Planned SMP Improvements

[Sebastian Huber]

On 01/02/16 14:39, Darshit Shah wrote:
>> >Maybe we can implement
>> >
>> >Improve the SMP scheduler with arbitrary processor affinity support
>> >https://devel.rtems.org/ticket/2510
>> >
>> >as part of this years GSoC.
>> >
> I've already mentioned that I'd be interested in picking this up as a
> potential project. However, the entire implementation may not be
> possible in a single summer. It is rather complex, and while a naive
> implementation may be possible to write, we would want something more
> efficient. I'm hoping for some pointers on where I should start
> looking for a better understanding of the codebase.

It would definitely help to have a co-mentor from the MPI (e.g. one of 
the authors of the paper). The SMP scheduler is a rather self-contained 
part of the operating system. I will probably add some internal 
documentation in the next months. It would be really nice if we could 
run the tests of the original paper.

>
> However, apart from this, even Eliminating the Giant Lock and
> Implementing OMIP could potentially be GSoC projects. These may not be
> extremely large changes, and you could probably finish them in a
> couple of weeks, but they would take someone new much longer.
> Reasoning about the new code will not be trivial.
>

Implementing OMIP is very complex. The Giant lock elimination is complex 
in some areas (thread restart/delete) and quite boring in most areas. I 
guess some trouble will arise in case I accidentally change some code 
patterns.

-- 
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber at embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.