On Tuesday 03 November 2009 20:36:37 Paul Gilmartin wrote:
> On 11/03/09 19:25, Leslie Turriff wrote:
> > On Tuesday 03 November 2009 20:09:49 Paul Gilmartin wrote:
> >> On 11/03/09 17:42, Leslie Turriff wrote:
> >>> Philosophical question: Has anyone posted a requirement to SHARE
> >>> or some such to get an interface to an NTP source into the firmware so
> >>> that the TOD clock can be synchronized to a reliable source (other than
> >>> the operator's wristwatch or the computer room's wall clock)?
> >>>
> >>> I've wondered about that ever since Linux-390 happened. :-)
> >>
> >> Often discussed on IBM-MAIN. IBM's position appears to be, "We have
> >> a [S]NTP server, but our clocks, with STP, synched to NIST, etc. are so
> >> much better than any others that it would be inappropriate for us to
> >> be an NTP client." IBM doesn't get a very good grade in "Plays well
> >> with others", IMO.
> >
> > But that sounds to me like a different issue. Does that, in fact,
> > provide the capability for the machine to obtain an accurate TOD value
> > from some external NTPish source, without operator intervention?
>
> May I paraphrase: "We don't need no steenkin external NTPish source!"
>
> I don't know what Linux-390 does. I'd be inclined let the TOD clock
> do whatever it damned well pleases, create a software clock in the kernel,
> and let NTP steer the offset between that and the physical TOD. But
> that would glitch badly whenever the hardware clock was reset to the
> correct time.
>
Of course! It needs to be done long before any OS gets IPLed, since the
(real) TOD clock is set during the POR process.
Leslie
