What's the timeslice size for kernel 2.6.0-test2, IA32?

What's the timeslice size for kernel 2.6.0-test2, IA32?

[Stephen Anthony]

I haven't been able to find this information anywhere.  I know HZ was 
increased to 1000, but was the timeslice decreased to 1 ms (from 10 ms) 
as well?

Steve

Re: What's the timeslice size for kernel 2.6.0-test2, IA32?

[Con Kolivas]

On Sat, 2 Aug 2003 01:51, Stephen Anthony wrote:
> I haven't been able to find this information anywhere.  I know HZ was
> increased to 1000, but was the timeslice decreased to 1 ms (from 10 ms)
> as well?

Depends on nice of the task. Nice 0 tasks get 102ms.

Con

Re: Re: What's the timeslice size for kernel 2.6.0-test2, IA32?

[Stephen Anthony]

>> I haven't been able to find this information anywhere.  I know HZ was
>> increased to 1000, but was the timeslice decreased to 1 ms (from 10 ms)
>> as well?
>
> Depends on nice of the task. Nice 0 tasks get 102ms.

I don't think I asked the right question :)  If I call usleep(x) or 
nanosleep(x) with kernel 2.4.21, and x < 10, the sleep would still last 
10 ms because of the timeslice.  All sleeps would be a multiple of 10 ms.

If I call usleep(x) or nanosleep(x) in 2.6.0-test2, what 'multiple' can I 
expect?  Maybe I mean granularity instead of timeslice.  Basically, I 
want to know how 'soft' of a real-time system the new kernel is.

It would be great if sleeps were 1ms accurate instead of 10ms.  It would 
make synchronization code a lot easier.

Steve

Re: Re: What's the timeslice size for kernel 2.6.0-test2, IA32?

[Mark Mielke]

On Fri, Aug 01, 2003 at 02:04:46PM -0230, Stephen Anthony wrote:
> It would be great if sleeps were 1ms accurate instead of 10ms.  It would 
> make synchronization code a lot easier.

Doesn't this depend on what HZ you define for the kernel?

If you want 1ms sleep, just set HZ to 1000HZ+, and give your process a
high priority?

mark

-- 
mark@mielke.cc/markm@ncf.ca/markm@nortelnetworks.com __________________________
.  .  _  ._  . .   .__    .  . ._. .__ .   . . .__  | Neighbourhood Coder
|\/| |_| |_| |/    |_     |\/|  |  |_  |   |/  |_   | 
|  | | | | \ | \   |__ .  |  | .|. |__ |__ | \ |__  | Ottawa, Ontario, Canada

  One ring to rule them all, one ring to find them, one ring to bring them all
                       and in the darkness bind them...

                           http://mark.mielke.cc/

Re: What's the timeslice size for kernel 2.6.0-test2, IA32?

[Ben Greear]

Mark Mielke wrote:
> On Fri, Aug 01, 2003 at 02:04:46PM -0230, Stephen Anthony wrote:
> 
>>It would be great if sleeps were 1ms accurate instead of 10ms.  It would 
>>make synchronization code a lot easier.
> 
> 
> Doesn't this depend on what HZ you define for the kernel?
> 
> If you want 1ms sleep, just set HZ to 1000HZ+, and give your process a
> high priority?
> 
> mark
> 

 From user space, at least, you can configure /dev/rtc to 1024HZ and then
select on it's file descriptor.  That will give you fairly reliable 1ms (or so)
sleeps, especially if you set your niceness to a high-priority setting.

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: What's the timeslice size for kernel 2.6.0-test2, IA32?

[george anzinger]

Ben Greear wrote:
> 
> Mark Mielke wrote:
> 
>> On Fri, Aug 01, 2003 at 02:04:46PM -0230, Stephen Anthony wrote:
>>
>>> It would be great if sleeps were 1ms accurate instead of 10ms.  It 
>>> would make synchronization code a lot easier.
>>
>>
>>
>> Doesn't this depend on what HZ you define for the kernel?
>>
>> If you want 1ms sleep, just set HZ to 1000HZ+, and give your process a
>> high priority?

As it currently stands in the 2.6 kernel for the i386, HZ is defined 
as 1000.  Since the PIT interrupt source can not hit this, the actual 
timer interrupt period is 999848 nano seconds.  The minimum sleep 
interval nanosleep will take is 1 of these and since the time may 
start between ticks it adds another to give a min sleep time of 
999848+~1/2 of that depending on where the request falls in the time 
period.

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml