On Fri, 18 Feb 2000, Ulrich Weigand wrote:
> Linux deliberately does not allow processes to decide themselves
> on which processor they want to run: the reason for this is that
> the optimal processor allocation strategy depends on what *else*
> is running at the same time, and any single application simply
> doesn't *have* that information.
I would argue that it's more an issue of simplicity. In Solaris, an LWP
can pin itself on a given processor. There are certainly circumstances
when this is useful.
But Linux doesn't want to be Solaris; although it has progressed to a
reasonable degree of complexity, decisions are still made favoring
simplicity over functionality. For example, the threading model is notable
because it is so simple to implement, as opposed to DEC UNIX's scheduler
activations, or Solaris' LWP/thread multiplexing.
Phil
