On Sat, Jun 07, 2008 at 01:04:16PM +0200, Alfredo Braunstein wrote: > > Typically -j3 is a better choice than -j2 on dual core machines, as not > > everything is processor-bound. > > Sure, maybe the reason. Note that multiple concurrent processes intensively > accessing one resource (say memory or disk) may give worse performance than > sequential runs (be it because of memory page switches, suboptimal cache > usages etc). Seemingly this is not the case for compilation (or just lyx > compilation)...
I think it's rather generally true for compilation and scales fairly well to distributed compilation so it seems to be rather a "I/0 thing" I.e. if you have eight dual-core machines in the farm, the optimum is around -j 20. [But I never tried to actually benchmark that, so this is pretty empirical...] Andre'
