Rajko M. wrote: > On Wednesday 05 December 2007 06:03:54 pm Carlos E. R. wrote: > >> The Wednesday 2007-12-05 at 22:48 +0100, Theo v. Werkhoven wrote: >> >>> don't know where you got that from, but ever since the IBM AT, PC's have >>> had a hardware clock on the mainboard, independent of the OS or user >>> programs. >>> The only thing that can make those thing fail is an empty battery or >>> broken crystal. >>> >>> The crystals that make these hardware clocks tick aren't the best in >>> terms of stability or accuracy of course, but with ntp for a daily >>> dosage of 'realtime', there's nothing wrong with the basic concept. >>> >> Er... >> >> However, that's not the clock the operating system uses when you ask for >> the time: not in Linux, not in msdos. Not in a X86 type PC, regardless of >> the operating system. >> >> There are two clocks on a PC. One is the CMOS clock, that runs out of >> battery, independent of system load, on an external chip somewhere in the >> mainboard (it is actually the same chip that holds the bios configuration >> data). However, the resolution of this clock is small, a second I think, >> and is slow to read. It can not be used as the system clock. >> >> This clock is only read once, when the system boots up, in order to set up >> the system clock. >> >> The system clock run originally as a timer or oscillator chip that >> interrupted the CPU about 19.2 times a second, and the CPU simply counted >> those interrupts, updating the "system clock". Today there are variations >> of this method, but the basis is the same. >> >> Suse programs this clock (in the kernel) to interrupt 250 times per >> second. Other possible settings are 100, 300 and 1000Hz. The kernel has >> been known to loose clock interrupts at the fast settings, specially the >> 1000 Hz one if a reiserfs is also used (because it dissable interrupts in >> some critical sections that last too long). This is known and documented; >> you can find references to this problem in the ntp bugzilla. >> >> >> Therefore, yes, a busy cpu can make the clock go slow. This is a fact. The >> kernel should compensate, though. >> >> However, my problem is often worse when the cpu is not busy at all. >> > > It seems that you can try to compile kernel with HZ enabled and set for > instance to 250 Hz and see if that helps. > > This is current status with all 10.3 kernels: > > ~> grep HZ /boot/config-2.6.22.13-0.3-default > CONFIG_NO_HZ=y > # CONFIG_HZ_100 is not set > CONFIG_HZ_250=y > # CONFIG_HZ_300 is not set > # CONFIG_HZ_1000 is not set > CONFIG_HZ=250 > > BTW, with current computer clocks in GHz range, 1000 Hz system clock rate > shouldn't be a problem (one tick on every few millions cycles), and so far I > know it is even recommended for desktop systems, although I'm not sure is > there any recent evaluation of gain. > >
FWIW I always recompile the suse kernel source with HZ=1000 on my gaming machine. Seriously, I get more frags, and get fragged less, that way... Joe -- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
