lkml.org 
[lkml]   [2013]   [Jul]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH v2] sched: smart wake-affine
Since v1:
1. Invoke record_wakee() at a better timing. (Thanks to PeterZ)
2. Use node-size instead of online-cpu-number
as the factor, to reduce overhead and make
logical more reasonable. (Thanks to PeterZ)
3. Name fix and more comments. (Thanks to PeterZ)
4. Tested again with the tip 3.10.0-rc7.

wake-affine stuff is always trying to pull wakee close to waker, by theory,
this will bring benefit if waker's cpu cached hot data for wakee, or the
extreme ping-pong case.

And testing show it could benefit hackbench 15% at most.

However, the whole stuff is somewhat blindly and time-consuming, some
workload therefore suffer.

And testing show it could damage pgbench 50% at most.

Thus, wake-affine stuff should be more smart, and realise when to stop
it's thankless effort.

This patch introduced 'nr_wakee_switch', which will be increased each
time the task switch it's wakee.

So a high 'nr_wakee_switch' means the task has more than one wakee, and
bigger the number, higher the wakeup frequency.

Now when making the decision on whether to pull or not, pay attention on
the wakee with a high 'nr_wakee_switch', pull such task may benefit wakee,
but also imply that waker will face cruel competition later, it could be
very cruel or very fast depends on the story behind 'nr_wakee_switch',
whatever, waker therefore suffer.

Furthermore, if waker also has a high 'nr_wakee_switch', imply that multiple
tasks rely on it, then waker's higher latency will damage all of them, pull
wakee seems to be a bad deal.

Thus, when 'waker->nr_wakee_switch / wakee->nr_wakee_switch' become higher
and higher, the deal seems to be worse and worse.

The patch therefore help wake-affine stuff to stop it's work when:

wakee->nr_wakee_switch > factor &&
waker->nr_wakee_switch > (factor * wakee->nr_wakee_switch)

The factor here is the node-size of current-cpu, so bigger node will lead
to more pull since the trial become more severe.

After applied the patch, pgbench show 40% improvement at most.

Test:
Tested with 12 cpu X86 server and tip 3.10.0-rc7.

pgbench base smart

| db_size | clients | tps | | tps |
+---------+---------+-------+ +-------+
| 22 MB | 1 | 10598 | | 10796 |
| 22 MB | 2 | 21257 | | 21336 |
| 22 MB | 4 | 41386 | | 41622 |
| 22 MB | 8 | 51253 | | 57932 |
| 22 MB | 12 | 48570 | | 54000 |
| 22 MB | 16 | 46748 | | 55982 | +19.75%
| 22 MB | 24 | 44346 | | 55847 | +25.93%
| 22 MB | 32 | 43460 | | 54614 | +25.66%
| 7484 MB | 1 | 8951 | | 9193 |
| 7484 MB | 2 | 19233 | | 19240 |
| 7484 MB | 4 | 37239 | | 37302 |
| 7484 MB | 8 | 46087 | | 50018 |
| 7484 MB | 12 | 42054 | | 48763 |
| 7484 MB | 16 | 40765 | | 51633 | +26.66%
| 7484 MB | 24 | 37651 | | 52377 | +39.11%
| 7484 MB | 32 | 37056 | | 51108 | +37.92%
| 15 GB | 1 | 8845 | | 9104 |
| 15 GB | 2 | 19094 | | 19162 |
| 15 GB | 4 | 36979 | | 36983 |
| 15 GB | 8 | 46087 | | 49977 |
| 15 GB | 12 | 41901 | | 48591 |
| 15 GB | 16 | 40147 | | 50651 | +26.16%
| 15 GB | 24 | 37250 | | 52365 | +40.58%
| 15 GB | 32 | 36470 | | 50015 | +37.14%

CC: Ingo Molnar <mingo@kernel.org>
CC: Peter Zijlstra <peterz@infradead.org>
CC: Mike Galbraith <efault@gmx.de>
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
---
include/linux/sched.h | 3 +++
kernel/sched/fair.c | 47 +++++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 50 insertions(+), 0 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 178a8d9..1c996c7 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1041,6 +1041,9 @@ struct task_struct {
#ifdef CONFIG_SMP
struct llist_node wake_entry;
int on_cpu;
+ struct task_struct *last_wakee;
+ unsigned long nr_wakee_switch;
+ unsigned long last_switch_decay;
#endif
int on_rq;

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index c61a614..a4ddbf5 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2971,6 +2971,23 @@ static unsigned long cpu_avg_load_per_task(int cpu)
return 0;
}

+static void record_wakee(struct task_struct *p)
+{
+ /*
+ * Rough decay(wiping) for cost saving, don't worry
+ * about the boundary, really active task won't care
+ * the loose.
+ */
+ if (jiffies > current->last_switch_decay + HZ) {
+ current->nr_wakee_switch = 0;
+ current->last_switch_decay = jiffies;
+ }
+
+ if (current->last_wakee != p) {
+ current->last_wakee = p;
+ current->nr_wakee_switch++;
+ }
+}

static void task_waking_fair(struct task_struct *p)
{
@@ -2991,6 +3008,7 @@ static void task_waking_fair(struct task_struct *p)
#endif

se->vruntime -= min_vruntime;
+ record_wakee(p);
}

#ifdef CONFIG_FAIR_GROUP_SCHED
@@ -3109,6 +3127,28 @@ static inline unsigned long effective_load(struct task_group *tg, int cpu,

#endif

+static int wake_wide(struct task_struct *p)
+{
+ int factor = nr_cpus_node(cpu_to_node(smp_processor_id()));
+
+ /*
+ * Yeah, it's the switching-frequency, could means many wakee or
+ * rapidly switch, use factor here will just help to automatically
+ * adjust the loose-degree, so bigger node will lead to more pull.
+ */
+ if (p->nr_wakee_switch > factor) {
+ /*
+ * wakee is somewhat hot, it needs certain amount of cpu
+ * resource, so if waker is far more hot, prefer to leave
+ * it alone.
+ */
+ if (current->nr_wakee_switch > (factor * p->nr_wakee_switch))
+ return 1;
+ }
+
+ return 0;
+}
+
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
{
s64 this_load, load;
@@ -3118,6 +3158,13 @@ static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
unsigned long weight;
int balanced;

+ /*
+ * If we wake multiple tasks be careful to not bounce
+ * ourselves around too much.
+ */
+ if (wake_wide(p))
+ return 0;
+
idx = sd->wake_idx;
this_cpu = smp_processor_id();
prev_cpu = task_cpu(p);
--
1.7.4.1


\
 
 \ /
  Last update: 2013-07-03 08:21    [W:0.631 / U:0.152 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site