DragonFly kernel List (threaded) for 2008-06
Re: User API still a goal?
Nicolas Thery wrote:
> 2008/6/10 Robert Luciani <email@example.com>:
>> If the "micro threads" don't have to be POSIX compliant, aren't there
>> already many languages which provide plenty of concurrency within
>> threads? Some are even SMP aware and don't do stupid scheduling decisions.
> Out of curiosity, where do these languages get SMP awareness from? Do
> they need some kernel-level support?
The one I'm sure of is Erlang, in that its virtual machine uses one
POSIX thread for each processor and then can distribute its
coroutine-like concurrency primitives amongst them. Obviously programs
written in erlang use the language-specific functions which can be
handled internally in the VM before resulting in bad performance.
Here at Chalmers University we're making gcc addons that thread
independent code segments so there's no chance for deadlocks etc.. so
that the number of processors is irrelevant. I'm sure that the new Intel
compilers and C++0x compilers would do things along the same lines.
Chalmers University of Technology, SWE
Department of Computer Science and Engineering