[distcc] Adding (relative) slow machines

[Martin Pool]

On 25 Feb 2003, Christian Rubbert <distcc at rubbert.de> wrote:
> Hey!
> I'm wondering, does adding slower machines generally slow down the
> build process?

Well, I know a good way to find out. :-)  Why don't you measure it and
tell us?

Make sure to list the slower machines last.
 
> I got a 1 GHz AMD Athlon, a 2 GHz P4, a Celeron 300 and a 400 MHz one.
> Now i'm wondering if those two slow machines do their bit of help,
> or if they hold up the rest because of some scheduling issues?

It could go either way.

It we're lucky, they could do a bit of work and save the other
machines some time.  Though the P4 (with bigger caches and more
execution units (?) as well as faster cycles) is probably so much
faster than the Celerons that it may not show up.

The problem is that Make issues tasks in a way that's not under
distcc's control or knowledge.  If the last compilation for a
directory is very large and happens to be sent to the 300MHz machine
then the other machines might be waiting around a long time for it to
finish.

You might try something like (assuming the Athlon is local)

  HOSTS='localhost/2 p4/4 cel3/1 cel4/1'

to put only one job on those machines.

Please let us know what you discover.

-- 
Martin 

What is striking is how little innovation there has been in the bits
of the market that Microsoft dominates, and how much where it has
little influence.
	-- The Economist
	   http://www.economist.com/displayStory.cfm?Story_ID=1429528