On 8 Jan 2003, Paul Millar paulm@astro.gla.ac.uk wrote:
Hi Martin,
On a similar note, I've been investigating OpenMOSIX on and off (mostly off) for doing distributed compiling. In general, I've found it not particularly stable (although its improving) and the performance not particularly great -- although I've not had time to investigate the causes of these problems. Besides, compiling code is quite a tricky task for it to schedule/migrate.
I haven't tried it myself but I would have expected the short, intense jobs generated by compilation to be a problem.
A reduction in compilation time by a factor of 2.6 (for three machines) isn't bad at all!
If you have the time, could you repeat this with different values for the -j option? Timing how long it takes with -j2 (and perhaps -j3 and -j4) would be interesting.
Yes, I've been meaning to add that to the benchmark tool. I'll let you know.
Also, repeat the compilation several times for each value of -j and taking the mean and standard deviation would be interesting too.
In informal measurements it seems to be quite reproducible. The sd is typically only a few percent of the overall time. But that might be an interesting option to add too.