Re: [PATCH 2.6.11-rc4-mm1] connector: Add a fork connector

[Jesse Barnes <jbarnes@engr.sgi.com>]

On Wednesday, March 2, 2005 6:51 am, Paul Jackson wrote:
> Guillaume wrote:
> >   I also run the lmbench and results are send in response to another
> > thread "A common layer for Accounting packages". When fork connector is
> > turned off the overhead is negligible.
>
> Good.
>
> If I read this code right:
> > +static inline void fork_connector(pid_t parent, pid_t child)
> > +{
> > + static DEFINE_SPINLOCK(cn_fork_lock);
> > + static __u32 seq;   /* used to test if message is lost */
> > +
> > + if (cn_fork_enable) {
>
> then the code executed if the fork connector is off is a call to an
> inline function that tests an integer, finds it zero, and returns.
>
> This is sufficiently little code that I for one would hardly
> even need lmbench to be comfortable that fork() wasn't impacted
> seriously, in the case that the fork connector is disabled.

But if it *is* enabled, it takes a global lock on every fork.  That can't 
scale on a big multiprocessor if lots of CPUs are doing lots of forks...

Jesse