changing timeslice in linux

Xin Tong

unread,

Aug 19, 2012, 3:00:01 PM8/19/12

to

I have 2 questions about linux 2.6 x86_64 scheduler.

1. is the default scheduling algorithm SCHED_NORMAL in linux ?
2. how do i change the time slice in linux source code ?

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

Mike Galbraith

unread,

Aug 20, 2012, 2:00:02 AM8/20/12

to

On Sun, 2012-08-19 at 11:58 -0700, Xin Tong wrote:
> I have 2 questions about linux 2.6 x86_64 scheduler.
>
> 1. is the default scheduling algorithm SCHED_NORMAL in linux ?

Yes.

> 2. how do i change the time slice in linux source code ?

You shouldn't need to. You can tune "slice" by adjusting
sched_latency_ns and sched_min_granularity_ns, but note that "slice" is
not a fixed quantum. Also note that CFS preemption decisions are based
upon instantaneous state. A task may have received a full (variable)
"slice" of CPU time, but preemption will be triggered only if a more
deserving task is available, so a "slice" is not the "max uninterrupted
CPU time" that you may expect it to be.. but it is somewhat similar.

-Mike

Xin Tong

unread,

Aug 20, 2012, 11:00:03 AM8/20/12

to

On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <efa...@gmx.de> wrote:
> On Sun, 2012-08-19 at 11:58 -0700, Xin Tong wrote:
>> I have 2 questions about linux 2.6 x86_64 scheduler.
>>
>> 1. is the default scheduling algorithm SCHED_NORMAL in linux ?
>
> Yes.

Is there any document describing what the sched_normal is ? it is more
difficult than SCHED_RR to infer its meaning just based on its name ?

>
>> 2. how do i change the time slice in linux source code ?
>
> You shouldn't need to. You can tune "slice" by adjusting
> sched_latency_ns and sched_min_granularity_ns, but note that "slice" is
> not a fixed quantum. Also note that CFS preemption decisions are based
> upon instantaneous state. A task may have received a full (variable)
> "slice" of CPU time, but preemption will be triggered only if a more
> deserving task is available, so a "slice" is not the "max uninterrupted
> CPU time" that you may expect it to be.. but it is somewhat similar.
>
> -Mike
>

This can be done without recompiling and reinstalling the kernel ?
maybe one simply needs to write to a device file under /dev ?

Xin

Namhyung Kim

unread,

Aug 20, 2012, 8:30:01 PM8/20/12

to

On Mon, 20 Aug 2012 07:51:52 -0700, Xin Tong wrote:
> On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <efa...@gmx.de> wrote:
>> On Sun, 2012-08-19 at 11:58 -0700, Xin Tong wrote:
>>> I have 2 questions about linux 2.6 x86_64 scheduler.
>>>
>>> 1. is the default scheduling algorithm SCHED_NORMAL in linux ?
>>
>> Yes.
> Is there any document describing what the sched_normal is ? it is more
> difficult than SCHED_RR to infer its meaning just based on its name ?
>>
>>> 2. how do i change the time slice in linux source code ?
>>
>> You shouldn't need to. You can tune "slice" by adjusting
>> sched_latency_ns and sched_min_granularity_ns, but note that "slice" is
>> not a fixed quantum. Also note that CFS preemption decisions are based
>> upon instantaneous state. A task may have received a full (variable)
>> "slice" of CPU time, but preemption will be triggered only if a more
>> deserving task is available, so a "slice" is not the "max uninterrupted
>> CPU time" that you may expect it to be.. but it is somewhat similar.
>>
>> -Mike
>>
> This can be done without recompiling and reinstalling the kernel ?
> maybe one simply needs to write to a device file under /dev ?

You can find those files under /proc/sys/kernel/.

Thanks,
Namhyung

Xin Tong

unread,

Aug 28, 2012, 4:20:01 PM8/28/12

to

i changed the TIMESLICE defined in the sched.c

109 /*
110 * These are the 'tuning knobs' of the scheduler:
111 *
112 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
113 * Timeslices get refilled after they expire.
114 */
115 #define DEF_TIMESLICE (20 * HZ / 1000)

I compiled and installed the modified kernel, but when i looked at the
, i get a different number

root@elm:~# cat /proc/sys/kernel/sched_rt_period_us
1000000

Xin

Xin Tong

unread,

Aug 28, 2012, 4:20:02 PM8/28/12

to

another question is that when i try to launch many (24) RT processes.
some of them do not seem to get scheduled. i query their scheduling
properties. they all have SCHED_RR with priority 99, but some of them
simply are not taking any CPU.

Xin