[ntp:questions] Re: ntpdate functions successors

David L. Mills mills at udel.edu
Fri Oct 8 17:12:45 UTC 2004


Harlan,

On NEVER stepping backwards. The IBM 9037 Sysplex Timer (MVS and zOS 
mainframes) will slew the system time over a five-second range. If for 
some reason it is necessary to slew beyond that range, the machine 
hollers for an operator to enter the time within five seconds. If more 
than five seconds off, corrections must be made five seconds at a time. 
Otherwise, the operator sets the correct time and the machine jumps 
directly to that time.

The operations manual suggests all time sensitive applications should be 
turned off before and restarted after the operator-induced jump. And, if 
the 9037 does something really nasty, all processors enter a 
non-restartable wait state and must be manually rebooted (IBMspeak is IPL).

I submit we have the identical capabilities already in the software 
distribution by suitable selection of command-line options and tinker 
commands. We should instruct anybody requiring no step backwards to do 
the same things as IBM mainframe operators. I suggest a command line 
option that enforces exactly the 9037 behavior.

I like this idea very much. I'll even post a link to the 9037 operations 
manual and defer to that manual anytime somebody gripes about this or 
that behavior about slew/step. Heck, I am told the IBM operators relish 
the two-hour (sic) interval late this month when the clocks are set back 
as an opportunity for annual maintenance.

Dave

Harlan Stenn wrote:
> Dave,
> 
> Thanks, now we're on the same page.
> 
> This means that somebody can use:
> 
>  ntpd -qc startup.conf
> 
> where startup.conf is, for example, a regular ntp.conf file that adds
> "disable ntp" and that will handle this case.
> 
> I think this covers all of the behaviors that were asked for in the list of
> "I used to do X with ntpdate, how do I do it with ntpd?"
> 
> The only other case may be making sure that the clock NEVER steps backwards,
> but I'm not sure that ntpd is the correct place to solve that problem.
> 
> H




More information about the questions mailing list