[HECnet] US DST

[Thomas DeBellis]

The recent post about multi-threaded NNTP for VMS (whaaay cool) jogged 
my memory about something I had been wondering about.  This morning, a 
large number of clocks got switched ahead an hour in the United States.  
Tops-20 did it 'relatively flawlessly' as it has done so since the early 
1980's.  Back then, a lot of systems didn't.  None of our IBM systems did.

A lot of that has changed nowadays with NTP clients; even if the base 
operating system doesn't support time change, an NTP client can address 
that.  So all my systems advanced appropriately, as did an old radio 
clock.  I'm not sure how *nix does it, but I don't remember Ultrix 
having the code on our 8650 (or 8700). Tops-20 will do the change 
whether or not a client exists as the code is in the monitor.

At the time (1980's), I don't believe that VMS had internal code to do 
the time change and NTP did not exist as such (some network time 
services existed, but I don't remember whether there was a generally 
available VMS TCP/IP package; we didn't have it).  I would assume that 
this is handled now.  I hesitate to ask about RSX and RSTS, but I would 
assume that the C compiler for RSX (with Johnny's TCP/IP interface) 
could support an NTP port.  Anyway, the question is: how does VMS handle 
the time change?

By 'relatively flawlessly', I mean Galaxy.  If you submit jobs with an 
/AFTER: switch, but use an elapsed time, you can get the wrong start 
time.  For example, in a daily batch job to do incremental saves, 
TODAY+2:00:00 means tomorrow morning at 2:00:00 AM, assuming the current 
job was started at 2:00:00.  However today, the job  started at 3:00:00 
AM.  No big deal.

However, this used to seriously mess up our operations, who started 
shift expecting requests for tape mounts.   I looked at the issue awhile 
back in the 80's, but realized that the problem wasn't in Quasar because 
it is only dealing with internal date format.  The issue had to be fixed 
by putting some logic into the Exec to differentiate between a specified 
time and a calculated time.  As it wasn't immediately straightforward, I 
never got to it.  Maybe one of these days...