[598] in Kerberos
Re: service instance names
daemon@TELECOM.MIT.EDU (Jeffrey I. Schiller)
Thu Jan 12 23:02:11 1989
From: Jeffrey I. Schiller <jis@ATHENA.MIT.EDU>
To: jtkohl@ATHENA.MIT.EDU
Cc: kerberos@ATHENA.MIT.EDU
In-Reply-To: John T Kohl's message of Wed, 11 Jan 89 22:35:25 EST <8901120335.AA09492@LYCUS.MIT.EDU>
Experience at Athena however shows that the last holdouts of
old software tend to be services! Example: E40-PO has the most ancient
software on it that I know of. The file (NFS) servers are only now
being updated to 6.0 style software. SUOMI (the primary hesiod server)
is still running 5.4A.
Now an obvious question is "Why is this?" and the answer is twofold:
1) Servers have more state information, and therefore the nuke the
harddisk approach to software updates (which is more or less the way
workstations are handled) isn't appropriate which implies that
updating them is a more manual process. Also shutting down a server to
do the update involves inconveniencing a significant number of people.
2) Critical servers are the most conservatively managed for the
consequences of failure are much worse then on a workstation.
In any event for these two reasons there are always out-of-date server
systems, therefore any upgrade of kerberos software has to be upward
compatible in terms of supporting old servers as well as old clients.
-Jeff
