Am 09.10.2010 um 13:19 schrieb David Chisnall: > meaning that we spend a lot more time sleeping in userspace waiting for the > kernel to wake us up
Let me ask a fundamental question: do we really have a performance problem to solve here? Are any measurements available that a typical application spends more than some % in this code? In my understanding, the timers and the untilDate: are handled only once per loop and then the kernel sleeps. Sometimes for minutes until the user moves the mouse or clicks a key. The same will happen with any alternative mechanism. I think your proposal will only simplify calculation how long to sleep and setting up the kernel to do so. But not improve the sleep time. The only situation where I can see that setting up the triggers as fast as possible is if you have timers that expire in fractions of milliseconds. But do we have a single example where this is required? Another area could be file handles for network stacks. Do we have a performance bottleneck there? I have a little the impression that the proposal is just for clearness of the code or "innovation" for innovation's sake? I fear a little by leaving POSIX we deviate from the least common denominator of all UNIX systems and introduce more dependencies. I don't want to argue against innovation, but it must really improve something. So I would propose that you work on a code branch that demonstrates that it works (better). Best regards, Nikolaus _______________________________________________ Gnustep-dev mailing list [email protected] http://lists.gnu.org/mailman/listinfo/gnustep-dev
