be wanting to be migrated to another CPU.

LKML-Reference: <1236709131.25234.576.camel@laptop>

---
 kernel/sched.c |   24 +++++++++++++++++++++++-
 1 files changed, 23 insertions(+), 1 deletions(-)
kernel/sched.c +23 -1
index af5cd1b..2f28351 100644
@@ -4620,6 +4620,28 @@ static inline void schedule_debug(struct task_struct *prev)
 #endif
 }
 
+static void put_prev_task(struct rq *rq, struct task_struct *prev)
+{
+	if (prev->state == TASK_RUNNING) {
+		u64 runtime = prev->se.sum_exec_runtime;
+
+		runtime -= prev->se.prev_sum_exec_runtime;
+		runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
+
+		/*
+		 * In order to avoid avg_overlap growing stale when we are
+		 * indeed overlapping and hence not getting put to sleep, grow
+		 * the avg_overlap on preemption.
+		 *
+		 * We use the average preemption runtime because that
+		 * correlates to the amount of cache footprint a task can
+		 * build up.
+		 */
+		update_avg(&prev->se.avg_overlap, runtime);
+	}
+	prev->sched_class->put_prev_task(rq, prev);
+}
+
 /*
  * Pick up the highest-prio task:
  */
@@ -4698,7 +4720,7 @@ need_resched_nonpreemptible:
 	if (unlikely(!rq->nr_running))
 		idle_balance(cpu, rq);
 
-	prev->sched_class->put_prev_task(rq, prev);
+	put_prev_task(rq, prev);
 	next = pick_next_task(rq);
 
 	if (likely(prev != next)) {