In the early 90s I worked with a lot of MCA-based IBM PC's on token ring. (16). Consistency was key. Use the same model machine with the same devices do not deviate. We had a system based on OS/2 1.3 that had to run on Model 77 machines. Period. When it worked it was great but there was very little flexibility from a hardware perspective.
Very nice! Yea, I picked up my very first MCA machine earlier this year, never had one before, and literally had not configured one since 1997. I picked it up because it came with a nice Model M keyboard! But I have grown to love it, as quirky as it is. Maybe I'll drag it out and work on it a bit today.
I think you are possibly overstating the scalability issues at large enterprise installations. One of the benefits of the MCA architecture was not only the jumper less configuration, but the ability to poll the bus to locate devices and their configuration register settings. This meant that a well written driver could not only find its adapter in the system, but read its configuration off the adapter itself. This meant that as long as you had a level of standardisation (e.g. the fleet uses the same network card) once your O/S loaded the drivers would find the card and use the configuration on the card (not from a config file or command line arguments stored on the hard drive). All you needed to do to manage this hardware was have in your tool bag reference disks for each PS/2 model you had in your fleet and ensure that you had all the ADF files needed. In the extremely unlikely event that the machine needed configuration (CMOS batteries lasted at least 5 years so this would be an extremely rare event) all you needed to do was auto-configure the machine and you were done.
PCI learned from the PS/2 and added a "type" byte in the configuration data area that identifies the type of device the card is (in addition to 2 16-bit vendor ID and product ID values). That's why PCI devices can often detect what kind of device it is like "Fixed Disk Controller" or "Universal Serial Controller" on startup. But even then, the VID and PID are meaningless numbers. There is a database of PCI identifiers for this purpose online and operating systems like Linux build it into their source code so they can provide details. USB expanded on it by adding strings to the USB descriptor so operating systems could print out something human readable.
That brought back some memories, back in 1993 to 1997 I supported a range on ps/2 systems on token ring. The reference disks were a complete nightmare. I still have vivid memories talking to ibm tech support about a ibm 16/4 token ring card in an ibm ps/2 model 95 server running os/2warp and using the netBui and netbios protocol for a ibm lotus notes server and ibm said they couldn’t support it because of possible non ibm products
From a technical point the PS/2 architecture might have been somewhat great back in the day but the way of configuring devices feels rather annoying. Not even being allowed to boot your system after inserting a simple card seems sluggish and I cannot understand how something like that could've been made like this. Who thought, it was a good idea to have a floppy disk at hand for that? As you named it: in bigger companies that would've meant, you'd have a single disk for EVERY computer in the company which was running under that architecture. Even testing a card was impossible without that cursed reference disk. Who thought, that was a good idea? If I had to choose between a slower architecture with setting some jumpers (which I had no idea of back in the day) and the PS/2 way of doing it, I'd definitely have chosen the PC AT way.
At least in theory, their way of predefining the resources in the ADF files, and having the reference disk setup program trying to figure it's way in allocating conflict free resources, without the user needing to care, isn't a bad idea per se. It was just caried out very badly. My guess is, it was time and money driving the decision to go with floppy disks. If I were to draw a Dilbert, it would propably go like this: Dilbert: I propose to store the Reference Setup in ROM, and add a 64-128k flash to store the ADF files. Pointy-haired Boss: Nah, flash is too expensive. Why don't you store it to the hard drive? Dilbert: Because Wally wrote the BIOS to invalidate the CMOS configuration if it detects new hardware. It looses everything and can't boot of the hard drive any more. He would take 2 weeks to rewrite it. Pointy-haired Boss: Nah, that's too long. Go with floppies. Customer's will deal with it, and nobody got ever fired for buying IBM.
Back in the early 90s having MCA over ISA was actually nice. Until PCI and software configurable irq / io there was less pain. Sure it looks clunky now but I preferred it over ISA config at the time. The real limitation to MCA was IBMs greed. The extreme cost prevented a lot of cards being ported over or made one that did hard to afford. On my model 80 you could just hit F1 at an error and continue booting. Only that card that caused the error would be ignored (not sure if overlapping io spaces would be a problem )
In the early 90s I worked with a lot of MCA-based IBM PC's on token ring. (16). Consistency was key. Use the same model machine with the same devices do not deviate. We had a system based on OS/2 1.3 that had to run on Model 77 machines. Period. When it worked it was great but there was very little flexibility from a hardware perspective.
Very nice! Yea, I picked up my very first MCA machine earlier this year, never had one before, and literally had not configured one since 1997. I picked it up because it came with a nice Model M keyboard! But I have grown to love it, as quirky as it is. Maybe I'll drag it out and work on it a bit today.
@@RetroTechChris inspiration, huh? ;-)
@@THEPHINTAGECOLLECTOR haha sure thing!
I think you are possibly overstating the scalability issues at large enterprise installations. One of the benefits of the MCA architecture was not only the jumper less configuration, but the ability to poll the bus to locate devices and their configuration register settings. This meant that a well written driver could not only find its adapter in the system, but read its configuration off the adapter itself. This meant that as long as you had a level of standardisation (e.g. the fleet uses the same network card) once your O/S loaded the drivers would find the card and use the configuration on the card (not from a config file or command line arguments stored on the hard drive). All you needed to do to manage this hardware was have in your tool bag reference disks for each PS/2 model you had in your fleet and ensure that you had all the ADF files needed. In the extremely unlikely event that the machine needed configuration (CMOS batteries lasted at least 5 years so this would be an extremely rare event) all you needed to do was auto-configure the machine and you were done.
I have never seen a youtube video covering IBM's ABIOS. Cheers! 👍
PCI learned from the PS/2 and added a "type" byte in the configuration data area that identifies the type of device the card is (in addition to 2 16-bit vendor ID and product ID values). That's why PCI devices can often detect what kind of device it is like "Fixed Disk Controller" or "Universal Serial Controller" on startup. But even then, the VID and PID are meaningless numbers. There is a database of PCI identifiers for this purpose online and operating systems like Linux build it into their source code so they can provide details. USB expanded on it by adding strings to the USB descriptor so operating systems could print out something human readable.
That brought back some memories, back in 1993 to 1997 I supported a range on ps/2 systems on token ring. The reference disks were a complete nightmare. I still have vivid memories talking to ibm tech support about a ibm 16/4 token ring card in an ibm ps/2 model 95 server running os/2warp and using the netBui and netbios protocol for a ibm lotus notes server and ibm said they couldn’t support it because of possible non ibm products
This is a really good video, thank you for sharing
@@NiceCakeMix you‘re welcome!
Please try AIX and OS/2 on this P70! 😊
What. A. Nightmare. Im glad i missed this era.
This is something beyond plug and pray. More like plug and make a sacrifice to IBM.
As it was back in the day lol
From a technical point the PS/2 architecture might have been somewhat great back in the day but the way of configuring devices feels rather annoying. Not even being allowed to boot your system after inserting a simple card seems sluggish and I cannot understand how something like that could've been made like this. Who thought, it was a good idea to have a floppy disk at hand for that? As you named it: in bigger companies that would've meant, you'd have a single disk for EVERY computer in the company which was running under that architecture.
Even testing a card was impossible without that cursed reference disk.
Who thought, that was a good idea?
If I had to choose between a slower architecture with setting some jumpers (which I had no idea of back in the day) and the PS/2 way of doing it, I'd definitely have chosen the PC AT way.
At least in theory, their way of predefining the resources in the ADF files, and having the reference disk setup program trying to figure it's way in allocating conflict free resources, without the user needing to care, isn't a bad idea per se. It was just caried out very badly.
My guess is, it was time and money driving the decision to go with floppy disks.
If I were to draw a Dilbert, it would propably go like this:
Dilbert: I propose to store the Reference Setup in ROM, and add a 64-128k flash to store the ADF files.
Pointy-haired Boss: Nah, flash is too expensive. Why don't you store it to the hard drive?
Dilbert: Because Wally wrote the BIOS to invalidate the CMOS configuration if it detects new hardware. It looses everything and can't boot of the hard drive any more. He would take 2 weeks to rewrite it.
Pointy-haired Boss: Nah, that's too long. Go with floppies. Customer's will deal with it, and nobody got ever fired for buying IBM.
Back in the early 90s having MCA over ISA was actually nice. Until PCI and software configurable irq / io there was less pain. Sure it looks clunky now but I preferred it over ISA config at the time.
The real limitation to MCA was IBMs greed. The extreme cost prevented a lot of cards being ported over or made one that did hard to afford.
On my model 80 you could just hit F1 at an error and continue booting. Only that card that caused the error would be ignored (not sure if overlapping io spaces would be a problem )
no AIX fun then? : )
@@bamdadkhan eventually, once I get it up and running.
MCA is a bag of hurt
I hated MCA. I gone back to ISA until PCI came around for me.