In article <200409212243.i8LMhvC17297@srapc342.sra.co.jp>,
	SODA Noriyuki <soda@sra.co.jp> writes:
> Acutally I did some benchmarks about basic therad primitives
> last year.

Does that involve using the kernel? Userland scheduling works probably
great as long as your threads don't need to use the kernel.

> I suspect the reason why Solaris M:N thread is slower is because
> generally it doesn't assign enough lwps to pthreads.

It did that dynamically. I think the 1:1 model works very well for
Solaris because the Solaris kernel internally uses threads and
preemptive scheduling, too. So with the 1:1 model each thread can
use the kernel directly whenever it wants to without blocking any
other userland thread.

	Kind regards

-- 
Matthias Scheler                                  http://scheler.de/~matthias/