NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/os/vms/20576

Internet Message Format | 1993-01-08 | 3.6 KB

Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!ohstpy!bclcl1.im.battelle.org!hansher
Newsgroups: comp.os.vms
Subject: Re: Date incorrect after reboot
Message-ID: <[e-mail]>
From: [e-mail] (Kevin Hansher, Battelle Computing Center)
Date: 8 Jan 93 08:15:04 EST
References: <[e-mail]> <[e-mail]> <[e-mail]>
Organization: Battelle, Columbus, OH
Lines: 58

In article <[e-mail]>, [e-mail] (Ehud Gavron 602-570-2000 x. 2546) writes:
> In article <[e-mail]>, [e-mail] (Stephen Tihor) writes...
> #(a) not everyone shutsdown their system betweeen jan1 and april 1.
> #(b) SHUTDOWN only does a set time on the node being shutdown, in a cluster
> #you want to do the set time on all nodes within the cluster before any 1
> #is shutdown.
>
>     B is irrelevant if these nodes use a common system disk and
>     system image (SYS.EXE) which is where the data is stored.
>
>     Ehud
>
> --
> Ehud Gavron        (EG76)
> [e-mail]
> ..  If I had heart failure right now, I couldn't be a more fortunate man!!

If I am really interpreting correctly what is said here (i.e. that (b) is
not needed to be done in a cluster with a common system disk and system
image SYS.EXE), then I believe that information to not be true.  I believe
this is exactly what causes one of the many time problems that are reported
at various sites.  It has to do with the Time-of-Day Processor Register (TODR)
not being reset on every other node using a common system image.  The TODR
cannot contain the entire system time, so it is defined to contain a value
relative to a base time of January 1 of the current year (which is stored in
SYS.EXE).  When a SET TIME is issued, VMS writes the current time to the system
image file and resets the TODR as an offset from that base year.  If the SET
TIME was only issued on one node after the start of a NEW year, let's say as a
normal shutdown, then later if the whole cluster for some reason would crash
(power failure???) and a node with a large offset in the TODR boots FIRST,
then the other nodes booting afterwards will have their current time set one
year into the future.

The other case that can happen is to shutdown and reboot a node after the
start of a new year (SYS.EXE updated).  Everything looks ok.  Now another node
crashes without a SET TIME being done.  Node reboots, and time will be sync'd
up due to the cluster state transistion coordinator node setting time to be
the same as the rest of the cluster.  SHOW TIME looks good.  But check out a
SHOW SYSTEM or try to get the boottime of the crashed node.  Ever wonder why
that "wierd" boottime value appeared on the SHOW SYSTEM display?

Again the current work around for all of this is to make sure that you issue
a SET TIME on !!!ALL!!! nodes after the start of a new year to reset the
base time and the TODR's on ALL nodes.

I know I for one in my past have seen the second problem and wondered why in
the world???  And obviously a reboot seemed to clear it up.

Hopefully, this has been clear.  I must confess that during the writing of
this, I looked up an excellent DSNlink article that does a much better job
at explaining this then I could ever do.  Thanks guys!

--
------------------------------------------------------------------------------
Kevin Hansher            Symposium Session Quality Coordinator -
Battelle                U.S. DECUS VMS Systems SIG
505 King Avenue                    Internet:    [e-mail]
Columbus,  OH  43201        Voice: 614-424-4008  Fax:  [phone]