2008-04-03 12:34:59     2007R1 Preemtion model

Document created by Aaronwu Employee on Aug 6, 2013
Version 1Show Document
  • View in full screen mode

2008-04-03 12:34:59     2007R1 Preemtion model

Hans Eklund (SWEDEN)

Message: 53609    While testing svn head last week i noticed the "Preemtion model" option in the kern config menus. What is the preemtion model used in 2007R1 with 2.6.19 kernel? I read this interesting white paper,

 

http://www.linuxdevices.com/articles/AT8906594941.html

 

mostly covering 2.4 kernels, kernel without the real-time patch could sometimes have 50-100ms scheduling latency, is it still so in 2.6 uClinux systems? Adeos/xenomai doc pages says uClinux did not get any realtime patches for 2.6, so is uClinux kernel 2.6 as "bad" as any 2.4 kernel?

 

Would there be a way of lowering the scheduling latency? We are not looking for hard realtime here, just minimizing it to say 5-10 ms in the vast majority of the cases. Working with FIFO threads did not make the situation any better, as my colleague talked about in another thread today. Raising HZ to 1000 did not help either.

 

http://blackfin.uclinux.org/gf/project/uclinux-dist/forum/?action=ForumBrowse&forum_id=39&_forum_action=ForumMessageBrowse&thread_id=26641

 

/Hans

QuoteReplyEditDelete

 

 

2008-04-03 13:17:33     Re: 2007R1 Preemtion model

Mike Frysinger (UNITED STATES)

Message: 53612    it's the same as any other architecture: if preemption is not an option, then you get the default of no preempt

QuoteReplyEditDelete

 

 

2008-04-07 05:04:44     Re: 2007R1 Preemtion model

Hans Eklund (SWEDEN)

Message: 53741    Ok, so 2008R1 could potentially increase soft-real time performance. Has any tests been made to compare the different schedulers impact on the Blackfin uClinux platform when it comes to scheduler latency?

 

It would be a quite big step for our project to move from 2007R1.1-rc3 to 2008R1, but if the new preemtive scheduler would solve this problem it might just be worth it.

 

/hans

QuoteReplyEditDelete

 

 

2008-04-07 14:26:06     Re: 2007R1 Preemtion model

Mike Frysinger (UNITED STATES)

Message: 53756    we have not done any such tests

 

with 2008R1 you get voluntary preemption by default, but the other option (forced preemption) is probably unsafe

Attachments

    Outcomes