* [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
@ 2009-04-30 6:11 KOSAKI Motohiro
2009-04-30 6:18 ` Balbir Singh
2009-05-01 1:10 ` Andrew Morton
0 siblings, 2 replies; 9+ messages in thread
From: KOSAKI Motohiro @ 2009-04-30 6:11 UTC (permalink / raw)
To: LKML, Bharata B Rao, Balaji Rao, Dhaval Giani, KAMEZAWA Hiroyuki,
Peter Zijlstra, Balbir Singh, Ingo Molnar, Martin Schwidefsky
Cc: kosaki.motohiro
Changelog:
since v1
- use percpu_counter_sum() instead percpu_counter_read()
-------------------------------------
Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
for avoiding performance degression.
For archs which define VIRT_CPU_ACCOUNTING, every tick would result
in >1000 units of cputime updates and since this is much much greater
than percpu_batch_counter, we end up taking spinlock on every tick.
This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
cputime in per-cpu cache.
it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
Cc: Balaji Rao <balajirrao@gmail.com>
Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
---
kernel/sched.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
Index: b/kernel/sched.c
===================================================================
--- a/kernel/sched.c 2009-04-30 11:37:47.000000000 +0900
+++ b/kernel/sched.c 2009-04-30 14:17:00.000000000 +0900
@@ -10221,6 +10221,7 @@ struct cpuacct {
};
struct cgroup_subsys cpuacct_subsys;
+static s32 cpuacct_batch;
/* return cpu accounting group corresponding to this container */
static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
@@ -10250,6 +10251,9 @@ static struct cgroup_subsys_state *cpuac
if (!ca->cpuusage)
goto out_free_ca;
+ if (!cpuacct_batch)
+ cpuacct_batch = jiffies_to_cputime(percpu_counter_batch);
+
for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
if (percpu_counter_init(&ca->cpustat[i], 0))
goto out_free_counters;
@@ -10376,7 +10380,7 @@ static int cpuacct_stats_show(struct cgr
int i;
for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
- s64 val = percpu_counter_read(&ca->cpustat[i]);
+ s64 val = percpu_counter_sum(&ca->cpustat[i]);
val = cputime64_to_clock_t(val);
cb->fill(cb, cpuacct_stat_desc[i], val);
}
@@ -10446,7 +10450,7 @@ static void cpuacct_update_stats(struct
ca = task_ca(tsk);
do {
- percpu_counter_add(&ca->cpustat[idx], val);
+ __percpu_counter_add(&ca->cpustat[idx], val, cpuacct_batch);
ca = ca->parent;
} while (ca);
rcu_read_unlock();
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 6:11 [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count KOSAKI Motohiro
@ 2009-04-30 6:18 ` Balbir Singh
2009-04-30 8:28 ` KOSAKI Motohiro
2009-04-30 8:47 ` Peter Zijlstra
2009-05-01 1:10 ` Andrew Morton
1 sibling, 2 replies; 9+ messages in thread
From: Balbir Singh @ 2009-04-30 6:18 UTC (permalink / raw)
To: KOSAKI Motohiro
Cc: LKML, Bharata B Rao, Balaji Rao, Dhaval Giani, KAMEZAWA Hiroyuki,
Peter Zijlstra, Ingo Molnar, Martin Schwidefsky
* KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> [2009-04-30 15:11:15]:
>
> Changelog:
> since v1
> - use percpu_counter_sum() instead percpu_counter_read()
>
>
> -------------------------------------
> Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
>
> cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
> for avoiding performance degression.
>
> For archs which define VIRT_CPU_ACCOUNTING, every tick would result
> in >1000 units of cputime updates and since this is much much greater
> than percpu_batch_counter, we end up taking spinlock on every tick.
>
> This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
> cputime in per-cpu cache.
> it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
>
> Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
> Cc: Balaji Rao <balajirrao@gmail.com>
> Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
> Cc: Ingo Molnar <mingo@elte.hu>
> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> ---
> kernel/sched.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> Index: b/kernel/sched.c
> ===================================================================
> --- a/kernel/sched.c 2009-04-30 11:37:47.000000000 +0900
> +++ b/kernel/sched.c 2009-04-30 14:17:00.000000000 +0900
> @@ -10221,6 +10221,7 @@ struct cpuacct {
> };
>
> struct cgroup_subsys cpuacct_subsys;
> +static s32 cpuacct_batch;
>
> /* return cpu accounting group corresponding to this container */
> static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
> @@ -10250,6 +10251,9 @@ static struct cgroup_subsys_state *cpuac
> if (!ca->cpuusage)
> goto out_free_ca;
>
> + if (!cpuacct_batch)
> + cpuacct_batch = jiffies_to_cputime(percpu_counter_batch);
> +
> for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
> if (percpu_counter_init(&ca->cpustat[i], 0))
> goto out_free_counters;
> @@ -10376,7 +10380,7 @@ static int cpuacct_stats_show(struct cgr
> int i;
>
> for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> - s64 val = percpu_counter_read(&ca->cpustat[i]);
> + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> val = cputime64_to_clock_t(val);
> cb->fill(cb, cpuacct_stat_desc[i], val);
> }
> @@ -10446,7 +10450,7 @@ static void cpuacct_update_stats(struct
> ca = task_ca(tsk);
>
> do {
> - percpu_counter_add(&ca->cpustat[idx], val);
> + __percpu_counter_add(&ca->cpustat[idx], val, cpuacct_batch);
> ca = ca->parent;
> } while (ca);
> rcu_read_unlock();
>
>
What do the test results look like with this? I'll see if I can find
some time to test this patch. On a patch read level this seems much better
to me, Peter?
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
--
Balbir
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 6:18 ` Balbir Singh
@ 2009-04-30 8:28 ` KOSAKI Motohiro
2009-04-30 8:47 ` Peter Zijlstra
1 sibling, 0 replies; 9+ messages in thread
From: KOSAKI Motohiro @ 2009-04-30 8:28 UTC (permalink / raw)
To: balbir
Cc: kosaki.motohiro, LKML, Bharata B Rao, Balaji Rao, Dhaval Giani,
KAMEZAWA Hiroyuki, Peter Zijlstra, Ingo Molnar,
Martin Schwidefsky
> * KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> [2009-04-30 15:11:15]:
>
> >
> > Changelog:
> > since v1
> > - use percpu_counter_sum() instead percpu_counter_read()
> >
> >
> > -------------------------------------
> > Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
> >
> > cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
> > for avoiding performance degression.
> >
> > For archs which define VIRT_CPU_ACCOUNTING, every tick would result
> > in >1000 units of cputime updates and since this is much much greater
> > than percpu_batch_counter, we end up taking spinlock on every tick.
> >
> > This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
> > cputime in per-cpu cache.
> > it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
> >
> > Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
> > Cc: Balaji Rao <balajirrao@gmail.com>
> > Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> > Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
> > Cc: Ingo Molnar <mingo@elte.hu>
> > Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
> > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > ---
> > kernel/sched.c | 8 ++++++--
> > 1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > Index: b/kernel/sched.c
> > ===================================================================
> > --- a/kernel/sched.c 2009-04-30 11:37:47.000000000 +0900
> > +++ b/kernel/sched.c 2009-04-30 14:17:00.000000000 +0900
> > @@ -10221,6 +10221,7 @@ struct cpuacct {
> > };
> >
> > struct cgroup_subsys cpuacct_subsys;
> > +static s32 cpuacct_batch;
> >
> > /* return cpu accounting group corresponding to this container */
> > static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
> > @@ -10250,6 +10251,9 @@ static struct cgroup_subsys_state *cpuac
> > if (!ca->cpuusage)
> > goto out_free_ca;
> >
> > + if (!cpuacct_batch)
> > + cpuacct_batch = jiffies_to_cputime(percpu_counter_batch);
> > +
> > for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
> > if (percpu_counter_init(&ca->cpustat[i], 0))
> > goto out_free_counters;
> > @@ -10376,7 +10380,7 @@ static int cpuacct_stats_show(struct cgr
> > int i;
> >
> > for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> > - s64 val = percpu_counter_read(&ca->cpustat[i]);
> > + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> > val = cputime64_to_clock_t(val);
> > cb->fill(cb, cpuacct_stat_desc[i], val);
> > }
> > @@ -10446,7 +10450,7 @@ static void cpuacct_update_stats(struct
> > ca = task_ca(tsk);
> >
> > do {
> > - percpu_counter_add(&ca->cpustat[idx], val);
> > + __percpu_counter_add(&ca->cpustat[idx], val, cpuacct_batch);
> > ca = ca->parent;
> > } while (ca);
> > rcu_read_unlock();
> >
> >
>
> What do the test results look like with this? I'll see if I can find
> some time to test this patch. On a patch read level this seems much better
> to me, Peter?
>
> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
this patch decrease slow down risk on large server. but this patch
doesn't have functional change. you can't make functional test.
AFAIK, percpu_counter_sum() don't make any performance degression,
but you have good stress test, please tell me it.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 6:18 ` Balbir Singh
2009-04-30 8:28 ` KOSAKI Motohiro
@ 2009-04-30 8:47 ` Peter Zijlstra
2009-04-30 8:52 ` KOSAKI Motohiro
2009-04-30 8:55 ` Ingo Molnar
1 sibling, 2 replies; 9+ messages in thread
From: Peter Zijlstra @ 2009-04-30 8:47 UTC (permalink / raw)
To: balbir
Cc: KOSAKI Motohiro, LKML, Bharata B Rao, Balaji Rao, Dhaval Giani,
KAMEZAWA Hiroyuki, Ingo Molnar, Martin Schwidefsky
On Thu, 2009-04-30 at 11:48 +0530, Balbir Singh wrote:
> > Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
> >
> > cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
> > for avoiding performance degression.
> >
> > For archs which define VIRT_CPU_ACCOUNTING, every tick would result
> > in >1000 units of cputime updates and since this is much much greater
> > than percpu_batch_counter, we end up taking spinlock on every tick.
> >
> > This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
> > cputime in per-cpu cache.
> > it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
> >
> > Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
> > Cc: Balaji Rao <balajirrao@gmail.com>
> > Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> > Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
> > Cc: Ingo Molnar <mingo@elte.hu>
> > Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
> > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > ---
> > kernel/sched.c | 8 ++++++--
> > 1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > Index: b/kernel/sched.c
> > ===================================================================
> > --- a/kernel/sched.c 2009-04-30 11:37:47.000000000 +0900
> > +++ b/kernel/sched.c 2009-04-30 14:17:00.000000000 +0900
> > @@ -10221,6 +10221,7 @@ struct cpuacct {
> > };
> >
> > struct cgroup_subsys cpuacct_subsys;
> > +static s32 cpuacct_batch;
> >
> > /* return cpu accounting group corresponding to this container */
> > static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
> > @@ -10250,6 +10251,9 @@ static struct cgroup_subsys_state *cpuac
> > if (!ca->cpuusage)
> > goto out_free_ca;
> >
> > + if (!cpuacct_batch)
> > + cpuacct_batch = jiffies_to_cputime(percpu_counter_batch);
> > +
> > for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
> > if (percpu_counter_init(&ca->cpustat[i], 0))
> > goto out_free_counters;
> > @@ -10376,7 +10380,7 @@ static int cpuacct_stats_show(struct cgr
> > int i;
> >
> > for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> > - s64 val = percpu_counter_read(&ca->cpustat[i]);
> > + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> > val = cputime64_to_clock_t(val);
> > cb->fill(cb, cpuacct_stat_desc[i], val);
> > }
> > @@ -10446,7 +10450,7 @@ static void cpuacct_update_stats(struct
> > ca = task_ca(tsk);
> >
> > do {
> > - percpu_counter_add(&ca->cpustat[idx], val);
> > + __percpu_counter_add(&ca->cpustat[idx], val, cpuacct_batch);
> > ca = ca->parent;
> > } while (ca);
> > rcu_read_unlock();
> >
> >
>
> What do the test results look like with this? I'll see if I can find
> some time to test this patch. On a patch read level this seems much better
> to me, Peter?
I don't really fancy percpu_counter_sum() usage. I'm thinking its ok to
degrate accuracy on larger machines and simply use
percpu_counter_read().
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 8:47 ` Peter Zijlstra
@ 2009-04-30 8:52 ` KOSAKI Motohiro
2009-04-30 9:02 ` Balbir Singh
2009-04-30 8:55 ` Ingo Molnar
1 sibling, 1 reply; 9+ messages in thread
From: KOSAKI Motohiro @ 2009-04-30 8:52 UTC (permalink / raw)
To: Peter Zijlstra
Cc: kosaki.motohiro, balbir, LKML, Bharata B Rao, Balaji Rao,
Dhaval Giani, KAMEZAWA Hiroyuki, Ingo Molnar, Martin Schwidefsky
> > > for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> > > - s64 val = percpu_counter_read(&ca->cpustat[i]);
> > > + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> > > val = cputime64_to_clock_t(val);
> > > cb->fill(cb, cpuacct_stat_desc[i], val);
> > > }
> >
> > What do the test results look like with this? I'll see if I can find
> > some time to test this patch. On a patch read level this seems much better
> > to me, Peter?
>
> I don't really fancy percpu_counter_sum() usage. I'm thinking its ok to
> degrate accuracy on larger machines and simply use
> percpu_counter_read().
I have same opinion with peter. Balbir, What do you think?
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 8:47 ` Peter Zijlstra
2009-04-30 8:52 ` KOSAKI Motohiro
@ 2009-04-30 8:55 ` Ingo Molnar
1 sibling, 0 replies; 9+ messages in thread
From: Ingo Molnar @ 2009-04-30 8:55 UTC (permalink / raw)
To: Peter Zijlstra
Cc: balbir, KOSAKI Motohiro, LKML, Bharata B Rao, Balaji Rao,
Dhaval Giani, KAMEZAWA Hiroyuki, Martin Schwidefsky
* Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> On Thu, 2009-04-30 at 11:48 +0530, Balbir Singh wrote:
>
> > > Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
> > >
> > > cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
> > > for avoiding performance degression.
> > >
> > > For archs which define VIRT_CPU_ACCOUNTING, every tick would result
> > > in >1000 units of cputime updates and since this is much much greater
> > > than percpu_batch_counter, we end up taking spinlock on every tick.
> > >
> > > This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
> > > cputime in per-cpu cache.
> > > it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
> > >
> > > Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
> > > Cc: Balaji Rao <balajirrao@gmail.com>
> > > Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> > > Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > > Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > > Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
> > > Cc: Ingo Molnar <mingo@elte.hu>
> > > Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
> > > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > > ---
> > > kernel/sched.c | 8 ++++++--
> > > 1 file changed, 6 insertions(+), 2 deletions(-)
> > >
> > > Index: b/kernel/sched.c
> > > ===================================================================
> > > --- a/kernel/sched.c 2009-04-30 11:37:47.000000000 +0900
> > > +++ b/kernel/sched.c 2009-04-30 14:17:00.000000000 +0900
> > > @@ -10221,6 +10221,7 @@ struct cpuacct {
> > > };
> > >
> > > struct cgroup_subsys cpuacct_subsys;
> > > +static s32 cpuacct_batch;
> > >
> > > /* return cpu accounting group corresponding to this container */
> > > static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
> > > @@ -10250,6 +10251,9 @@ static struct cgroup_subsys_state *cpuac
> > > if (!ca->cpuusage)
> > > goto out_free_ca;
> > >
> > > + if (!cpuacct_batch)
> > > + cpuacct_batch = jiffies_to_cputime(percpu_counter_batch);
> > > +
> > > for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
> > > if (percpu_counter_init(&ca->cpustat[i], 0))
> > > goto out_free_counters;
> > > @@ -10376,7 +10380,7 @@ static int cpuacct_stats_show(struct cgr
> > > int i;
> > >
> > > for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> > > - s64 val = percpu_counter_read(&ca->cpustat[i]);
> > > + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> > > val = cputime64_to_clock_t(val);
> > > cb->fill(cb, cpuacct_stat_desc[i], val);
> > > }
> > > @@ -10446,7 +10450,7 @@ static void cpuacct_update_stats(struct
> > > ca = task_ca(tsk);
> > >
> > > do {
> > > - percpu_counter_add(&ca->cpustat[idx], val);
> > > + __percpu_counter_add(&ca->cpustat[idx], val, cpuacct_batch);
> > > ca = ca->parent;
> > > } while (ca);
> > > rcu_read_unlock();
> > >
> > >
> >
> > What do the test results look like with this? I'll see if I can
> > find some time to test this patch. On a patch read level this
> > seems much better to me, Peter?
>
> I don't really fancy percpu_counter_sum() usage. I'm thinking its
> ok to degrate accuracy on larger machines and simply use
> percpu_counter_read().
yes - and the values will converge anyway, right? So it's just a
small delay, not even any genuine loss of accuracy.
Ingo
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 8:52 ` KOSAKI Motohiro
@ 2009-04-30 9:02 ` Balbir Singh
0 siblings, 0 replies; 9+ messages in thread
From: Balbir Singh @ 2009-04-30 9:02 UTC (permalink / raw)
To: KOSAKI Motohiro
Cc: Peter Zijlstra, LKML, Bharata B Rao, Balaji Rao, Dhaval Giani,
KAMEZAWA Hiroyuki, Ingo Molnar, Martin Schwidefsky
* KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> [2009-04-30 17:52:16]:
> > > > for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
> > > > - s64 val = percpu_counter_read(&ca->cpustat[i]);
> > > > + s64 val = percpu_counter_sum(&ca->cpustat[i]);
> > > > val = cputime64_to_clock_t(val);
> > > > cb->fill(cb, cpuacct_stat_desc[i], val);
> > > > }
> > >
> > > What do the test results look like with this? I'll see if I can find
> > > some time to test this patch. On a patch read level this seems much better
> > > to me, Peter?
> >
> > I don't really fancy percpu_counter_sum() usage. I'm thinking its ok to
> > degrate accuracy on larger machines and simply use
> > percpu_counter_read().
>
> I have same opinion with peter. Balbir, What do you think?
>
Sure, but the larger the delta gets, the less useful the metric gets
:) I am OK with going back to percpu_counter_read() if that is the
consensus.
--
Balbir
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-04-30 6:11 [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count KOSAKI Motohiro
2009-04-30 6:18 ` Balbir Singh
@ 2009-05-01 1:10 ` Andrew Morton
2009-05-01 1:45 ` KOSAKI Motohiro
1 sibling, 1 reply; 9+ messages in thread
From: Andrew Morton @ 2009-05-01 1:10 UTC (permalink / raw)
To: KOSAKI Motohiro
Cc: linux-kernel, bharata, balajirrao, dhaval, kamezawa.hiroyu,
a.p.zijlstra, balbir, mingo, schwidefsky, kosaki.motohiro
On Thu, 30 Apr 2009 15:11:15 +0900 (JST)
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
>
> Changelog:
> since v1
> - use percpu_counter_sum() instead percpu_counter_read()
>
>
> -------------------------------------
> Subject: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
>
> cpuacct_update_stats() is called at every tick updating. and it use percpu_counter
> for avoiding performance degression.
>
> For archs which define VIRT_CPU_ACCOUNTING, every tick would result
> in >1000 units of cputime updates and since this is much much greater
> than percpu_batch_counter, we end up taking spinlock on every tick.
>
> This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
> cputime in per-cpu cache.
> it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
Does this actually matter?
If we're calling cpuacct_update_stats() with large values of `cputime'
then presumably we're also calling cpuacct_update_stats() at a low
frequency, so the common lock-taking won't cause performance problems?
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count
2009-05-01 1:10 ` Andrew Morton
@ 2009-05-01 1:45 ` KOSAKI Motohiro
0 siblings, 0 replies; 9+ messages in thread
From: KOSAKI Motohiro @ 2009-05-01 1:45 UTC (permalink / raw)
To: Andrew Morton
Cc: linux-kernel, bharata, balajirrao, dhaval, kamezawa.hiroyu,
a.p.zijlstra, balbir, mingo, schwidefsky
>> For archs which define VIRT_CPU_ACCOUNTING, every tick would result
>> in >1000 units of cputime updates and since this is much much greater
>> than percpu_batch_counter, we end up taking spinlock on every tick.
>>
>> This patch change batch rule. now, any cpu can store "percpu_counter_bach * jiffies"
>> cputime in per-cpu cache.
>> it mean this patch don't have behavior change if VIRT_CPU_ACCOUNTING=n.
>
> Does this actually matter?
>
> If we're calling cpuacct_update_stats() with large values of `cputime'
> then presumably we're also calling cpuacct_update_stats() at a low
> frequency, so the common lock-taking won't cause performance problems?
VIRT_CPU_ACCOUNTING change cputime_t meaning. but don't change calling
update time frequency.
example,
ia64, HZ=1000, VIRT_CPU_ACCOUNTING=y (1 cputime == 1ns, ie 1 jiffies
== 1000000 cputime)
every tick updating makes 1000000 cputime. (see jiffies_to_cputime)
-----------------------------------------------------------------------------------------
void account_process_tick(struct task_struct *p, int user_tick)
{
cputime_t one_jiffy = jiffies_to_cputime(1);
cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy);
struct rq *rq = this_rq();
if (user_tick)
account_user_time(p, one_jiffy, one_jiffy_scaled);
else if (p != rq->idle)
account_system_time(p, HARDIRQ_OFFSET, one_jiffy,
one_jiffy_scaled);
else
account_idle_time(one_jiffy);
}
-----------------------------------------------------------------------------------------
but tick updating frequency don't changed.
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2009-05-01 1:46 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2009-04-30 6:11 [PATCH v2] cpuacct: VIRT_CPU_ACCOUNTING don't prevent percpu cputime count KOSAKI Motohiro
2009-04-30 6:18 ` Balbir Singh
2009-04-30 8:28 ` KOSAKI Motohiro
2009-04-30 8:47 ` Peter Zijlstra
2009-04-30 8:52 ` KOSAKI Motohiro
2009-04-30 9:02 ` Balbir Singh
2009-04-30 8:55 ` Ingo Molnar
2009-05-01 1:10 ` Andrew Morton
2009-05-01 1:45 ` KOSAKI Motohiro
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).