Patch "1b9508f6 sched: Rate-limit newidle" reduced the CPU time spent in
idle_balance() by refusing to balance if the average idle time was less
than sysctl_sched_migration_cost. Since then, more refined methods for
reducing CPU time have been added, including dynamic measurement of search
cost in curr_cost and a check for this_rq->rd->overload. The original
check of sysctl_sched_migration_cost is no longer necessary, and is in
fact harmful because it discourages load balancing, so delete it.
1) An internal Oracle RDBMS OLTP test on an 8-socket Exadata shows a 2.2%
gain in throughput.
2) Hackbench results on 2 socket, 44 core and 88 threads Intel x86 machine
(lower is better):
+--------------+-----------------+-------------------------+
| | Without Patch |With Patch |
+------+-------+--------+--------+----------------+--------+
|Loops | Groups|Average |%Std Dev|Average |%Std Dev|
+------+-------+--------+--------+----------------+--------+
|100000| 4 |9.701 |0.78 |8.919 (+8.07%) |1.07 |
|100000| 8 |17.186 |0.77 |17.043 (+0.83%) |0.83 |
|100000| 16 |30.378 |0.55 |29.565 (+2.67%) |0.29 |
|100000| 32 |54.712 |0.54 |52.158 (+4.67%) |0.22 |
+------+-------+--------+--------+----------------+--------+
3) Sysbench MySQL results on 2 socket, 44 core and 88 threads Intel x86
machine (higher is better):
+-------+--------------------+----------------------------+
| | Without Patch | With Patch |
+-------+-----------+--------+-------------------+--------+
|Num | Average | | Average | |
|Threads| throughput|%Std Dev| throughput |%Std Dev|
+-------+-----------+--------+-------------------+--------+
| 8 | 133658.2 | 0.66 | 134232.2 (+0.43%) | 1.29 |
| 16 | 266540 | 0.48 | 268584.6 (+0.77%) | 0.37 |
| 32 | 466315.6 | 0.15 | 468594.2 (+0.49%) | 0.23 |
| 64 | 720039.4 | 0.23 | 717253.8 (-0.39%) | 0.36 |
| 72 | 757284.4 | 0.25 | 764984.0 (+1.02%) | 0.38 |
| 80 | 807955.6 | 0.22 | 831372.2 (+2.90%) | 0.10 |
| 88 | 863173.8 | 0.25 | 887049.0 (+2.77%) | 0.56 |
| 96 | 882950.8 | 0.32 | 892913.8 (+1.13%) | 0.41 |
| 128 | 895112.6 | 0.13 | 901195.0 (+0.68%) | 0.28 |
+-------+-----------+--------+-------------------+--------+
4) tbench sample results on 2 socket, 44 core and 88 threads Intel x86
machine:
With Patch:
Throughput 562.783 MB/sec 2 clients 2 procs max_latency=0.365 ms
Throughput 1394.5 MB/sec 5 clients 5 procs max_latency=0.531 ms
Throughput 2741.27 MB/sec 10 clients 10 procs max_latency=0.692 ms
Throughput 5279.49 MB/sec 20 clients 20 procs max_latency=1.029 ms
Throughput 8529.22 MB/sec 40 clients 40 procs max_latency=1.693 ms
Without patch:
Throughput 557.142 MB/sec 2 clients 2 procs max_latency=0.264 ms
Throughput 1381.59 MB/sec 5 clients 5 procs max_latency=0.335 ms
Throughput 2726.84 MB/sec 10 clients 10 procs max_latency=0.352 ms
Throughput 5230.12 MB/sec 20 clients 20 procs max_latency=1.632 ms
Throughput 8474.5 MB/sec 40 clients 40 procs max_latency=7.756 ms
Signed-off-by: Rohit Jain <rohit.k.j...@oracle.com>
---
kernel/sched/fair.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 2fe3aa8..52cf36e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8782,8 +8782,7 @@ static int idle_balance(struct rq *this_rq, struct
rq_flags *rf)
*/
rq_unpin_lock(this_rq, rf);
- if (this_rq->avg_idle < sysctl_sched_migration_cost ||
- !this_rq->rd->overload) {
+ if (!this_rq->rd->overload) {
rcu_read_lock();
sd = rcu_dereference_check_sched_domain(this_rq->sd);
if (sd)
--
2.7.4
