Re: Looking for point to compare SA and 1:1 threading model

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, May 27, 2009 at 12:44:53PM +0100, Jim Whitehead II wrote:
> I am current in the process of researching the effect of various
> threading models on programs that are structured in the OCCAM.  In
> particular I am looking to make a comparison of scheduler activations
> versus a 1:1 threading model for a very specific class of programs.
> Obviously FreeBSD and NetBSD both implemented some form of scheduler
> activations in the 90s, but both abandoned the threading model in
> favor of the current 1:1 model.
> 
> I am looking to find a point in the NetBSD version history where I can
> make a fair comparison between the two models.  In an effort to avoid
> causing a debate regarding the advantages of disadvantages of each
> (something that I see has been exhaustive on this list alone), I'm
> wondering if anyone can provide help with the following questions:
> 
> 1. Did the SA system in NetBSD ever get multiple CPU support [1] for
> threaded applications?

not officially, although some threaded application worked just fine with
PTHREAD_CONCURRENCY > 1. Unfortunably this got disabled in netbsd-5.


> 2. How is the kern.no_sa_support option implemented?  Specifically, is
> there a large amount of overhead associated that will bias a
> comparison between having the option on and off?

Look for sa_system_disabled in /sa/compat_sa.c. overhead is low.

> 3. Is there any point in the SCM history, or released versions that
> would be appropriate for such a comparison?

I think the best would be to test in on the 5.0 release. Use a
netbsd-4 chroot to test sa programs.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--