lkml.org 
[lkml]   [2010]   [Oct]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: [PATCH 2/3] sched: drop group_capacity to 1 only if remote group has no running tasks
Hi Suresh,

Sorry for the delayed reply.

On Tue, Sep 28, 2010 at 4:04 PM, Suresh Siddha
<suresh.b.siddha@intel.com> wrote:
> On Mon, 2010-09-27 at 17:29 -0700, Nikhil Rao wrote:
>> When SD_PREFER_SIBLING is set on a sched domain, drop group_capacity to 1
>> only if the remote sched group has no running tasks. This addresses the case
>> where you have two tasks on one socket and the other socket is idle, in which
>> case you drop the capacity to 1. If the remote group has >=1 running task, then
>> there is no difference from a cache-sharing perspective.
>>
>> Signed-off-by: Nikhil Rao <ncrao@google.com>
>> ---
>>  kernel/sched_fair.c |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
>> index de8a6a0..33a7985 100644
>> --- a/kernel/sched_fair.c
>> +++ b/kernel/sched_fair.c
>> @@ -2548,7 +2548,7 @@ static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
>>                * first, lower the sg capacity to one so that we'll try
>>                * and move all the excess tasks away.
>>                */
>> -             if (prefer_sibling)
>> +             if (prefer_sibling && !sgs.sum_nr_running)
>>                       sgs.group_capacity = min(sgs.group_capacity, 1UL);
>>
>>               if (local_group) {
>
> Nikhil, Doesn't this break the case of:
>
> two sockets with dual-core and HT. Four tasks currently scheduled as:
> three on socket-0 (two threads on core-0 running two tasks and 1 thread
> on core-1 running one task). One on socket-1 (one thread on core-0
> running a task, with other core-1 idle)
>
> We would like to move the task from core-0 socket-0 to core-1 socket-1,
> while we are load balancing at the socket level (it might be smp or numa
> level depending on system).
>
> thanks,
> suresh
>

Thanks for raising this issue. Yes, when you have a quad-core,
dual-socket machine, the additional check will prevent group_capacity
from dropping down to 1. In this situation, we want to decrease
group_capacity if the local group has extra capacity (i.e.
this_nr_running < this_group_weight) [credit goes to Venki for this
insight]. This also works when you have a niced task, which is what
this patch was trying to fix. I have attached a modified version of
the patch below. Does this look OK?

-Thanks,
Nikhil

---
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index de8a6a0..e0f697a 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -2030,6 +2030,7 @@ struct sd_lb_stats {
unsigned long this_load;
unsigned long this_load_per_task;
unsigned long this_nr_running;
+ unsigned long this_group_capacity;

/* Statistics of the busiest group */
unsigned long max_load;
@@ -2548,7 +2549,8 @@ static inline void update_sd_lb_stats(struct
sched_domain *sd, int this_cpu,
* first, lower the sg capacity to one so that we'll try
* and move all the excess tasks away.
*/
- if (prefer_sibling)
+ if (prefer_sibling && !local_group &&
+ sds->this_nr_running < sds->this_group_capacity)
sgs.group_capacity = min(sgs.group_capacity, 1UL);

if (local_group) {
@@ -2556,6 +2558,7 @@ static inline void update_sd_lb_stats(struct
sched_domain *sd, int this_cpu,
sds->this = sg;
sds->this_nr_running = sgs.sum_nr_running;
sds->this_load_per_task = sgs.sum_weighted_load;
+ sds->this_group_capacity = sgs.group_capacity;
} else if (update_sd_pick_busiest(sd, sds, sg, &sgs,
this_cpu)) {
sds->max_load = sgs.avg_load;
sds->busiest = sg;
--
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: 2010-10-11 23:23    [W:0.172 / U:0.344 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site