On Wed, 13 Nov 2002, Billy Biggs wrote: > > Most of those apps should be ok in 2.5 with normal timers - the select > > resolution is 1Khz in 2.5.x > > So, just to make _sure_ I've got this straight, does this mean that > usleep() will suddenly be 1ms-accurate, or do I need to do something > special? Also, is there some easy way to detect this accuracy? Something to look at, which I've wanted to see in the stock kernel for a long time, is "firm timers": http://www.cse.ogi.edu/~luca/firm.html The idea is to use the APIC when available to increase timing resolution down to the FBS frequency (66 to 133MHz typically). Instead of waiting for the HZ counter to come around and wake things up, the APIC can be programmed in mere cycles (PCI memory writes instead of outb) to interrupt ridiculously close to the actual requested wakeup time. IIRC there's some similar work going on at Montavista or somesuch, the goal being to totally remove the HZ tyranny from Linux <g> Erik Walthinsen <omega@xxxxxxxxxxxxxxxxxx> - System Administrator __ / \ GStreamer - The only way to stream! | | M E G A ***** http://gstreamer.net/ ***** _\ /_
