Subject: Re: 1:1 threading model
To: Nathan J. Williams <firstname.lastname@example.org>
From: Bang Jun-Young <email@example.com>
Date: 01/10/2003 20:01:56
On Wed, Jan 08, 2003 at 03:01:16PM -0500, Nathan J. Williams wrote:
> Bang Jun-Young <firstname.lastname@example.org> writes:
> > Ulrich Drepper said that he succeeded starting and stopping 100,000
> > threads in 2 seconds in his article. Other people on the NPTL mailing
> > list reported good results with NPTL. So it doesn't look like just a
> > belief.
> Well, without any contrasting results from a M:N library with a
> similar amount of effort put into it, the "100,000 in 2 seconds"
> result is meaningless. It sounds big, sure.
Hmm, although I'm not a Linux evangelist, that's not true. Development
version of Linux already has kernel support for NGPT, currently de facto
standard M:N threads library for Linux. Since NGPT is backed by IBM, I can
say similar (or, maybe, larger) amount of effort is being put into it.
However, by no means I'm saying that M:N or scheduler activations is worse
than 1:1 model. I just liked to know if coexistence between two models in
the same kernel was possible.
> But compared to what?
The current clone(2) implementation in Linux.
Bang Jun-Young <email@example.com>