-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Wednesday 2007-12-05 at 19:20 -0600, Rajko M. wrote:
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.
I have it at 300 now, and had it at 1000 for a little while. The default is 250. I don't think that's the problem, though, because it would be worse when the system was the busiest. I might try lowering it to 100, but that might be a problem for multimedia. There are other settings I might try twiddling: [*] Tickless System (Dynamic Ticks) [ ] High Resolution Timer Suppor ┌────────────────────────────────── Tickless System (Dynamic Ticks │ CONFIG_NO_HZ: │ │ This option enables a tickless system: timer interrupts will │ only trigger on an as-needed basis both when the system is │ busy and when the system is idle. │ │ Symbol: NO_HZ [=y] │ Prompt: Tickless System (Dynamic Ticks) │ Defined at kernel/time/Kconfig:8 │ Depends on: GENERIC_TIME && GENERIC_CLOCKEVENTS │ Location: │ -> Processor type and features │ Selects: TICK_ONESHOT │ CONFIG_HIGH_RES_TIMERS: │ │ This option enables high resolution timer support. If your │ hardware is not capable then this option only increases │ the size of the kernel image. │ │ │ Symbol: HIGH_RES_TIMERS [=n] │ Prompt: High Resolution Timer Support │ Defined at kernel/time/Kconfig:17 │ Depends on: GENERIC_TIME && GENERIC_CLOCKEVENTS │ Location: │ -> Processor type and features │ Selects: TICK_ONESHOT The thing is, I don't know how these settings were in 10.2 or earlier.
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.
There are two things. No matter how high the frequency is, the cpu should cope. If the kernel code has to disable interrupts for so long a time as to loose interrupts, that's a bad design, IMO. At worst, interrupts would queue and be attended later, in priority order. Second, the clock counter should not require the cpu at all, at least not to keep track of time. The counting should be independent. But that's an architecture design change. IMO, of course. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.4-svn0 (GNU/Linux) iD8DBQFHV3IMtTMYHG2NR9URAkOFAJ9M/RhZTg2TfUm1NY78YBQm4J1I/QCfepAS deB3rDuRcu7mlYXgsBNGN0o= =tOdr -----END PGP SIGNATURE-----