>
> Ok, I never cared much about it the number of jobs to run in paralel
> with make, but I heard, that you should use as many jobs as
> processors/cores you have. One page I googled, even suggested to use
> 1.5 times the number of processors.
>
I recently helped redo our make process at work, which included using parallel builds, and doing it through Eclipse.
Eclipse has a "use optimal number of jobs" setting, but all this does is pass the -j flag with NO argument to make. Apparently this actually spawns ALL jobs which can be parallelized, all at once when make begins. Somehow on my and several coworkers' machines, this works just fine, and gives equivalent performance to the fastest time given from experimenting with various explicit numbers of jobs (which happened to be 1.5 times the number of processors for us...we all have 4-core machines). But...not passing an explicit number of jobs fails miserably on other machines, which more predictably run out of memory and/or process IDs and the build crashes.
This discussion prompted me to investigate further, coming up with 'psrinfo' on Solaris and the NUMBER_OF_PROCESSORS environment variable on Windows which can be used in the same way as the tip being discussed here.
I wonder if we want to modify the tip to use 1.5*CPUs instead of 1+CPUs? This would also prevent us from starting multiple jobs on a single-processor system, if we round down or use integer math. It should be fairly easy for a user who decides they want a different number of processes to modify the arithmetic.
--
You received this message from the "vim_use" maillist.
Do not top-post! Type your reply below the text you are replying to.
For more information, visit http://www.vim.org/maillist.php
No comments:
Post a Comment