Scheduler and SMP

Scheduler and SMP

[Bruno Bauer]

Hello all,

I'm new in kernel development and i be making tests in scheduler for a
study course.
I created a new syscall that set a task for no migrate from cpu (I
added a new flag no_migrate on task_struct).
In my tests, a tar/bzip task with no_migrate flag seted run at 5% more fast.

How do this be explained? Overhead of migration?

My changes are in can_migrate_task and balance_tasks (sched.c) and
copy_process (fork.c).

Thanks,

PS: Sorry, my English is poor.

--
Bruno Germano Bauer

Re: Scheduler and SMP

[Jiri Kosina]

On Sat, 13 Jun 2009, Bruno Bauer wrote:

> I'm new in kernel development and i be making tests in scheduler for a
> study course.

I guess kernelnewbies malinglist would be more appropriate for you than 
lkml.

> I created a new syscall that set a task for no migrate from cpu (I added 
> a new flag no_migrate on task_struct). 

Why didn't you use existing infrastructure (sched_setaffinity()) for that?

> In my tests, a tar/bzip task with no_migrate flag seted run at 5% more 
> fast.
> How do this be explained? Overhead of migration?

Probably. Depends on the exact changes you have made, which you haven't 
shared with us.

-- 
Jiri Kosina
SUSE Labs