lkml.org 
[lkml]   [2007]   [Nov]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: High priority tasks break SMP balancer?
On 19/11/2007, Micah Dowty <micah@vmware.com> wrote:
> [ ... ]
>
> micah@micah-64:~$ cat /proc/schedstat
> version 14
> timestamp 4366722208
> cpu0 0 0 0 785868111 0 828020771 12621812 22703872 14096041 9504937730116 23603703739504 815398959
> domain0 03 8766051 8760123 95 14544377 6374 0 13 8760110 7369 6771 0 20835105 605 0 1 6770 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7026998 0 24826
> cpu1 0 0 0 783505304 0 860700497 26129151 41776419 34749418 10682815341474 1753703225423 834571346
> domain0 03 8685595 8677710 224 7037051 8127 0 76 8677634 7500 7320 7 471762 179 0 8 7312 ===> 0 0 0 0 0 0 0 0 <=== 1 0 1 0 0 0 0 0 0 8607831 0 36753

> micah@micah-64:~$ cat /proc/schedstat
> version 14
> timestamp 4366723231
> cpu0 0 0 0 785868111 0 828021128 12621812 22703965 14096130 9509029340381 23603703795868 815399316
> domain0 03 8766051 8760123 95 14544377 6374 0 13 8760110 7374 6776 0 20835105 605 0 1 6775 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7027010 0 24826
> cpu1 0 0 0 783505304 0 860708521 26132612 41780521 34753508 10683770853336 1753758309667 834575909
> domain0 03 8685718 8677833 224 7037051 8127 0 76 8677757 7500 7320 7 471762 179 0 8 7312 ===> 0 0 0 0 0 0 0 0 <=== 1 0 1 0 0 0 0 0 0 8607835 0 36753

You seem to have a configuration with domains which don't have
SD_BALANCE_NEWIDLE on (CONFIG_NUMA?) as there are no events (all
zeros above) for CPU_NEWLY_IDLE.

this one is being triggered whenever a cpu becomes idle (schedule()
--> idle_balance() --> load_balance_newidle()).

(this flag is a bit #1 == 2)

cat /proc/sys/kernel/sched_domain/cpu0/domain0/flags

?

it can be changed with "echo new_value > ..."

does a change of it make any difference?

moreover, /proc/sys/kernel/sched_domain/cpu1/domain0/newidle_idx seems
to be responsible for a source of the load for calculating the busiest
group. e.g. with newidle_idx == 0, the current load on the queue is
used instead of cpu_load[].


>
> Thanks,
> --Micah
>

--
Best regards,
Dmitry Adamushko
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2007-11-19 23:25    [W:0.051 / U:0.500 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site