Arjan van de Ven wrote:
the theory seems to be that HZ=1000 gives a better interactive response. (you can measure a lot of things better with a 1ms sample period than with a 10ms sample period). Another reason is that with HZ=1000 userspace can sleep more accurate intervals, which is especially helpful for video playback (if the least you can sleep is 20ms, doing 60Hz playback sucks, but if you can do 2ms in 1ms increments you get a lot better) I'm not so convinced either way to be honest; for a while I had HZ=100 patched into the kernel, but people complained so I set it back to 1000 way back. Now that I have a laptop with powersaving I think I'll build myself a kernel with HZ=100 again.
I never understood the kernel timer concept, why not build a wakeup list and sleep exactly the time to the next event??