Re: Simple Network Time Protocol (SNTP)

On Fri, 10 May 2002, Smith, Dave wrote:

> Can someone using SNTP on an iSeries (V5R1) please respond to the following
> questions. They are being asked based on the fact that there is a "machine"
> time (QHOUR, QMINUTE, Qetc...) and a "system software" clock on the beloved
> iSeries.  SNTP keeps the system software clock in sync with our NTP server.
> I am aware some API's use the system software clock but most logging and
> commands and HLL programs use the machine time. Is that true?

Well I don't use ntp on the iSeries but I do use it on a lot of other
machines, so maybe I can say something decent.

I'm unsure specifically of the iSeries, but on other hardware platforms
the hardware clock can be accessed, but isn't normally.  So logging and
HLL languages (unless using specific APIs) don't access the hardware
clock.

> 1.  How often should these clocks be re-synced up?  Do they run at the same
> speed so that one can not get out of sync with the other?  Do the clocks
> re-sync at IPL?

Usually both clocks drift.  Some CPUs make the hardware clock drift
faster and some loads make the software clock drift more.  ntp keeps the
software clock correct.  My machines set the hardware clock to match the
software clock when shutting down and set the software clock to the
hardware clock when booting.

> 2.  What do you recommend in the spring and the fall when the machine times
> and the coordinated universal time offset need to manually be changed on the
> AS/400?  The customer I am working for generally changes the 'Q' values at
> the end of the next business day (when all their processes are done running
> in the spring and fall when the time changes for daylight savings.

NTP handles this for me.  Since I don't use it on the iSeries I don't know
specifics for it.

> 3.  Are there system (IBM) programs running that might use the API's using
> the software clock?  Thus how important is making the two time exactly
> match?  I know the potential is out there to someday run into a problem, but

I don't think it is terribly important that the two match.  If I don't
restart my machines then the hardware clock will be wrong by an hour for
six months of the year (yes my machines stay up long enough that the
hardware clock is wrong for six months and then becomes correct again).
As long as they are synced when the machine restarts there should be no
problem.  But an uncontrolled power loss can create trouble.

> 4: Any idea what IBM has in the works to automatically keep the two times in
> sync?  Is this pretty standard in the "Midrange/Mainframe/UNIX" world to
> have two clocks running? What about the general "network" (Microsoft)
> community?

Well on unix anyway it is pretty very common.  There are some hardware
timers associated with the hardware clock that can be used in programs to
generate time intervals.  On my machines there is a program that syncs the
two clocks.  But I only run it on shut down/start up.

> 5. Is the TIME op code in an RPG program retrieving the software clock or
> the hardware clock?

Software.

> In the end, I guess I just want to know what SNTP alone buys me?  Why am I
> forced to (write or download) the programs to keep the clocks in sync?  It
> makes no sense to me.

The hardware clock is mainly used to set the software clock when starting
the machine.  I don't even worry about the hardware clock usually.

James Rich
james@eaerich.com