[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference noted::dtss

Title:DTSS_NOTE
Moderator:TUXEDO::BARYIAMES
Created:Mon Jul 31 1989
Last Modified:Thu May 29 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:624
Total number of notes:2671

615.0. "clerks and epochs?" by CHEFS::DAVIDSONS () Thu Jan 16 1997 17:51

T.RTitleUserPersonal
Name
DateLines
615.1Yes, you will need all systems to operate as servers...STEVMS::PETTENGILLmulpTue Jan 28 1997 02:1120
A clerk does not pay attention to the epoch; it just tries to find the
best time from all servers it can see.

This design choice was made so that there was nothing to configure on the clerk.
I guess this was too VMS a view of the world which assumed that system designers
were confident enough in their system designs so that they would make sure that
its time representation would be valid for the next thousand operating system
generations.

I do question using any time service at all if you're doing Y2K work.  I would
expect that you will be setting the time back and forth around the year 2000
in order to observe the problems of transitioning to year 2000 as well as
operation after the year 2000.  Neither DECdts nor NTP is really prepared to
deal with time not being monotonically increasing.  They do deal with systems
with faulty time, but that is significantly different than time being faulty.
If you change the time of a group of systems backward, there is no assurance
that the group of systems as a whole will behave correctly.  For example, I can
think of several components of these systems that will interpret a timestamp
in the future as a sign of corruption of persistent storage (ie., file systems
and databases).
615.2thanksCHEFS::DAVIDSONSWed Jan 29 1997 17:116
    OK, we'll configure all systems as server...
    
    BTW. We're not doing Y2K work rather record and replay.
    
    Thanks,
    	Stuart.