[distcc] Enabling lzo compression?

[Dara Hazeghi]

--- Martin Pool <mbp at sourcefrog.net> wrote:
> On 19 Jul 2003, Dara Hazeghi <dhazeghi at yahoo.com>
> wrote:
> > Hello,
> > 
> > I want to do some measurements of current distcc
> > performance with and without lzo compression.
> > Unfortunately I haven't been able to find what
> > flag/option does this. Any hints? Thanks,
> 
>   DISTCC_HOSTS='localhost otherhost,lzo'
> 
> Mentioned in NEWS, and will be in the manual.

Ah, thanks. Guess I should look harder next time.

I did a little testing building gcc's cc1 with

DISTCC_HOSTS="draco/3 penguin/1 cygnus/1

where draco = 2x733MHZ P-III, penguin = 333MHZ P-II
and cygnus (localhost) = 667MHZ ppc. Machines were
connected via 10baseT.

I'm guessing that with more machines lzo would make a
bigger difference. I was most surprise that the
performance hit of ssh was so little.

make CC=gcc -j1
real    10m18.664s
user    8m11.180s
sys     1m16.070s

make CC=dcc -j1
real    8m39.949s
user    0m32.820s
sys     0m22.330s

make CC=dcc -j1 (lzo)
real    8m5.593s
user    0m33.510s
sys     0m22.360s

make CC=dcc -j1 (ssh)
real    10m17.584s
user    1m37.000s
sys     0m28.440s

make CC=dcc -j1 (ssh,lzo)
real    9m47.885s
user    1m33.850s
sys     0m27.600s

make CC=dcc -j5
real    3m16.084s
user    1m44.180s
sys     0m34.360s

make CC=dcc -j5 (lzo)
real    3m23.241s
user    0m34.370s
sys     0m22.520s

make CC=dcc -j5 (ssh)
real    4m12.116s
user    1m36.030s
sys     0m29.480s

make CC=dcc -j5 (ssh,lzo)
real    4m55.420s
user    1m35.130s
sys     0m27.690s


Cheers,

Dara