On Mon, 3 Nov 2003, Eric Sammons wrote:

> I am currently interested in how folks are performing time
> synchronization updates?

NTP is commonly used in discrete server environments.  The "transportation
company" I am associated with uses it in their Linux under z/VM
environment, with the result that the time on the Linux images is rarely
the same time as z/VM has (since no-one there has investigated bringing
z/VM into the world of NTP).

> And finally, if the time is managed in this fashion are the linux guests
> continuously updated with the correct time or is this a once at boot
> thing?

With NTP, the theory is that the NTP daemon runs for a sufficient length
of time to determine how much quicker or slower the given system's clock
runs compared to the time reference.  Once this is determined, the daemon
regularly makes tiny changes to the system time to keep it in-sync with
real time.

My concern about this under z/VM -- particularly when the timer patch is
installed -- is that we do not know how often or how consistently the
clock skew value will get calculated (and applied) when the amount of CPU
the guest is allocated may vary according to load.  The NTP protocol is
remarkably complex but I'm not sure it takes virtualisation into account.

At the airline, we've been having strange time synchronisation issues on
guests running NTPD, and they're not even using the timer patch.

Apart from this, you have yet another process sitting on your guest that
wakes it up periodically.

I wonder whether a better approach would be to run a regular ntpdate
command instead of running an NTP daemon.  The zSeries clock should not
drift that much, so once a day should be more than enough.

Vic

Reply via email to