Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!snorkelwacker.mit.edu!shelby!MIT.EDU!Saltzer
From: Saltzer@MIT.EDU (Jerome H Saltzer)
Newsgroups: comp.protocols.kerberos
Subject: Re: Time synchronization
Message-ID: <9102061912.AA04465@PTT.LCS.MIT.EDU>
Date: 6 Feb 91 19:12:43 GMT
References: <11981@darkstar.ucsc.edu>
Sender: Saltzer@mit.edu
Organization: Internet-USENET Gateway at Stanford University
Lines: 35


Question:

  How does Project Athena synchronize the clocks on its many
workstations and fileservers?

Answer #1:

(Jonathan explained about the nntp version of timed)

Answer #2:

Before nntp was installed, there were a few halfway measures used to
help keep from going completely insane:

1.  All servers ran a network time service that simply reported the
time on the local clock.

2.  All workstations, whenever rebooted, set their clocks from a
particular one of the time servers:  the one running on the Kerberos
server.  Users were told that if they saw an error message that
mentioned both tickets and clocks they should reboot their
workstation and try again.

3.  A program was written to compare the times reported by any two
servers and display the number of seconds of difference.  This program
was periodically exercised, comparing every server with the Kerberos
server, looking for server clocks that had drifted.  When a drifted
clock was found, that machine was rebooted.  

This set of measures certainly didn't solve the problem completely,
but it kept things more or less under control for three years, until
the version of timed that uses the nntp protocol became available.

					Jerry Saltzer