[net.decus] 4.2 Upgrade procedure

nrh@lzwi.UUCP (N.R.HASLOCK) (10/02/85)

I am still in the process of performing an upgrade to 4.2

I have had a couple of new problems that you might like to hear
about.

First and foremost the processor is a member of a cluster using a
shared system disk so the upgrade was done on a copy of it on
another drive.

VMS gets most upset if you try to mount two volumes with the same
label in a cluster, and if it is the volume you are trying to boot
from you get an incomplete boot and an assortment of
incomprehensible numbers on the console. I've seen it before and I
can recognise and admit to the mistake but I do not like the result.
( I said it was a copy of the system disk, didn't I! ).

The procedures tell you to set some SYSGEN parameters related to
clusters, but they don't tell you that the install procedure will
exit if you don't set them the way they tell you to.

They tell you to turn off clustering for the processor that you are
going to use to make the upgrade. This is reasonable if the system
disk is local to the processor. If it is not local, I believe that
clustering should be enabled, so that if it is necessary to fix the
disk, it can be done safely by some other cluster member. ( You'll
never need to do that says DEC. Ha say I!! ).

The upgrade ran happily. ( N.B. Having read the earlier warnings, I
had not set P1 to MINIMUM which is why it happily read my
distribution tape from $6$mua0: ).

Then I tried to apply the mandatory update.

Does the message
 "This kit requires 2000 additional blocks after installation."
imply to you that it needs 2000 additional blocks before it will
install. Well, even if DEC thinks it does, I spent many minutes
trying to find out why it failed before this interpretation occured
to me.

The next question is why? The answer is that autogen, in its
infinite wisdom, had decided that I had to have large page and swap
files on my system disk, and so it had created them sometime during
the upgrade, leaving 1800 blocks free. I, because I have learnt
better, keep minimal page files for each of 5 roots on my system
disk and install large page and swap files on support disks during
my site dependant startup. AUTOGEN wanted at least 104400 blocks for
a page file and 78800 blocks for a swap file, i.e. 183200 blocks per
root plus 66000 blocks for a sysdump.dmp file equals 249200 per
root. This requirement times five is 1246000 blocks which is a bit
more than I can squeeze onto an RA81. This will be SPR'd real soon.
( If the dump file looks big, we allow the possibility of booting an
8600 from any root. )

I fixed that problem and continued. I was using a tape distribution
of the mandatory update kit and, not following the instructions,
told it to install products *. It access violated and then hit an
operand reserved to DIGITAL while dumping the Sig args and stack to
the console. It did this while trying to update BASRTL. I still
haven't uncovered why, but my partner in crime suggested that it
might have something to with the installed or not state of BASRTL.

The problem with VMSINSTAL is that if you tell it products * it will
loop asking for products, forever, until you do something drastic
and undocumented ( the help file does not tell you how to escape )
like hit control_y. This might only happen if one or more kits fail
to install correctly.

I hope you have better success.
-- 
--
{ihnp4|vax135|allegra}!lznv!nrh
	Nigel		The Mad Englishman or
			The Madly Maundering Mumbler in the Wildernesses

Everything you have read here is a figment of your imagination.
Noone else in the universe currently subscribes to these opinions.

"Its the rope, you know. You can't get it, you know."

tihor@acf4.UUCP (Stephen Tihor) (10/07/85)

You can override how large (or if) AUTOGEN wants to make page, swap and dump 
files by using the parameters PAGEFILE, SWAPFILE, and DUMPFILE (blocks).
Setting a parameter to zero tells autogen to not make or modify the file if 
missing.  (We have been running with DUMPFILE=0 and SAVEDUMP=1 for a while).

nrh@lzwi.UUCP (N.R.HASLOCK) (10/10/85)

In article <5870002@acf4.UUCP>, tihor@acf4.UUCP (Stephen Tihor) writes:
> You can override how large (or if) AUTOGEN wants to make page, swap and dump 
> files by using the parameters PAGEFILE, SWAPFILE, and DUMPFILE (blocks).
> Setting a parameter to zero tells autogen to not make or modify the file if 
> missing.  (We have been running with DUMPFILE=0 and SAVEDUMP=1 for a while).

That is wonderful, but how do you persuade the upgrade procedure to
define those parameters that way?
-- 
--
{ihnp4|vax135|allegra}!lznv!nrh
	Nigel		The Mad Englishman or
			The Madly Maundering Mumbler in the Wildernesses

Everything you have read here is a figment of your imagination.
Noone else in the universe currently subscribes to these opinions.

"Its the rope, you know. You can't get it, you know."