Re: [PATCH] Re: boot time, process start time, and NOW time

[john stultz]

On Mon, 2004-08-30 at 16:00, Tim Schmielau wrote:
> So I think we should not apply the patch, but rather back out the patch 
> that rebased uptime on a ntp-corrected timesource.
> There are too many statistics that are still based on jiffies or clock 
> ticks, and we cannot immediately change that without a large rework
> (although this might eventually happen according to John's proposal).
> And mixing two different timesources just won't work, regardles where we 
> draw the borderline between them.
> 
> George, please excuse my lack of understanding. What again where the
> precise reasons to have an ntp-corrected uptime?

If I remember correctly, folks were complaining that boot time was
drifting due to the same issue. 

So yes, a full rework of the time subsystem is needed, but it alone
won't fix all of these problems, its just the first step. Once we have a
sane time base that isn't dependent on regular timer ticks, we then need
to make the timer subsystem and every other subsystem to use that time
base instead of Jiffies/HZ. 

This isn't going to happen instantly by any means. I'm trying to get the
time of day rework finished as soon as I can, but I've got the day job
to do as well. In the mean time, we can staple gun any user visible
exported HZ/jiffies values so they are accurate (using ACTHZ or
gettimeofday), and also look into changing HZ to a less error-ful
value.  HZ=1001 has been suggested and looks quite promising (although
/net/schec/estimator.c wants a power of 4).

thanks
-john

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/