Scientific Data Systems

From Wikipedia, the free encyclopedia

Scientific Data Systems, or SDS, was an American computer company founded in September 1961 by Max Palevsky, a veteran of Packard Bell and Bendix, along with eleven other computer scientists. SDS was an early adopter of integrated circuits in computer design and the first to employ silicon transistors. The company concentrated on larger scientific workload focused machines and sold many machines to NASA during the Space Race. Most machines were both fast and relatively low priced. The company was sold to Xerox in 1969, but mismanagement and dwindling sales caused Xerox to close the division in 1975 at a loss of hundreds of millions of dollars. During the Xerox years the company was officially Xerox Data Systems, or XDS.

Throughout the majority of the 1960s the US computer market was dominated by Snow White (IBM) and the Seven Dwarves, NCR, Burroughs, Control Data Corporation, General Electric, Honeywell, RCA and UNIVAC. SDS entered this well developed market and soon carved out their own niche, a surprising development. Much of this success was due to the use of silicon-based transistors in their earliest designs, the 24-bit SDS 910 and SDS 920 which included a hardware (integer) multiplier. These are arguably the first commercial systems based on silicon, which offered much better performance for no real additional cost. Additionally the SDS machines shipped with a selection of software, notably a FORTRAN compiler, developed by Digitek, that made use of the systems' Programmed OPeratorS (POPS) and could compile, in 4K 24-bit words, programs in a single pass without the need for magnetic tape secondary storage. For scientific users writing small programs, this was a real boon and dramatically improved development turnaround time.

The 910 and 920 were supplanted by the SDS 9300, announced in June 1963. Among other changes, the 9300 included a floating point processor for higher performance. The performance increase was dramatic, the 910/920 needed 16 microseconds to add two 24-bit integers, the 9300 only 1.75, almost 10 times as fast. The 9300 also increased maximum memory from 16 kWords to 32 kWords. It was otherwise compatible with the earlier machines.

In December 1963 they announced the SDS 930, a major re-build of the 9xx line using ICs in the central processor. It was comparable to the 9300 in basic operations, but was generally slower overall due to the lack of the 9300's memory interlace capability and hardware floating point unit (although a hardware floating point "correlation and filtering unit" was available as an expensive option). The 930 cost less than 1/2 that of the original 9300, at about $105,000. Cut-down versions of the 920 also followed, including the 12-bit SDS 92, and the IC-based 925.

A project to develop a virtual memory system for time sharing use at the University of California, Berkeley resulted in the SDS 940, basically the 930 with additional hardware to provide dynamically paged memory and interruptible instructions. The 940 would go on to be a major part of Tymshare's network system growth in the 1960s. A 945 was announced in July 1968 as a modified 940 with less I/O but more compute power, but it is unclear whether this shipped. The 945 never shipped; I had the one-and-only 945 stainless steel bar. I think it was actually just like the 940 but with a smaller drum in place of the disc.

In December 1966 SDS shipped the entirely new Sigma series, starting with the 16-bit SDS Sigma 2 and the 32-bit SDS Sigma 7, both using common hardware internally. The rise of the 8-bit ASCII character standard was pushing all vendors to the 8-bit standard from their earlier 6-bit ones, and SDS was one of the first to enter this market. Various versions of the Sigma 7 followed, including the cut-down Sigma 5 and re-designed Sigma 6. The Xerox Sigma 9 was a major re-design with instruction lookahead and other advanced features, while the Sigma 8 and Sigma 9 mod 3 were low-end machines offered as a migration path for the Sigma 5.

Even with these successes, when Xerox bought the company in 1969 they sold only about 1% of the computers in the US, something Xerox never seemed to improve. When they were purchased, about 1,000 SDS machines of all types were in the market, and by the time the division closed in 1975 this had increased to only about 2,100. By this point the newer Xerox 550 and 560 models, extensively re-designed Sigmas, were about to come to market and were extensively back ordered. Most rights were sold to Honeywell in July, 1975 who produced Sigmas for a short period, and provided support into the 1980s.

Several manufacturers attempted to enter the Sigma 9 replacement market. The first successful design was the Telefile T-85, but it is not clear how many were sold. Other efforts, including the Modutest Model 9, Ilene Model 9000 and Real-time RCE-9 were designed, but it is not clear if they were ever produced past the prototype stage.

In 1979 Jack Mitchell and Henry Harold, former SDS engineers, along with some other ex-SDS people re-started the company along with some funding from Max Palevsky. They introduced a microprocessor based computer called the SDS 420 built on a 6502-based processor design with up to 256KB of memory and a proprietary OS along with the BASIC programming language. The 420 Series had little to do with scientific computing and more with word processing and business services. The second incarnation failed, and the new SDS went out of business in 1984.

[edit] SDS Software

The primary operating system for the 900 series was called Monarch. A Real-Time monitor was also available. In March 1982 Honeywell gave the remaining software for the 900 series to a group in Kansas City that offered to continue making copies for people still using the systems. Honeywell had stopped supporting the systems many years before this. In September 2006, this collection was donated to the Computer History Museum along with all of the program's original documentation, and copies of most of the SDS user's manuals. This is one of the largest collections of software to have survived from the 1960's intact. Unfortunately, the timesharing software for the 940 series was not present in the Honeywell LADS Library and does not appear to have survivied. Copies of the original system developed at UC Berkeley exist as file system backups. Most of the customers for 940 systems (in particular Tymshare) made extensive modifications to the 940 system software, and no copies of that version of the software are known to have survived.

A simulator for the Sigma series is known to exist, but the status of Sigma series software is unknown. Documentation on the web implies that early versions were not copyrighted, while later versions developed by Honeywell were.