rickert@CS.NIU.EDU (Neil Rickert) (06/08/90)
Well we finally received the Umax 4.3 software about two weeks ago. I eagerly
opened the box, and pulled out the INSTALLATION NOTES so that I could begin
my planning. I particularly wanted to do some preparation before partitioning
the disk.
Unfortunately there was no such information. All the INSTALLATION NOTES said
was to run a script which would do it all automatically.
Thanks a whole bunch, Encore.
Well not to be daunted, I did the best I could, and used there installation
script. What an incredible pain.
I had to do some partitioning by hand, after running the script, just to
have big enough user partitions. But, as I feared, the non-selectable
choices of the install script were all wrong for me. It means that some
time in the new few weeks I will need to repartition the disk, and again
restore all the file systems.
The trouble didn't end with the partitioning problems. (I had anticipated
them at least).
Here is a catalog of problems to watch out for.
1. When the system first boots, the 'sendmail' daemon is started.
Fortunately I anticipated this, and immediately killed it, and
renamed 'sendmail' so it would happen again. I can imagine the
next time a neighbouring machine runs its queue and all the mail
starts coming in to a system with only 'root' in the password file,
so all mail is rejected.
2. The system started 'inetd', which means that anyone could probably
login while the non-password protected 'root' is the only user.
Fortunately we did not have any problems.
3. After following all the instructions to recover old password files,
group files, etc, I naturally tried rebooting the system. With
great amazement I discover that it boots from the 'bkuproot' partition
instead of the 'root' partition. (Of course the 'bkuproot' partition
still has only 'root' in the passwd file.)
4. When, because of the preceding problem, you boot from 'bkuproot' instead
of from 'root', the 'fstab' file says that '/dev/root' is mounted on '/'
whereas actually '/dev/bkuproot' is mounted. I don't know whether this
is why 'fsck' kept complaining about the root partition.
5. Trying to correct the boot problem, I tried running 'devconfig'. This
software turns out to be brain dead. You basically can't run it (in
interactive mode) from a printing terminal. It generates a line feed
after every input character.
Worse still, 'devconfig' can get you hopelessly stuck. When creating
a new device definition for a partition you built by hand, if you
accidently accept the default '0' for the slot number, it doesn't
complain till you enter a partition number. Then it gets into an
endless loop of demanding a correct partition number, and rejecting
everthing you enter. You can't get out with ^C or ^Z. You have to
login to another terminal and kill the job to escape from this disaster.
If you shut down the network daemons because you didn't want any logins
till you had finished configuring, and the console was all you had, this
meant hitting ^P, and rebooting the system.
Basically I had to do raw boots only until I had the system running
well enough that I could login at a video terminal and run devconfig
to fix up the bad DCT entries.
All this would not have happened if I had been able to install by hand
instead of running that automatic script.
Well good luck to anyone who hasn't yet installed 4.3. I hope this helps
you prepare.
--------------
And thanks a whole bunch, Encore. Next time can you just give us the
information we need to do the installation by hand if we want to.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
Neil W. Rickert, Computer Sci Dept, Northern Illinois U., DeKalb IL 60115
InterNet, unix: rickert@cs.niu.edu Bitnet, VM: T90NWR1@NIUCS
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=ran@oz.crl.dec.com (Randy Osborne) (06/09/90)
In article <9006081439.AA01527@cs.niu.edu>, rickert@CS.NIU.EDU (Neil Rickert) writes: ... lots about problems installing UMAX 4.3 We had the same sort of problems here installing UMAX 4.3. |> I had to do some partitioning by hand, after running the script, just to |> have big enough user partitions. But, as I feared, the non-selectable We had to do this anyway since we wanted to increase the size of the paging partitions. We didn't want to gamble that Encore might have actually fixed a kernel bug in UMAX 4.3 that existed in 4.2 (and which seemed to also exist in a beta release of UMAX 4.3 at MIT). The bug occurred when the system ran out of swap space. Certain processes would hang in the kernel and because they were in the kernel, they were unkillable and thus we could not get them to release their storage. Except by booting the machine. Anybody else ever had problems like this? With 64Mb of swap space I was able to cause the problem quite consistently by starting up a Lisp with a 30MB heap. This problem seems to be behind us now that we have 256MB of swap space. Others may also want to consider increasing the swap space (unless you can get some definitive info from Encore that the problem has been fixed - we couldn't). |> well enough that I could login at a video terminal and run devconfig |> to fix up the bad DCT entries. I found that devconfig was even brain damaged running interactively at a video terminal. It wouldn't let me store changes into the DCT because of a detected "error" in boot order it found in my changes (which were correct). I had to edit the Umax.config file directly. The cause of all the mess with booting off the bkuproot device seems to be because the installation script makes most of the non-terminal devices in /dev bootable, and leaves bkuproot as the first bootable device in the DCT (which is searched linearly for a boot device at boot time). I told Judy Leach at Encore about this mess, but I guess my advice for others would be to just edit the Umax.config file directly after booting off bkuproot and then reboot. Oh yes, there's no documentation for the Umax.config file format and Judy Leach at Encore couldn't find any either - but it's all in ASCII and you'll figure it out as I did. This seems to be what Encore customers have to do. Randy Osborne