08/12/04

Kidder, Chapter 2

pages 28 - 48

1. DEC brought out a 32-bit minicomputer and Data General did not have a competitive product - both the NOVA and the newer Eclipse were 16-bit machines. Mainframes were 32-bit machines.
2. Two reasons why 32-bit machines are desirable in comparison with machines with fewer bits:
   • Machines that move data around in the larger 32-bit chunks tend to be faster than those that move data in smaller chunks
   • Machines that use a 32-bit word length can address many more memory locations than those with a 16-bit word length. The situation is analogous to that for telephone numbers. A 7-digit phone number is good enough for most cities but a country requires a 10-digit number (including area code). Reaching a telephone outside the country requires a 12-digit code (including a 2-digit country code). As customer applications become more complex, more memory is required and a 32-bit machine is needed to address the large number of memory locations.
3. Data General had correctly foreseen the need for a 32-bit machine and had authorized a group in Research Triangle Park in North Carolina to design such a machine.
4. Tom West and his group at Westborough desperately wanted to work on a 32-bit machine. Many of them, indeed, were bitter that they had been assigned the relatively prosaic task of making small improvements in the old 16-bit Eclipse.
5. West's group proposed to develop a 32-bit machine that would provide upward software compatibility with the 16 bit Eclipse: that is, the 32-bit machine would run the software that users already owned, as well as newer, faster 32-bit software.
6. The idea of upward software compatibility had been pioneered successfully by IBM when it "bet the company" by going to the huge expense of developing an entire new line of machines (the historic IBM 360 series), all at the same time, in which the software for the smaller machines in the line would run on all larger machines in the line. The result was almost a death grip on the customers, who were much more likely to buy a new larger machine from IBM than from a different company because they could continue to use their old software on a larger IBM machine even as they added new, more powerful, software. If they bought a larger machine from a different company, on the other hand, their old software was useless unless they went to the expense of maintaining two different computer systems. Just as important, the cost to IBM of writing the software decreased with this approach because the software, once written, ran on several different machines and did not need to be rewritten for each one.
7. Tom West's group first proposed to achieve software compatibility by using a "mode bit" to switch between 16- and 32-bit operation. Essentially, the computer would be two computers in the same case - one a 16-bit machine and the other a 32-bit machine. The mode bit would switch machines depending on what software the user wanted to run.
8. De Castro did not accept this proposal. He indicated he wanted no mode bit, although he did not explain why. West concluded that de Castro thought that the mode bit would lead to inefficient complexity that would increase costs - and Data General was committed to producing fast machines at the lowest possible cost.
9. West and the people at Data General knew that a key factor in keeping cost down was "chip count." They also knew that using larger (and hence fewer) printed circuit boards tended to reduce the chip count. In fact, this approach gave NOVA, Data General's first machine, a key advantage over DEC's PDP-8 machine. West was so concerned about the chip count, and possible reduced complexity, of DEC's new VAX 11/780 32-bit machine that he made a special trip to see, and take apart, one that a friend had bought. He felt relieved that it appeared unnecessarily complex. He felt his team had a chance to make one that was just as capable, but was simpler and cost less.
10. West and his team formulated a new strategy. They proposed a 32-bit Eclipse. It would run 16-bit Eclipse software but would be a true 32-bit machine. To de Castro and Data General, he sold it as merely a logical upgrade of the Eclipse line, for which he and his team were responsible. De Castro did not stop the project, nor did he endorse it enthusiastically. He said the project should be completed in one year. Presumably, he wanted to limit the investment in the project and, in addition, felt that if it were not ready within that time, the North Carolina machine would be ready. Perhaps he viewed the project as an inexpensive way to toss a bone to the Westborough engineers and make them feel better.
11. To his team, however, Tom West sold the new machine as an exciting opportunity to build the fastest and cheapest machine available; that is, the opportunity to make a jump in the state-of-the-art.
12. At this time, it is important to note that West, nor anyone on his team, knew in any detail how they would go about building such a machine.