On 7/27/07, Rafael Vanoni <[EMAIL PROTECTED]> wrote: > It does sound a little black box, but if I'm not mistaken the idea was > no to limit the number of user threads, despite the limited number of > kernel threads. libthread decided which user thread would run and to > which kernel thread they'd be assigned to. Allowing the existence of > more user threads than kernel ones. So it makes sense to say that it > scales well.
Also, when a thread yields in a M:N thread model, the userland scheduler can pick another thread to run. On the other hand, a 1:1 system needs to get the kernel involved, and thus needs to cross the user/kernel boundary. There is still some performance advantage in M:N, but modern processors are just so fast that the advantage is much smaller, but handling signals in M:N is a bigger headache. Rayson > > Rafael > _______________________________________________ > perf-discuss mailing list > [email protected] > _______________________________________________ perf-discuss mailing list [email protected]
