login
Search
On a single cpu system with SMT, in the scenario of one SMT thread being idle with another SMT thread running a task and doing a non sync wakeup of another task, we see (from the traces) that the woken up task ends up running on the busy thread instead of the idle thread. Idle balancing that comes in little bit later is fixing the scernaio. But fixing this wake balance and running the woken up task directly on the idle SMT thread improved the performance (phoronix 7zip compression workload) by ~9% on an atom platform. During the process wakeup, select_task_rq_fair() and wake_affine() makes the decision to wakeup the task either on the previous cpu that the task ran or the cpu that the task is currently woken up. select_task_rq_fair() also goes through to see if there are any idle siblings for the cpu that the task is woken up on. This is to ensure that we select any idle sibling rather than choose a busy cpu. In the above load scenario, it so happens that the prev_cpu (that the task ran before) and this_cpu (where it is woken up currently) are the same. And in this case, it looks like wake_affine() returns 0 and ultimately not selecting the idle sibling chosen by select_idle_sibling() in select_task_rq_fair(). Further down the path of select_task_rq_fair(), we ultimately select the currently running cpu (busy SMT thread instead of the idle SMT thread). Check for prev_cpu == this_cpu before calling wake_affine() and no need to do any fancy stuff(and ultimately taking wrong decisions) in this case. Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> --- Changes from v1: Move the "this_cpu == prev_cpu" check before calling wake_affine() --- kernel/sched_fair.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) Index: tip/kernel/sched_fair.c =================================================================== --- tip.orig/kernel/sched_fair.c +++ tip/kernel/sched_fair.c @@ -1454,6 +1454,7 @@ static int select_task_rq_fair(struct ta int ...
Address these issues in the current select_idle_sibling() logic.
a) Once we select the idle sibling, we use that domain (spanning the cpu that
the task is currently woken-up and the idle sibling that we found) in our
wake_affine() comparisons. This domain is completely different from the
domain(we are supposed to use) that spans the cpu that the task currently
woken-up and the cpu where the task previously ran.
b) We do select_idle_sibling() check only for the cpu that the task is
currently woken-up on. If select_task_rq_fair() selects the previously run
cpu for waking the task no, doing a select_idle_sibling() check
for that cpu also helps and we don't do this currently.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
Changes from v1:
Drop treating the current cpu with one running task as an idle cpu in the
presence of sync wakeup.
---
kernel/sched_fair.c | 69 +++++++++++++++++++++++++++-------------------------
1 file changed, 36 insertions(+), 33 deletions(-)
Index: tip/kernel/sched_fair.c
===================================================================
--- tip.orig/kernel/sched_fair.c
+++ tip/kernel/sched_fair.c
@@ -1407,28 +1407,48 @@ find_idlest_cpu(struct sched_group *grou
* Try and locate an idle CPU in the sched_domain.
*/
static int
-select_idle_sibling(struct task_struct *p, struct sched_domain *sd, int target)
+select_idle_sibling(struct task_struct *p, int target)
{
int cpu = smp_processor_id();
int prev_cpu = task_cpu(p);
int i;
+ struct sched_domain *sd;
/*
- * If this domain spans both cpu and prev_cpu (see the SD_WAKE_AFFINE
- * test in select_task_rq_fair) and the prev_cpu is idle then that's
- * always a better target than the current cpu.
+ * If the task is going to be woken-up on this cpu and if it is
+ * already idle, then it is the right target.
*/
- if (target == cpu && !cpu_rq(prev_cpu)->cfs.nr_running)
+ if (target == cpu && !cpu_rq(cpu)->cfs.nr_running)
+ return ...Right, so we since merged 8b911acd, in which Mike did almost this but not quite, the question is over: cpu == prev_cpu vs this_cpu == prev_cpu. Mike seems to see some workloads regress with the this_cpu check, does your workload work with the cpu == prev_cpu one? --
Mike saw a regression with the sync check that was in the previous version (v1). Anyways, the current code in -tip has the check that I wanted and which addresses the netbook (2 SMT cpu's) performance issue. But the current logic in select_task_rq_fair() is not quite correct, especially we can wake the task on a busy core rather than on an idle core, as the latest changes are making the wake up decisions entirely on an idle HT sibling if there is one. Also there are couple of more issues which I have explained in the previous version of the patch. I have updated my patch on top of the latest -tip, which addresses all these issues. Let me know your thoughts. Thanks. --- From: Suresh Siddha <suresh.b.siddha@intel.com> Subject: sched: fix select_idle_sibling() logic in select_task_rq_fair() Issues in the current select_idle_sibling() logic in select_task_rq_fair() in the context of a task wake-up: a) Once we select the idle sibling, we use that domain (spanning the cpu that the task is currently woken-up and the idle sibling that we found) in our wake_affine() decisions. This domain is completely different from the domain(we are supposed to use) that spans the cpu that the task currently woken-up and the cpu where the task previously ran. b) We do select_idle_sibling() check only for the cpu that the task is currently woken-up on. If select_task_rq_fair() selects the previously run cpu for waking the task, doing a select_idle_sibling() check for that cpu also helps and we don't do this currently. c) In the scenarios where the cpu that the task is woken-up is busy but with its HT siblings are idle, we are selecting the task be woken-up on the idle HT sibling instead of a core that it previously ran and currently completely idle. i.e., we are not taking decisions based on wake_affine() but directly selecting an idle sibling that can cause an imbalance at the SMT/MC level which will be later corrected by the periodic load ...
Why does that matter? If we find an idle shared cache cpu before we hit the spanning domain, we don't use affine_sd other than maybe (unlikely) Yes, the pressing decision for this one wakeup is can we wake to a shared cache and thus avoid cache misses. IMHO, the point of the affinity decision isn't instant perfect balance, it's cache affinity if at all possible without wrecking balance. Load balancing moves tasks for optimal CPU utilization, tasks waking each other pull to a shared domain.. a tug-of-war that balances buddies over time. wake_affine()'s job is only to say "no, leave it where it was for now". I don't see any reason to ask wake_affine()'s opinion about an idle CPU. We paid for idle shared cache knowledge. We certainly wouldn't want to leave the wakee on it's previous CPU only because that CPU is idle, it would have to be idle and sharing cache. That said, Nehalem may ramp better with select_idle_sibling() turned off at the HT level, and ramp was it's motivation. Maybe you could continue checking until out of shared cache country, but that's more expensive. The logic may not be perfect, but it really needs to become cheaper, not more expensive. -Mike --
Ok. This is not a big issue with the new idle cpu change, as atleast we don't endup calling wake_affine() with the wrong sd. I have never tried to understand any code surrounded by CONFIG_FAIR_GROUP_SCHED so can't comment if the using affine_sd for updating group scheduler shares is correct or not. But please look below for the issues with selecting the Just like the current code, my patch is doing the idle checks only once. Current code is doing idle checks for the woken-up cpu and my code is first selecting woken-up vs previously-ran and then doing idle sibling Last level cache sharing is much more important than small L1 and mid level caches. Also performance impact of keeping both the threads on a core busy in the context of an idle core and then periodic balancer For not wrecking balance we should do the wake_balance() and based on that decision, do the select_idle_sibling() for selecting an idle cpu in Consider this scenario. Today we do balance on fork() and exec(). This will cause the tasks to start far away. On systems like NHM-EP, tasks will start on two different sockets/nodes(as each socket is a numa node) and allocate their memory locally etc. Task A starting on Node-0 and Task B starting on Node-1. Once task B sleeps and if Task A or something else wakes up task B on Node-0, (with the recent change) just because there is an idle HT sibling on node-0 we endup waking the task on node-0. This is wrong. We should first atleast go through wake_affine() and if wake_affine() says ok to move the task to node-0, then we can look at the cache siblings for node-0 and select an appropriate cpu. thanks, suresh --
Yes, if task A and task B are more or less unrelated, you'd want them to stay in separate domains, you'd not want some random event to pull. The other side of the coin is tasks which fork off partners that they will talk to at high frequency. They land just as far away, and desperately need to move into a shared cache domain. There's currently no discriminator, so while always asking wake_affine() may reduce the risk of moving a task with a large footprint, it also increases the risk of leaving buddies jabbering cross cache. You can tweak it in either direction, and neither can be called "wrong", it's all compromise. Do you have a compute load bouncing painfully which this patch cures? I have no strong objections, and the result is certainly easier on the eye. If I were making the decision, I'd want to see some numbers. -Mike --
Mike, Apart from this small tweak that you added in wake_up() path there
is no extra logic that keeps buddies together for long. As I was saying,
fork/exec balance starts apart and in the partial loaded case (i.e.,
when # of running tasks <= # of sockets or # of total cores) the default
load balancer policy also tries to distribute the load equally among
sockets/cores (for peak cache/memory controller bw etc). While the
wakeup() may keep the buddies on SMT siblings, next load balancing event
will move them far away. If we need to keep buddies together we need
All I saw in the changelog when you added this new tweak was:
commit 8b911acdf08477c059d1c36c21113ab1696c612b
Author: Mike Galbraith <efault@gmx.de>
Date: Thu Mar 11 17:17:16 2010 +0100
sched: Fix select_idle_sibling()
Don't bother with selection when the current cpu is idle. ....
Is it me or you who need to provide the data for justification for your
new tweak that changes the current behavior ;)
I will run some workloads aswell!
thanks,
suresh
--
We very definitely need to keep buddies cache affine. Yes, maybe some additional logic is needed, as there is a conflict between load types. A kbuild spreading to the four winds is fine, while netperf jabbering Whoa. It was a simple question, no need to get defensive. You need not provide anything. Forget I even asked, it's not my decision. -Mike --
Mike, PeterZ,
Finally got sometime to get back to this and provide some data backing
up my patch under discussion. Here are my test results:
System is a two socket quad-core NHM-EP with SMT enabled and the
workload is specjbb2005.
Warehouses Throughput
tip tip+proposed-fix
1 35142 35027
2 73563 75977
3 105806 109836
4 133421 142490
5 152151 168888
6 164936 195392
7 184763 208155
8 192419 223846
PeterZ, I think the above clearly shows that we have a problem with the
current -tip code. Please consider the proposed patch (which can be
found at http://marc.info/?l=linux-kernel&m=127007936408754&w=2)
Mike, with the above data, can I have your Ack for the patch?
thanks,
suresh
--
Sure (not that you need it). A bit of abbreviated testing this morning
showed no big hairy differences. A bit of loss to very fast switchers,
but OTOH it improved ramp a bit for mysql/pgsql+oltp.
tip = v2.6.34-rc4-937-gba0b2c9
tip-x = tip + your patches
netperf TCP_RR
unpinned
tip 102877.39 102860.66 103210.25 avg 102982.76 1.000
tip-x 100926.59 100380.26 100536.35 avg 100614.40 .977
pinned
tip 100181.70 100288.34 99711.16 avg 100060.40 1.000
tip-x 99347.12 100551.80 99827.22 avg 99908.71 .998
tbench 8
tip 1195.51 1194.49 1197.46 avg 1195.82 1.000
tip-x 1183.19 1188.00 1188.13 avg 1186.44 .992
mysql+oltp
clients 1 2 4 8 16 32 64 128 256
tip 10424.24 20417.42 37151.84 36777.79 36442.06 36122.14 33803.15 30109.81 28368.75
11196.80 20443.43 37560.39 37358.14 36822.26 36132.95 34411.88 30349.47 28909.04
11204.02 20513.21 37550.11 37219.52 36840.44 36150.47 34445.02 30975.95 28840.59
tip avg 10941.68 20458.02 37420.78 37118.48 36701.58 36135.18 34220.01 30478.41 28706.12
tip-x 10331.62 20981.89 36768.45 36556.19 36069.56 35612.24 34100.70 30459.80 29043.76
11101.77 21153.77 37622.83 37228.68 36845.37 36256.04 34606.37 31287.36 28635.98
11076.66 21153.41 37703.80 37242.89 36842.93 36296.27 34733.06 30578.00 29313.10
tip-x avg 10836.68 21096.35 37365.02 37009.25 36585.95 36054.85 34480.04 30775.05 28997.61
vs tip .990 1.031 .998 .997 .996 .997 1.007 1.009 1.010
pgsql+oltp
clients 1 2 4 8 16 32 64 128 256
tip 14683.30 30452.18 53826.68 53793.77 ...So here we keep using !cfs.nr_running to mean idle, which might not at all be true when there's real-time tasks around. --
something like the below..
---
Index: linux-2.6/kernel/sched_fair.c
===================================================================
--- linux-2.6.orig/kernel/sched_fair.c
+++ linux-2.6/kernel/sched_fair.c
@@ -1375,26 +1375,25 @@ find_idlest_cpu(struct sched_group *grou
/*
* Try and locate an idle CPU in the sched_domain.
*/
-static int
-select_idle_sibling(struct task_struct *p, int target)
+static int select_idle_sibling(struct task_struct *p, int target)
{
int cpu = smp_processor_id();
int prev_cpu = task_cpu(p);
- int i;
struct sched_domain *sd;
+ int i;
/*
* If the task is going to be woken-up on this cpu and if it is
* already idle, then it is the right target.
*/
- if (target == cpu && !cpu_rq(cpu)->cfs.nr_running)
+ if (target == cpu && idle_cpu(cpu))
return cpu;
/*
* If the task is going to be woken-up on the cpu where it previously
* ran and if it is currently idle, then it the right target.
*/
- if (target == prev_cpu && !cpu_rq(prev_cpu)->cfs.nr_running)
+ if (target == prev_cpu && idle_cpu(prev_cpu))
return prev_cpu;
/*
@@ -1405,7 +1404,7 @@ select_idle_sibling(struct task_struct *
break;
for_each_cpu_and(i, sched_domain_span(sd), &p->cpus_allowed) {
- if (!cpu_rq(i)->cfs.nr_running) {
+ if (idle_cpu(i)) {
target = i;
break;
}
@@ -1479,16 +1478,14 @@ select_task_rq_fair(struct rq *rq, struc
want_sd = 0;
}
- if (want_affine) {
- /*
- * If both cpu and prev_cpu are part of this domain,
- * cpu is a valid SD_WAKE_AFFINE target.
- */
- if (cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))
- && (tmp->flags & SD_WAKE_AFFINE)) {
- affine_sd = tmp;
- want_affine = 0;
- }
+ /*
+ * If both cpu and prev_cpu are part of this domain,
+ * cpu is a valid SD_WAKE_AFFINE target.
+ */
+ if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
+ cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
+ affine_sd = ...Commit-ID: 99bd5e2f245d8cd17d040c82d40becdb3efd9b69 Gitweb: http://git.kernel.org/tip/99bd5e2f245d8cd17d040c82d40becdb3efd9b69 Author: Suresh Siddha <suresh.b.siddha@intel.com> AuthorDate: Wed, 31 Mar 2010 16:47:45 -0700 Committer: Ingo Molnar <mingo@elte.hu> CommitDate: Fri, 23 Apr 2010 11:02:02 +0200 sched: Fix select_idle_sibling() logic in select_task_rq_fair() Issues in the current select_idle_sibling() logic in select_task_rq_fair() in the context of a task wake-up: a) Once we select the idle sibling, we use that domain (spanning the cpu that the task is currently woken-up and the idle sibling that we found) in our wake_affine() decisions. This domain is completely different from the domain(we are supposed to use) that spans the cpu that the task currently woken-up and the cpu where the task previously ran. b) We do select_idle_sibling() check only for the cpu that the task is currently woken-up on. If select_task_rq_fair() selects the previously run cpu for waking the task, doing a select_idle_sibling() check for that cpu also helps and we don't do this currently. c) In the scenarios where the cpu that the task is woken-up is busy but with its HT siblings are idle, we are selecting the task be woken-up on the idle HT sibling instead of a core that it previously ran and currently completely idle. i.e., we are not taking decisions based on wake_affine() but directly selecting an idle sibling that can cause an imbalance at the SMT/MC level which will be later corrected by the periodic load balancer. Fix this by first going through the load imbalance calculations using wake_affine() and once we make a decision of woken-up cpu vs previously-ran cpu, then choose a possible idle sibling for waking up the task on. Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1270079265.7835.8.camel@sbs-t61.sc.intel.com> Signed-off-by: Ingo Molnar ...
