[comp.unix.i386] ESIX installation story

kcollins@convex1.convex.com (Kirby L. Collins) (07/31/90)

Well, I finally got my shiny new toy up and running ESIX Rev. D, and
I thought a summary of my experience might be helpful for other neophytes.
(BTW, I have a fair amount of experience administering UNIX systems, but
all larger systems with BSD flavor OS's).

I did this a little bit backwards, in that I bought ESIX first, and
then tried to buy hardware to fit.  A local PC vendor (Lucky Computer)
put together the following system for me:

Micronics 386-i 33MHz motherboard, 4MB RAM, 32K cache
Phoenix BIOS 1.10
DTC 8-bit floppy/SCSI controller
5.25 and 3.5" floppy drives
Toshiba 234 105MB hard drive
Mono graphics & parallel combo card
ATI I/O card with 2 serial and one parallel port.
DOS 4.01

I also bought a Seagate ST296N 85MB hard drive and an Intel 2400B internal 
modem at a local discounter, intending to install these myself 
(more about this later).

My thinking was to go with SCSI disk from the start...since I see most
of the workstation vendors using SCSI as the standard periperal
interface, and I expected that most of the interesting peripherals (large
disk, high capacity tape, optical) will have SCSI interfaces (whether this
is in fact true remains to be seen).  I also wanted to have enough disk 
so my wife could use the machine as a DOS machine in her business.

Unfortunately, I ran into a problem from the start, in that the system
ran DOS with no problems, but would not even begin to boot ESIX.  When
I inserted the ESIX boot disk in my 5.25 drive and reset, the system 
would hang on the first access to the floppy.  I called the tech
support people at Everex and thought the likely culprit was my
(admittedly el cheapo) DTC floppy/SCSI controller.  I went back to 
Lucky and tried to boot ESIX on one of their IDE based systems.
Here I at least got to the "booting ESIX" message, but soon after
the system reset itself, and got into an infinite loop, where it 
would boot to a certain point, then reset and try again.  ESIX 
tech support told me they have seen this problem before on systems
with IDE controllers and the Phoenix BIOS.

Well, I looked through the postings on comp.unix.i386, and saw 
numerous references to the Adaptec 1542B controller.  ESIX tech 
support told me it ought to work, so I traded in the DTC SCSI
on an Adaptec 1542B rev. H.  At the same time I had Lucky install
the Seagate hard drive for me.

Unfortunately, the Seagate drive and the Toshiba did not coexist
very well.  First of all, the Adaptec SCSI controller seems to 
expect all the SCSI devices to have sequential IDs, starting with
0.  The first time it tries an ID and finds no device, it gives 
up.  Since we had the drives jumpered for 1 and 2, it took us a
while to figure this out  (is this normal?   the DTC controller
didn't seem to care how the IDs were arranged, although it took 
almost 30 seconds(!) to probe the SCSI bus and find them).  Second,
the Seagate drive was so slow spinning up and initializing that
the Adaptec would give up on it unless it was the first hard 
drive (SCSI ID 0).  This meant that the Seagate HAD to be the
boot drive.

When I got the system home and tried to install ESIX, things had
improved somewhat...I could get completely through the boot
process (it takes two floppies just to boot), partition the 
Seagate drive, but the system would now hang while copying
the ESIX core to the hard drive.  After some consultation with
ESIX tech support (and trying various combinations of DOS fdisk
and the ESIX version of fdisk), I gave up, returned the Seagate
drive, and went back to Lucky to get another Toshiba drive instead.

One of the things that disturbed me was that the SCSI interface
was remapping the cyl/sec/head data to reduce the number of 
cylinders an application sees (apparently this is important to
some DOS applications).  Worse, the DOS and ESIX fdisk programs
didn't seem to agree on how many cylinders there were.  DOS said
100 cyls, 32 heads, 64 secs, ESIX said 101/32/64.  At least with
Lucky I had a knowledgeable vendor who could talk intelligently
about how the drives were really laid out.  When I hammered the
REAL drive geometry (there is a point in the boot process where
ESIX asks if you want the reported geometry or override) the 
ESIX install process finally started working (the geometry is
actually something like 1013cyls, 12 heads, 17 secs...sorry I
don't have my notes here at work, so that may not be quite right).

A couple of interesting caveats about fdisk...ESIX tech support told
me that I HAD to use the ESIX version of fdisk to partition the
hard drive.  Unfortunately, ESIX fdisk doesn't know about DOS 
partitions larger than 32 MB (new with DOS 4.0).  I wanted the 
following arrangement:

drive 1:  20MB DOS, 85MB ESIX
drive 2:  60MB DOS, 45MB ESIX /mnt

To do this, I had to use the ESIX fdisk to create both partitions
on the first drive, and the ESIX partition on the second drive.  I
then had to use the DOS 4.01 fdisk to create the 60MB DOS partition
on the second drive.  So far this seems to work, although I have
not tried to access the 60MB DOS partition from within ESIX (I suspect
none of those utilities will work, and I'm not anxious to roach the
DOS partition).  To boot ESIX or DOS, I can use either version of
fdisk to set the active partition on the first drive, and DOS doesn't
seem to care which partition on the second drive is active, it still
finds the right one.

Once ESIX is installed on the first drive, the 101/32/64 versus
real geometry doesn't seem to be an issue...the reported geometry
works fine on the second drive, and the display and set active
commands work fine without having to set the real layout.
(It may be obvious that I don't really understand this geometry
remapping business...Is there an article anywhere in the PC rags
that explains it?)

Now for the next issue...once the basic system is installed, the
documentation Everex provides for configuration is atrocious.  For 
instance, they provide a whole chapter on configuring the lp
spooling mechanism, but I had to dig around in /etc/conf/cf.d to
discover that lp1 was enabled, but not lp0 (which is why the
printer controller I jumpered to be the first parallel port did
not work).

Worse, the installation notes describe a whole raft of separately
installable packages, with no description of what's in them or
whether I need them or not.  I loaded the Security Administration
Package Version 2.0...and found that passwd restrictions had 
suddenly been enabled (scary...what else changed that I don't
know about?).

That aside...the support people for Everex really are great (as
were the people at Lucky, if any one in Dallas is interested).