[distcc] Scaling?

[Martin Pool]

On  5 Feb 2003, Jaan Oras <joras at itcollege.ee> wrote:

> I just tried distcc in our school and i found out that using 14 computers 
> the speedup was about 6 times. Why isnt the speedup bigger? What
> limits it?

A few possibilities, in order of likelihood:

 - some non-parallelizable work, such as linking

 - distcc's scheduler is leaving machines idle

 - makefile inefficiencies (can't issue enough parallel jobs)

 - limited by outgoing bandwidth on the client

 - distcc or cpp overhead

I hope to improve the scheduler a bit in the next release.

One thing you might try is to turn on verbose logging on the client,
and then examine the log file using grep to see which machines are
receiving jobs.  Are they spread across all 14?

-- 
Martin