[H-GEN] distcc at meetings

Greg Black gjb at gbch.net
Tue Apr 13 02:56:47 EDT 2004


On 2004-04-10, Mark Suter wrote:

> If we all try to run distcc at meetings, then the following will
> allow people to quickly take advantage.
> 
>     $ sudo nmap -sT 192.168.146.0/23 -p 3632 -oG - | awk '/open/ {print $2}'

I've been using ccache[1] (from the same factory as distcc and
samba) for ages, but had not previously actually tried distcc.

When I got home, I set it up on some of the boxes on my network
and did some testing.  The test compile was a full kernel
compile on a FreeBSD box, which is a big enough job to provide
meaningful data (and to profit from speeding up).

Before each compile, I ensured that all the results of previous
compiles were removed.

The machines used were a Celeron-900 256 MB laptop (the client);
a Celeron-2200 1 GB desktop and a Celeron-2600 1 GB desktop (the
servers).  The network is a 100 Mb/s Ethernet.  The servers have
sufficient extra grunt and disk speed that you'd expect them to
help the puny laptop client.

I first tried doing everything that could be offloaded on the
servers with the client just doing preprocessing, linking and
job management.  That took 402s.

Then I tried the same again, but including the laptop in the
mix; that took 375s.

Doing a clean compile on the laptop without using distcc, but
using ccache with a completely empty cache, took 731s.

But a second compile on the laptop, after a "make clean" and
using ccache again, took only 283s.

So distcc certainly saved time for a first build, but nowhere
near as much as ccache.  And, since ccache is so much simpler to
use (and can help without changing anything in your makefiles)
and has no scary networking issues, I think I'll just stick to
using ccache and forget what I learned about distcc.

Cheers, Greg

[1] Find ccache at <http://ccache.samba.org/>




More information about the General mailing list