Re: Latency Problems with 2.6.31/PREEMPT_RT on a 1.2GHz Celeron

[Mark Hounschell <dmarkh@cfl.rr.com>]

On 12/21/2009 04:16 AM, Nikita V. Youshchenko wrote:
>> On 12/18/2009 11:56 AM, Vladimir Cotfas wrote:
>>> All,
>>>
>>> I have a a driver that must process an interrupt every 100 uS without
>>> missing any. I am having latency issues and I am looking for some help
>>> debugging my problem.
>>
>> You'll probably never get this working reliably with the RT kernel.
>> Latencies are more deterministic with the RT kernel but slightly higher.
>> You'll have better luck with a vanilla kernel running your app in run
>> level 4. And then even better luck running on an SMP machine isolating
>> your process and IRQ to a single processor. And stay away from the VGA
>> console. You have better luck with faster hardware too.
> 
> We have seen <20us avg latencies on RT kernel on slower powerpc-based SoC.
> 
> The only problem is that sometimes we got single-shot latency >150 us - 
> that happened when sched tick was running for >100 us with interrupts 
> disabled.
> 

Averages are one thing. Guaranteeing he doesn't miss _any_ and still has
time to do whatever he needs to do before the next one, is different. Local
timer interrupts alone are enough to kill ya.

Mark