[ntp:questions] Re: IEEE 1588 support in NTP?

Ulrich Windl Ulrich.Windl at RZ.Uni-Regensburg.DE
Wed Nov 2 15:26:39 UTC 2005


"Richard B. Gilbert" <rgilbert88 at comcast.net> writes:

> David D. Hagood wrote:
> 

[...]

> Linux has been known to have timekeeping problems due to losing clock
> interrupts.   Each lost interrupt can introduce an error of  one or ten
> milliseconds.  If the kernel parameter HZ is set to 1000, the clock ticks 1000
> time a second and the system is likeliest to lose interrupts.  It works much
> better if HZ is set to 100.

I'd like to object: Some long time ago someone (I can't remember) made a test
over an extended period of time showing that Linux did not loose a single
interrupt. However depending on the drivers and crappy hardware (not to talk
about CPU frequency scaling and ACPI's various power saving modes), the system
may loose interrupts. For any hardware there is a rate of interrupts that
cannot be handled. For my 16MHz 386SX that rate was quite low.

Regards,
Ulrich




More information about the questions mailing list