An anonymous reader writes:
KernelTrap has an article about some recent benchmarks posted to the Linux Kernel mailing list, comparing the much discussed Completely Fair Scheduler to the original Linux process scheduler, as well as to Con Kolivas' staircase scheduler. Reviewing the resulting graphs, it seems that the Kolivas staircase scheduler can offer better performance. Ingo Molnar performed additional tests and concluded that the extra overhead comes from being completely fair and providing precise scheduling. He offered a patch to improve the performance of the new scheduler, but cautioned that as with other schedulers it comes at the cost of accuracy, 'tasks working off the timer tick could steal away cycles without being accounted for them fairly, and could starve out all other tasks.' Ingo added that the performance improving patch will not be merged into the Linux kernel as accuracy is more desirable than micro benchmark numbers, 'unless the difference was really huge in real life — but it isn't.' The bottom line is that the new process scheduler will offer a significantly smoother desktop experience as compared to the old Linux process scheduler.
