home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!munnari.oz.au!comp.vuw.ac.nz!zl2tnm!toyunix!don
- Newsgroups: comp.os.vms
- Subject: Re: Date incorrect after reboot
- Message-ID: <3890007@zl2tnm.gen.nz>
- From: don@zl2tnm.gen.nz (Don Stokes)
- Date: 8 Jan 93 15:53:20 GMT
- Sender: news@zl2tnm.gen.nz (GNEWS Version 2.0 news poster.)
- References: <1993Jan7.090247@mccall.com>
- Distribution: world
- Organization: The Wolery
- Lines: 93
-
- tp@mccall.com (Terry Poot) writes:
- > In article <1993Jan7.090347.14794@fys.ruu.nl>, hooft@fys.ruu.nl (Rob Hooft)
- > writes:
- > >Just make sure you have a cluster: good LAVC practice is to
- > >synchronize the clocks every once in a while: I think SET TIME/CLUSTER
- > >will also take care of the problem?
- >
- > I don't think so. SET TIME with no arguments copies the system time into the TOY
- > clock. SET TIME/CLUSTER synchronizes the cluster, but does NOT copy it into the
- > TOY clock on any system.
-
- Other way around. SET TIME without parameters gets the time from the TOY.
-
- The SET TIME sequence goes (roughly):
-
- If time specified:
- $SETTIME parameter
- else
- $SETTIME
- if /CLUSTER:
- For all cluster members who will talk to us:
- time = current time
- Instruct CSP (aka CLUSTER_SERVER) to do a
- $SETTIME time for us.
-
- $SETTIME goes like this:
- if time specified:
- t = new TODR value
- else
- t = current TODR
- if t > 1 year
- t = t - 1 year
- set TODR = t
- Write 01-JAN-<year> to SYS.EXE
- Set system time
-
-
- It's a good idea to do a SET TIME periodically, since the TOY tends to be
- far more accurate the system time; the former usually being based on a
- crystal oscillator, the latter on the who-knows-what in the CPU that bumps
- the interval timer. There are exceptions -- see below.
-
- If in a cluster, a SET TIME/CLUSTER (or its SYSMAN equivalent) should be
- done periodically to synchronise times across cluster members. Do it from
- whichever node you think has the most accurate TOY (aka TODR above) clock.
- Remember that the queue mangler (at least prior to V5.5) schedules on
- any node; what tends to happen is that the node with the fastest clock
- schedules jobs, regardless of which processor it runs on. The effect of
- this is that if you have self-resubmitting batch jobs that go:
-
- $ this = f$environment("PROCEDURE")
- $ SUBMIT/AFTER=TOMORROW 'this'
- $ [rest of job]
-
- your job ends up running slighty before tomorrow (because another node
- let it go), and you get two copies running TODAY.
-
- You should always kick such things off /AFTER="TOMORROW+0:05" to avoid
- clock skew, and periodically synchronise the clocks to keep the skew under
- five minutes.
-
- Tip to the wise: if you have Nautilus processors (85xx, 8700, 8800 88x0?),
- *don't* recalibrate from those. They're dreadfully inaccurate. I can
- personally vouch for 82x0/83x0s and MicroVAX IIs. (The former by experience,
- the latter by taking a squiz at the hardware....)
-
- > it by doing SET TIME/CLUSTER from a node that was right. A few days later we
- > lost power, and when the cluster came back up, that same node was again off by
- > an hour. I always try to remember now to use sysman to do a SET TIME on each
- > system after synchonizing the cluster time. One of these days, I'll even look up
- > the SYSMAN command to do that so I can do all from within SYSMAN. :-)
-
- In theory at least, the $SETTIMEs done by CSP on the remote nodes should
- be no different to the $SETTIME done on the local node.
-
- Was the offending node a Nautilus? I've seen lots of bugs on the damn
- things. The physical TODR is derived from the system time on the Pro
- consoles, and the support has often been flakey. It's a good idea to
- exit the console program and do a SHOW TIME on the console processor
- itself (and a SET TIME if necessary). I think the worst of the bugs have
- been fixed, but I've had to set the time manually on them before,
- regardless of where the VMS SET TIME was done from. Horrible things. I
- think the console software begins to work at version 10 or 11.....
-
-
- As to my earlier comment about SHUTDOWN.COM -- the implication (as I read
- it) was that the ever-lovin CSC were suggesting that a SET TIME be done
- before *intentional* shutdowns....
-
- --
- Don Stokes, ZL2TNM (DS555) don@zl2tnm.gen.nz (home)
- Network Manager, Computing Services Centre don@vuw.ac.nz (work)
- Victoria University of Wellington, New Zealand +64-4-495-5052
-