[distcc] Results: massively parallel builds, load balancing, etc.

[Martin Pool]

(Please just go ahead and post the code, or a link to it.  Ugly
experimental code is nothing to be ashamed of.)

Scheduling across larger clusters or with more than one active user
could certainly be better than in 2.17.  The main thing is to add some
kind of better idea of which servers are busy, either from distcc jobs
started by other clients, or from other work.

An alternative approach is to have the client connect in parallel to
several servers, and ask each of them for this information directly.

-- 
Martin