[1240] in Kerberos
Time Synchronization for VM and MVS (Chapter 2)
daemon@ATHENA.MIT.EDU (Richard Hintz)
Fri Feb 8 15:47:07 1991
Date: Fri, 08 Feb 91 12:13:19 PST
From: Richard Hintz <OPSRJH@uccvma.ucop.edu>
To: kerberos@ATHENA.MIT.EDU
I've gotten a couple of private notes so perhaps I should clarify
my concern. I admit I'm probably jabbering on here with a naive
view, so you should take that into consideration.
I am thinking of a situation such as:
1) Kerberos server runs on VM with manually set clock.
2) Application client which refers to Kerberos server is on a
different VM machine also with a manually set clock.
3) Application server which refers to Kerberos server is on yet
another VM machine with a manually set clock.
Just for fun I checked the system time of my home VM machine. It
was off from UTC (as reported by my ntp set Mac) by 2 minutes,
40 seconds (not the 2:40 seconds I mentioned to some).
I think most VM sites would believe this was accurate time.
For what it's worth, multiple MVS systems sharing a spool need
only be withing 120 seconds of each other to be sync'ed up.
At least, that the delivered system default which gives some
idea of what time sync means in the mainframe world.
Until the advent of the Sysplex timer, there was no reliable,
accepted way to bring an IBM mainframe into time sync with UTC.
My concern is for the case, which I believe would be fairly
realistic, when the total time offset from machine 1 to machine 3
would be greater than 5 minutes.
Richard Hintz opsrjh@uccvma.ucop.edu opsrjh@uccvma
University of California
