On Tue, Sep 14, 2004 at 09:50:08AM +0200, [EMAIL PROTECTED] wrote: > I can not say anything about the performance of 1:1 linux threading model. > I can not say that it is supperior or anything like that. I have not > done any benchmarks. I can just say that the design of this model has > some drawbacks.
Those drawbacks[0] must be balanced against the drawbacks of any other design, including specifically M:N and user space threading. Specifically, Linux context switches are so fast (relatively speaking) that the context switch overhead of 1:1 threading is considered to be worth it, when you take into account the added complexity of the M:N design and implementation[1]. > http://web.mit.edu/nathanw/www/usenix/freenix-sa/freenix-sa.html With a little bit of digging, you'll find papers that say 1:1 is better, user threading is better, and M:N is better. It all depends on the design, implementation and *benchmark*. [0] Proving the tautology "any design has drawbacks" is left as an exercise to the reader. [1] Thus spoke Ingo Molnar, who implemented the kernel parts of NPTL. I can find the exact mail on lkml if you insist. Cheers, Muli -- Muli Ben-Yehuda http://www.mulix.org | http://mulix.livejournal.com/
signature.asc
Description: Digital signature
