[distcc] Multi CPU machines

[Jonas Jensen]

On Tue, 2002-11-19 at 11:54, Jean-Eric Cuendet wrote:
> Thanks, but the patch is corrupted at the URL.
> Could you resend it to me?
> Thanks.

Here it is, decoded:
http://article.gmane.org/gmane.comp.compilers.distcc/416

> This patch is a solution to send more than one job at each CPU. But the solution should go beyond that.
> I think that the dispatch should be completely dynamic. I see our dev environment being composed of dev machines for developpers and compile farms to speedup the compilation. These machines should be loaded with compilation from all the developers at the same time, when they run compilation.
> But then, the spawned jobs should be sent depending on the current load of each compilation server.
> This could be achieved if the server send an average of its load before/after each compilation sent.
> How hard would it be to modify the distcc code so the server send the client the laodavg?

That would be Martin's call, but I think it falls under the "premature
optimization is the root of all evil" category, as described in the FAQ:
http://distcc.samba.org/faq.html#central-controller
Also, it probably not even an optimization, as distcc has near linear
scalability already: if a host is heavily loaded (or just slow) it'll
take longer to finish the job, and therefore it'll get fewer jobs.
I'd rather compile on a 50% loaded Athlon XP 2000+ than a 1% loaded
Pentium 200MHz.

-- 
Jonas Jensen <jbj at knef.dk>