[distcc] Do the number of jobs make any sort of difference?
martinpool at gmail.com
Fri Sep 9 03:56:17 GMT 2005
Note that -j200 is pretty much equivalent to -j with no parameter,
i.e. "run as many as you can".
I expect if you plut them you'd see a curve that goes up from 1 to
about 50 or 60 jobs, at which point the client machine becomes the
bottleneck - it can no longer dispatch jobs fast enough, so it
wouldn't matter if you had a thousand machines waiting for work.
The bottleneck can be
- client machine's cpu saturated (can't start tasks or run cpp fast
enough; can try turning off ssh)
- network saturated (in which case try compression)
- client machine's disk or memory saturated (can't bring in source
and write out objects fast enough)
- servers saturated
More information about the distcc