[distcc] n00b: What

Enric Martinez runlevel0 at gmail.com
Sat Nov 13 12:45:18 GMT 2004

Hi Martin,
> > The results are that at least nothing crashed, but I can't
> > tell if things where faster or not.
> Run 'top' on the server and see if compilers are running.

Sorry, I can see that you misunderstood me in someway.
It was 3:00 PM and I wasn't quite awake...

I also hit the 'send' button too quick, so the subject was wrong.

Was I was trying to do was to emulate and stress test a system in
which compilation was not shared, but indeed delegated. The target is
building a service
which would do the most possible part of the compile process, so that
a source based distribution would have no meaningful difference in
installation time as a binary based distro. This service will be based
for the first stage on 3-4 CPU (uniprocessors) connected via LAN and a
server to handle client access...

The results where impressing, what I was telling you about the timings
was that I indeed was not interested on how much time I saved, but on
the possibility of doing
jobs remotely and how the host behaved (max. 60% system load running
7-9 processes, while only 2 on the client, amazing).

Compile time must forcible have been shorter, as the host machine is
an Athlon-XP 2500+ Mobile 1.8 GHz, while the client is only a humble
Duron 1.3 GHz. I will run timed tests once we are working on the
*real* compile farm.

The network load was never above 2.7 MB/s, and the average was 0.7
MB/s, low enough for a DSL connection (I know that the results on a
LAN are different, but at least I can have an idea of what we have to

So, the results with my weird settings where not only the expected, I
even got no complaints during the whole emerge -u world (koffice,
parts of KDE, gcc ...)

The final target is, as I said, to set up a system which a user could
use remotely accessing via web server which would then the send the
jobs to a network of compile farms hosted by volunteers using low-end
hardware to run light-weight compile farms... We thing that we will
have the first stage up and running for xmas...

So, hats off, your app is really killer and IMHO it could represent a
revolution in the way we actually consider Linux distros, and perhaps
even in the way we today consider managing software. Think about the
implications this could have for enterprises which would be able to
have 100% customized and optimized software running with the same
effort or spend time as with binary distributions and not needing
extra hardware or waste cycles to compile things by themselves.

P.D.: I forgot to say that on the host machine distcc was running with
nice = -1...


More information about the distcc mailing list