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

[Hugues Balp]

Robert Wetzel wrote:
> hugues.balp at free.fr wrote:
>   
>>                   Hi all,
>>
>>  I wonder if the okl4 micro-kernel can garanty CPU isolation between cells ?
>>  and if the answer is yes, how ?
>>     
>
> __As far as I'm aware__, there is no system partitioning provided by the kernel
> itself (yet?) in terms of scheduling. It all comes down to configure the
> priorities and timeslices of the respective (main) threads properly to have
> a system working in the way you want.
>
> For example, if I read your intentions right, your OK-Linux needs the same
> or a higher priority than your decrypt cell. If it is higher (or to be more
> specific; The priority of the Linux processes/threads are higher than that
> of your decrypt cell), Linux may render your decrypt cell starving.
>
>   
Robert,

I have performed the same test with a priority of 255 for the ok linux cell

and a priority of 100 for the decrypt cell and the system doesnt' boot 
better...

You can reproduce this test by just editing the main.c file of the 
decrypt cell

( just keep the whil( 1 ) {} without the body ) and the end of file 
okl4_3.0/cells/examples/SConscript

as follows:

if example == "decrypt":
examples_env.set_cell_config(priority = 100, spaces = 256, clists = 256, 
mutexes = 256, args = args)
else:
examples_env.set_cell_config(spaces = 256, clists = 256, mutexes = 256, 
args = args).

Concerning the timeslices, I don' t see any examples that use them 
explicitely ( except in ktest samples )

and don't know concretely how to reconfigure those timeslices in the 
cross-cell-demo example...

Thanks in advance for your help,

Hugues Balp.
> To sort "priorities" at the same scheduling priority you can configure the
> timeslices for each thread, which is by default at 10ms (at least in OKL v2.1),
> i.e. that way you can give Linux 67% of the CPU and the decrypt cell the
> remaining 33% (assuming there are no other threads running on the same
> priority level as these two).
>
>
>
> Robert
>
> _______________________________________________
> Developer mailing list
> Developer at okl4.org
> https://lists.okl4.org/mailman/listinfo/developer
>
>
>