Fifteen Classic PC Design Mistakes

Harry writes "Once upon a time, it wasn't a given that PC owners should be able to format their own floppy disks. Or that ports should be standard, not proprietary. Or that it was a lousy idea to hardwire a PC's AC adapter, or to put the power supply in the printer so that a printer failure rendered the PC unusable, too. Over at Technologizer, Benj Edwards has taken a look at some of the worst design decisions from personal computing's early years — including ones involving famous flops such as the PCJr, obscure failures such as Mattel's Aquarius, and machines that succeeded despite flaws, like the first Mac. In most instances — but not all — their bad decisions taught the rest of the industry not to make the same errors again."

Comments

The 15 problems

[houghi]

Problem #1: No Power Supply Fan
Problem #2: Limited Apple II Compatibility
Problem #3: No Way to Format Disks
Problem #4: EM Pulse Erases Tapes
Problem #5: Printer Required
Problem #6: Rubber Keyboard
Problem #7: Non-Detachable AC Adapter
Problem #8: Miserable Keyboard
Problem #10: Sidecar Expansion
Problem #11: No User Expandability
Problem #12: Slow BASIC
Problem #13: Sidecar Expansion
Problem #14: Bulky Expansion Modules
Problem #15: Unreliable Proprietary Disk Drives

Re:Worst Mistake That Still Needs Fixing

[Anonymous Coward]

Actually, DisplayPort isn't proprietary, it's the successor to DVI. Mini-DisplayPort is part of the VESA specification and is entirely royalty-free.

Re:#1 failure...

[pz]

Why? What other processor(s) should have been used, and what would have been the benefits? No, not trolling. Just interested in what you said and would like more information.

The fundamental problem with Intel's instruction set architecture for the 8088/8086 line was that it was complex and intricate. To perform some instructions, the arguments had to be in very specific registers. Every register was, in some way or another, special purpose. The contemporary Motorola architecture, based on the 6800 and extended into the 68000 line, was completely the opposite: every register was, more-or-less, general purpose.

Writing a compiler for the Intel architecture is an exercise in masochism. Writing one for the Motorola architecture is one of simplicity and elegance. The Motorola instruction set documentation of the era was simple, clean, and definitive: it molded the way instruction sets were documented for generations afterward. The Intel documentation was difficult to understand at best.

One of the stark differences in the two instruction sets was the difference in instruction length variability. Intel instructions could be almost arbitrarily long. Motorola instructions were one or two bytes, with the one byte instructions being the ones most frequently used (inspired brilliance, that was). Also, for very related reasons, the number of cycles to execute an instruction was highly variable for Intel architectures, and more-or-less fixed for Motorola architectures.

I wrote assembly code for both architectures, back in the day. I hated, hated, hated writing for Intel chips, and breathed a sigh of relief whenever writing for Motorola chips. The inherent beauty in the Motorola instruction set created a certain kind of transparency making it possible --- seriously --- to see programmer intent when reading assembly code. With Intel chips, that was just not possible. With Motorola chips, you could reverse engineer code pretty easily; with Intel chips, it was painful.

The world would be a better place if IBM had selected Motorola.

Real mistakes

[Animats]

Those are mistakes an end user would see. Here are some deeper mistakes from an engineerings standpoint.

• Bus-type peripheral architecture. The IBM PC was a spinoff of the IBM Displaywriter, a dedicated word processor with no expandability. It inherited some design decisions from the Displaywriter that were reasonable for a word processor, but terrible for an expandable machine. Most notably, the IBM PC had the peripherals on the memory bus. That meant any DMA had to be on the I/O card, and thus any card could blither all over memory. Peripherals were thus trusted devices, and, in turn, drivers had to be trusted. IBM knew the right answer - channels, as on mainframes, and in the PS/2, they used a "microchannel" architecture. But it was too late - the industry had already standardized on "ISA cards". This is the fundamental reason cause of most operating system crashes - the I/O architecture gives drivers too much power.
• The Motorola MMU debacle.The Motorola 68000 first appeared in 1978, and it was a very good machine. Almost. There was a flaw. Instruction backout didn't quite work, and thus a paged MMU couldn't be added. So Motorola didn't ship an MMU with the 68000. The early UNIX workstations all used the 68000, and painful hacks were used to kludge together some kind of MMU to make it work. Apollo used two CPUs, one for the OS and one for the user, only one running at a time, to get around this. The Apple Lisa used one CPU with an Apple MMU built from many parts, and the compiler avoided generating any instructions with incrementation so that backout would work. Motorola came out with the M68010 in 1982, which fixed the bugs, but there was still no MMU. When Motorola finally shipped the 68451 MMU, it was a segmented MMU, and worse, slowed down the machine by one clock cycle per memory access. If Motorola had gotten it right by 1979 or so, the whole history of personal computing might have been Motorola-based using protected mode-UNIX.
• The Intel 286 CPU. Not enough memory management for a protected mode OS, too much segmentation machinery for an unprotected OS. That powered the IBM PC/AT and a whole generation of machines with the addressing system from hell. It could run a version of UNIX, but no process could exceed 64K in protected mode, although you could put a few megabytes on the machine.
• Baseband Ethernet. Coax-based Ethernet had some serious electrical problems. The thing really was unbalanced baseband, so you couldn't use capacitive coupling. The coax shield could only be grounded at one point, or you'd get ground loops. That created an electrical safety issue with the outside of coax connectors, and running coax between buildings was iffy. It was just bad electrical design. 10baseT, which is balanced, was far better from an electronics standpoint.