Bull and computer innovation.
In that page, we will go through a non-structured list of computer innovations and we will identify the reaction of Bull (and of its associated companies, particularly GE and Honeywell) engineers have reacted about or influenced them.
The original technology used in Bull's card machines was electro-mechanical relays. In the 1950's Bull started to adopt electronic vacuum tubes for calculator functions and solid state diodes for logic functions.
Magnetic core logic
SEA pioneered that technology with CAB500 in late 1950s. That technology was obsoleted by transistors for speed and by the emergence of integrated circuits for cost. It could be noted that NEC used "parametrons" in an early computer, but, there also there was no follow-on.
Bull introduced the germanium diodes for its logic circuitry (in conjunction with electromechanical relays) quite early in the industry.
Transistors were initially (in the early 1950s) not reliable enough to replace electronic tubes for sizeable computers. Their performances did neither obsolete tubes. So, the first computer for which Bull relied on transistors was the Gamma 60. For that machine and for the most part of the 1960s, computer manufacturers had to design all the logic circuits from the elementary gates to the more complex multiplier using their own knowledge of the performances of elementary transistors (both their specifications and the actual output of the plant production). That lead to a sizeable team of electronics engineers who had to be reused for other tasks when integrated circuits became available.
Integrated Circuits were introduced by Fairchild and Texas Instruments in the early 1960s. They became widely available around 1965. As both Bull-GE and CII rely at that time on their American affiliates, it was only around 1970 that their own design was converted to ICs. at that time the TTL technology from Texas was dominating the market that was essentially made of the computer industry.
Bipolar (ECL) technology from Siemens was planned to be used in some aborted projects at CII and Bull.
General Electric contemplated for a while to return to the semiconductor market by betting on the CML version of bipolar technology. However, CML did not get a market inside the Bull group until the 1980. Honeywell has bought the GE Colorado Springs laboratory in 1972 (?) and made a small production of CML. The technology was also acquired by NEC that used it successfully in their ACOS product line.
From the beginning of the 1970s, it was obvious that MSI functional modules will be eventually replaced by very large scale integrated circuits that will require a complete change in the methodology of building a computer, replacing the workbench by a virtual one in a computer. However, the size of Bull's main frame computers, starting around 100,000 gates was too large for contemplating a single chip processor before the late-1980s. So, an intermediary step was taken in the 1980s to design processor single boards equipped with a handful of CMOS chips. Bull made its implementation of CMOS versions of DPS-7 and DPS-8 at the same time or earlier than its major competitors.
Although Honeywell and Bull designers have been briefed by Intel, National Semiconductors and Motorola since the appearance of 4004, it was not before the early 1980s that use of those microprocessors took a share in mainframe design.
The AMD bit-sliced 2900 processor was used in a Bull disk controller in the late 1970s. But studies to use several of them for a proprietary processor failed
R2E, with the Micral 5, was the first in the world to offer a commercially available computer system powered by a microprocessor (Intel 8008). The Motorola 68000 family was adopted as the engine for SEMS UNIX system and took. It became also a general purpose engine for the 1985 generation of peripheral controllers for DPS-7 and later for network processors.
As the Motorola 68000 line was essentially discontinued in 1990, Bull (courted and wac courted by ) the companies trying to impose their RISC architecture and finally set for the PowerPC of IBM and Motorola, before leaning for Intel in the late 90s.
The concept of designing a processor with several instructions being in execution inside a single execution unit was relatively adopted by Bull that was not been addressing the intensively computing market segment. The ADP project in the mid-70s in Phoenix, that gave birth eventually to DPS-88, was the first pipelined design in the group, while the Auriga design circa 1986 was the first true pipelined DPS-7.
After the first use of cache in the 360/85, Honeywell decided to add it on Level 66, but the flushing had to be done by software, due to the reuse of the same multi-processor synchronization mechanism as the no-cache version.
The P7G DPS-7 used the first "store-through" cache modeled by Amdahl. Complex mechanisms were implemented to recover from a processor failure in a SMP system.
Magnetic Core Memory
The first DRAM ever produced by Intel was to be used in the central memory of ACS, the ill-fated Honeywell new product line, before the GE merger. Level 64 used for second generation Intel 4K-bit DRAM instead of core memory, as did level 62. Level 66 followed soon. Two generations of DRAM (e.g. 4Kb and 16Kb) were generally successively offered on a mainframe processor, the memory controller was designed for that.
Bull and Honeywell did develop its own boards and mounted the chips on different boards, initially using the same type of board as for logic. When processors came in one board , memory was either on the processor board or in separate boards.
SRAM was used initially for cache memory, and for control store in DPS-7 and later. On NEC most powerful machines (Zeus) the main memory was implemented in SRAM.
While Bob Bemer at GE was one of the pre-eminent apostle of ASCII, the first system using ASCII was Multics' 645 on a 9-bits byte. ASCII was later used in Large System first in conjunction with BCD and was generalized on GCOS8. While also used on Level-61, and terminals ASCII waited the generalization of networks and open systems to erode the EBCDIC postion.
Data compatibility with IBM S/360 was recognized as necessary for the business market in the second half of the 1960s. Computers designed after that date for the business market (GCOS 62 and GCOS64) used EBCDIC for that reason.
Jacques Stern, Bull's CEO in the early 1980s, and some Bul's scientists has been convinced that RISC computers would revolution the computer industry as much and as definitively as S/360 had obsoleted the 1950s vintage computers. However, the hardware designers were busy to reduce the cost of their proprietary GCOS systems by adopting the VLSI and MOS technology and were not enthusiasts about a revolutionary change in the architecture, even if HP and IBM were rumored to be interested. Product planners as well did not see RISC computers to be able to exploit their market momentarily.
For Jacques Stern, RISC computers would take the essential part of the UNIX open systems market (as they eventually did). he decided not to invest internally, but to outsource the design, initially from a small Siilicon Valley company Ridge Computers. It was expected that Ridge investments in architecture and compilers would be eventually benefit fro Bull's CMOS technology design and operating system experience.
Transferring to the compiler the responsibility of ordering the instructions is a characteristic of the still recent IA-64 architecture. A remote predecessor of that architecture could be traced up to the Gamma 60 that rely on the programmer the synchronisation of arithmetic, logic and other operations. Unhappily, compilers able to order object code were not yet available!
Disc cache buffer
Automated Tapes library
Data Cell drives
In the 1960s, the industry was in quest of very large random access memories to handle files such as insurance and income tax. Discs were too small and too expensive, tapes require batch processing. IBM and RCA produced magnetic strips devices. Bull naturally adopted the RCA device and connected it to the GE-400 (naming it Bullrac). Bullrac was somewhat more reliable as the IBM 2321, but the complexity of the mechanics caused this type of device before the 1970s.
A revival of the automatic loading of magnetic media happened in the 1980s with IBM tape cartridge library, followed by Masstor and StorageTek devices. Bull connected and successfully market those devices to DPS-8 and to DPS-7.
Paper tape readers
Paper tape punch
Diskette (floppy discs)
CRT display terminals
Dynamic partitioning of a SMP system
Real time processing
General Electric, as a major EDP user in the 1950s, was one of the first company to think seriously about the organization of files stored on disks. Charlie Bachmann designed IDS (integrated Data Store) in the early 1960s. IDS provided a data base organization that seemed to address all the requirements of batch as well as direct access processing, thanks to a mechanism of chained records, allowing hierarchical of structures as well. The mechanism was then more expensive that indexed-sequential offered at the time by IBM. Honeywell and Bull marketed IDS-2 -an evolved version, adopted by CODASYL)- successfully in the 1970s and 1980s.
However, IDS-2 data bases reflected the structure of enterprises and were not easy to restructure when the enterprises evolved. On the contrary, the relational model advocated by Codd of IBM, while more expensive in terms of processing constraints, succeded to dominate the world in the 1980s and later.
The strategy differed according the product lines of Honeywell and Bull. Multics and GCOS-8 developed their own implementation of the relational model establishing links with IDS data bases. GCOS-7 choose to port Oracle product as a server on their operating system. The latter implementation opened the way for specialized processor(s) for running the data base engine. Oracle's port was done by Bull's engineers but the contract with Oracle prohibits any know-how transmission of the data base engine technology.
High level Implementation languages
CAD Computer aided design
Revision : 27 juin 2001.