Re: [ANNOUNCE] ktimers subsystem

["Paul E. McKenney" <paulmck@us.ibm.com>]

On Fri, Sep 23, 2005 at 01:09:46AM +0200, Roman Zippel wrote:
> On Thu, 22 Sep 2005, Thomas Gleixner wrote:

[ . . . ]

> > > The main difference between them is that the latter is user 
> > > programmable. 
> > 
> > wallclock is reprogrammable too and it introduces a bunch of horrible
> > functions in posix-timers.c. grep for abs_list. I explained why its
> > horrible already.
> 
> I said _user_ programmable, wallclock time is usually NTP controlled.

I believe Thomas is concerned about workloads that need a short-term
stable timebase.  For example, a process-control application might need
to accurately measure a (say) 1500-millisecond time interval.  Both
user-programmability and NTP adjustments to a given timebase could
destroy the needed measurement accuracy.

Such a workload does not need the long-term tie to wallclock time that
NTP provides, but it does need the accurate short-term timekeeping that
NTP cannot provide -- NTP sacrifices short-term accuracy in order to
adjust the clock as needed to gain long-term stability.

Thomas, John, please jump in if I am missing the point here.

						Thanx, Paul