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
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
On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <[email protected]> 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
On Mon, 20 Aug 2012 07:51:52 -0700, Xin Tong wrote:
> On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <[email protected]> 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
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
On Mon, Aug 20, 2012 at 5:16 PM, Namhyung Kim <[email protected]> wrote:
> On Mon, 20 Aug 2012 07:51:52 -0700, Xin Tong wrote:
>> On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <[email protected]> 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
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
On Tue, Aug 28, 2012 at 1:15 PM, Xin Tong <[email protected]> wrote:
> 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
>
> On Mon, Aug 20, 2012 at 5:16 PM, Namhyung Kim <[email protected]> wrote:
>> On Mon, 20 Aug 2012 07:51:52 -0700, Xin Tong wrote:
>>> On Sun, Aug 19, 2012 at 10:54 PM, Mike Galbraith <[email protected]> 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