Re: Fwd: [autoconf] Re: autoconf - threads [paraller] - multicore CPUs

[Tim Post]

On Wed, 2008-07-23 at 11:21 +0200, litlle girl wrote:
> 2008/7/23 <address@hidden>:
> 
> > How do you split existing configure into 2 parts:
> > configrue1 and configure2?
> 
> Yes, this can be a problem, you're right.
> 
> But what about second idea: in ./configure
> odd commands run normally
> and even with &
> Last 3 (5?) commands run normally to have sure that tasks at second core
> have beed ended.
> Then gather the results from subtask.

That would be _really_ racey. Odd would often have to block while
waiting for something even needs to do, or vice-versa.

The end result would (likely) be a much slower configure using drop
files or SIGUSR* to play leap frog, yikes! You'd go from one core being
heavy to both cores sweating.

Have you ever used OpenSSI or Kerrighed? Writing scripts that distribute
work amongst nodes with some kind of remote fork is very problematic,
this would be very similar.

No matter what, we come down to the confines of the interpreted
languages being used.

> How to gather the results?
> put results from each sh scripts into: the array or separate files (and
> include them later?)

Unfortunately, arrays are not supported by many shells. That brings us
back to files without the benefit of (fast) exclusive locks.

Regards,
--Tim

-- 
Monkey + Typewriter = Echoreply ( http://echoreply.us )