THE EVOLUTION OF THE DECsystem-10 491

system for computer-aided instruction, and the BBN Tenex system all contributed concepts to the DECsystem-10 evolution in the 1960s.

In architecture, the Manchester Atlas [Bell, Newell, 1971: Ch. 23] was exemplary, not because it was a large machine that we would build, but because it illustrated a number of good design principles. Atlas was multiprogrammed with a well defined interface between the user and operating system, had a very large address space, and introduced the notion of extra codes to extend the functionality of its instruction set. Paging was a concept we just could not afford to implement without a fast, small memory. The IBM Channel concept was in use on their 7094; it was one we wanted to avoid since our minicomputers (e.g., PDP-l) were generally smaller than a single channel and could outperform the 7094 in terms of I/O concurrency and I/O programmability by a clean, simple interrupt mechanism.

The DEC product line in 1964 is summarized in Table 1. Sales totaled $11 million then, and it was felt that computers had to be offered in the $20,000 to $300,000 range. We were sensitive to the problems encountered by not having enough address bits, having watched DEC and IBM machines exceed their addressing capacities.

On the software side, most programmers at DEC had been large-machine (16 Kword to 32 Kword) users, although they had most recently programmed minicomputers where program size of 4 Kwords to 8 Kwords was the main constraint. There was not a good understanding of operating systems structure and design in either academia or industry. MIT's Multics project was just being formed and IBM's 360/TSS project did not start until 1965. Generally, there were no people who directly represented the users within the company, although all the designers were computer users. A number of users in the Cambridge (Mass.) community advised on the design (especially John McCarthy, Marvin Minsky, and Peter Sampson at the MIT Artificial Intelligence Laboratory).

Although there was little consensus that FORTRAN would be so important, it was clear that our machine would be used extensively to execute FORTRAN. The macroassemblers, basically unchanged even today, were used in various laboratories; our first one for the PDP-1 was done by MIT in 1961. We also felt that the list languages, especially LISP for symbolic processing, were important. There was virtually no interest in business data processing although we had all looked at COBOL.

We did not understand the concept of technology evolution very well, even though integrated circuits were both forecast and in development. Germanium transistors were available, and silicon transistors were just on the market. IBM was using machine wirewrap technology, while DEC back panels were hand-wired and soldered. The basic DEC logic circuits were saturating transistors as distinct from the more expensive current mode used by IBM in the 7094 and Stretch computers. Production core memories of 2 microseconds were beginning to appear, and their speed was improving. The PDP-l used a 5 microseconds core. Hence,