[okl4-developer] How to garanty CPU isolation between cells ?

Hugues Balp hugues.balp at free.fr
Thu Dec 11 09:00:17 EST 2008


Gernot Heiser wrote:
> HB> I have performed the same test with a priority of 255 for the ok linux cell
> HB> and a priority of 100 for the decrypt cell and the system doesnt' boot
> HB> better...
>
> You can't expect any system to work if you run apps (which is what OK
> Linux is in this case) at the highest priority (higher than any
> interrupt handlers, for example). I wouldn't expect anything to work
> in such a setup.
>
>   
Yes, I do agree and am sorry for this error !
I will have a look at the RBED scheduler....

After having changed the priorities I got the test working well for a 
versatile platform:
at first with a priority of 120 for linux and 80 for the decrypt cell 
(120, 80), and then trying to reduce
the priority levels between both, with a priority of 99 for linux and 98 
for the decrypt cell (99, 98).

What is quite strange for me is to check that the test doesn't work with 
the following priority values:
(101,100), (100, 99), (98, 97), (130,100) or (98, 80).... ? whereas it 
works for (120,80), (99,80), (99,98).
So I really need to better understand what is going on with the default 
scheduler...

When I test the couple (99,99) the boot begins well and then stops, as 
we could expect because
of the default timeslice values that should be the same between both 
cells...

I have not tested with different values of timeslices simply because I 
don't know how to do that...
Indeed I don't see an accessor for setting the time slice values in the 
OKL4 library reference Manual...
and don't know where to modify the default time slice value for the ok 
linux main thread...

Hugues.
> Assignment of RT priorities is tricky, and your really need to
> understand what you're doing (and there isn't really much helpful
> theory). Which is a reason why we're trying to move away from
> priority-based scheduling (see for example the work done at NICTA and
> UCSC on using the RBED scheduler in L4). Unfortunately, most real-life
> RT systems are based on priorities, so that's likely to remain the
> default setup for a while...
>
> Gernot
>
>
>   




More information about the Developer mailing list