Re: SMP Scheduling

Hello,

let my try to answer those questions of which I think I know the answer ;-)

>  - Threads have priorities. Threads in one priority class are served
>    round-robin. Higher priority queues are served first.

As far as I know, this is correct.

>  - Pistachio always creates threads on the first CPU unless told
>    otherwise via L4_ThreadControl.

Nope, I think that's not correct. Threads are assigned to the CPU on
which ThreadControl is executed and can be migrated afterwards using
"Set_ProcessorNo", which is derived from the "Schedule" syscall. Or did
I get your question wrong?

>  - Because preemption messages are sent with timeout 0, one would need
>    at least one thread per CPU that handles preemption
>    messages. Otherwise one could lose messages when the scheduler is
>    busy.

I'm not sure about this, but it sounds reasonable. It is generally a
good idea to offer at least one thread per service and CPU for reasons
of parallelism and to avoid communication between threads on different CPUs.

Perhaps [1] might also be interesing for you.

Hope this helps,
Philipp

[1]
Towards Effective User-Controlled Scheduling for Microkernel-Based Systems
Jan Stoess
ACM SIGOPS Operating System Review, Special Topics on Secure
Small-Kernel Systems, July 2007
http://i30www.ira.uka.de/research/documents/l4ka/2007/stoess07ul4sched.pdf