Engineering
Engineering is with sales and manufacturing the most important components of an Information Technology company. The structure of Engineering organization has evolved essentially with the history of mergers, while the local culture of the engineering labs went often essentially unchanged in spite of those mergers.
General organization
It was basically impossible to establish a central entity for Engineering while the legal and financial structure of the group was not unified. For instance, the Italian Engineering did not report to a central engineering entity before 1989.
At General Electric times, there had been an attempt, by Dick
Bloch in 1970, to establish a centralized structure to control the APL developments from
Stanford, CT.
Within Honeywell, a central coordination office (NPL technical Office) was handle the
specifications and planning of the new product line. However, the on-going projects were
controlled locally.
In 1974, for some time, Phoenix and Boston operation were merged, as some key personal
moved or returned to Phoenix. But, with the development of Level 6, the Boston
organization took back its autonomy in 1975.
With the CII-HB merger, the Engineering shops started to be divided into Large Systems
(ex-CII), Medium Systems (ex-HB), Small Systems and Peripherals. When it was decided to
migrate Siris systems towards GCOS7, a single multi-sites organization was established for
mainframes. GTI (Groupement Technique et Industriel) covered Engineering of and
Manufacturing included a Minicomputer ("Frenchization" of Level 6) operation,
the Small Computer and terminals operation and the Angers-Joué les Tours manufacturing
operation. Peripherals had a larger autonomy (Engineering and Manufacturing operating as a
single entity).
Eventually, in 1989, all the worldwide engineering labs were made reporting to a single structure under Christian Joly, then Lucio Pinto and Xavier Stefani. The intent was to mobilize enough resources to take a dominant stake of the UNIX market, while retreating smoothly from the mainframe business. That policy was successful in developing a SMP Escala UNIX system in common with IBM, and in becoming more independent of NEC in GCOS8 systems.
In 1994,this trend towards a centralized approach to engineering faded out while priority was shifted from long term projects to immediate profitability, and engineering became again oriented towards product lines immediate goals.
Hardware
The following description of engineering skills apply to central processors and
to peripheral hardware that have been designed internally. The use of externally acquired
off-the-shelf processors that began to phase in during the 1980s may eventually make those
skills redundant.
Many techniques contribute to hardware engineering:
Technology:
Originally, in the 1950s, a computer lab had to design a computer starting from elementary
electronic parts, like diodes, relays and tubes. The electronic lab was responsible for
the measurements of those parts, the fact that they fit together, the manufacturing
tolerances etc..
From 1965, the acquisition of integrated circuits (initially SSI, MSI medium scale
integration soon) made the above responsibility essentially on the side of the vendor.
With custom VLSI, the manufacturing was on the vendor side and the interface of the
computer manufacturer was a set of rules enforced by CAD. The Bull technologists had
essentially a responsibility of dialog with vendors and of vendor selection.
They had also
to deal with packaging of circuits on the board, selection of connectors taking in account
that boards manufacturing stayed with Bull manufacturing.
Liquid cooling was a peculiar technology issue that was used only in Phoenix. Other labs
were using air cooling only.
Implanters
Originally, the role of implanters was to prepare manufacturing drawings for the wire
interconnected components and boards.
When the speed of computer started to be distance limited (around 1965 too), a specific
task was to position components on printed circuits boards and boards in card-cages.
Initially, that was made manually. The systems of that period was often designed in worst
case and were speeded up after the first batch of machines when time was available for
measurements and computations.
With CAD tools, the task of implanters disappeared, their work being taken up by
computers.
Logic designers
That function is usually considered as the more noble in hardware engineering. It includes
ideas creation to translate an architecture in a system design, evaluation of the gross
design and attention to the detailed design. Progressively, when CAD tools became
sophisticated, logic designers became almost the only human involved in hardware design.
Microprogrammer
As the majority of Bull computers were microprogrammed, logic designers were not actually
address the software visible architecture but another machine. Architecture actual
implementation laid under the responsibility of specialized programmers. The talents
required from them were similar to process control programmers, they could not trade
performances against productivity and their responsibility for bugs was very high, when
microprograms were in ROM.
Tests and Diagnostics programmers
Initially, field engineers understood the details of the circuitry and were able to
diagnose a failed component by the drawings. Progressively, that became impossible because
the complexity of the system and engineering was demanded by Filed Engineering to have the
system to say what part was to change. Service processors were introduced in the mid-1970s
and they had to be programmed. In 1980, non-functional testing became the rule and logic
design had to be made with the part diagnosis in view. It ended by using that feature to
be used in the initialization of the system as well.
With the generalization of single chip processor around 1990 the amount of T&D work
decrease. Not only parts were more reliable, but the replaceable unit became as big as
a function easily identified by the customer him(her)self.
CAD programmers
CAD tools were introduced initially in Phoenix for Level 66 and Bull developed its CAD
factory on GCOS8 until the 1980s. VLSI design was done interactively on different types of
graphic workstations, main frames being used as back-end servers for CAD.
Architects
A limited number of individuals had been the so-called architects, defining the
instruction set of the system and its evolutions, receiving software people suggestions
and logic designers complaints, taking in account measurements by performance measuring
people.
Architecture design is a huge work when a new product line is introduced or when the
emulation of one architecture on another is asked for by marketing.
Architects too have to watch the competition, sometimes benchmarking some of its products
and helping Marketing to react.
Qualifiers
A specific functions requiring relatively important expenses is the checking and the
control to the conformity to standards. Manufacturing to the worldwide market means that
your product must pass specifications: from air flow in Lhasa to noise in Stockholm and
earthquakes in Japan. The work depends on technology, but the early 1980's technology
caused issues in EM emission and sensitivity, as well as noise.
Basic Software
Software Architects
Software architects are dealing with carving software into pieces to allow the development
to be shared, defining the right interfaces between components, insuring a coherent user
interface through an operating system and between different OS. They deal frequently with
their hardware counterpart, specially in performance issues.
Software designers and programmers
Within Bull, in USA and in Europe, the task of designing and programming had been
generally assigned to the same group of individuals. The alpha testing of components was
also the responsibility of developers.
Qualifiers
After component integration, the "software release" is transmitted to an
internal independent organization within engineering that perform "preliminary
beta" testing. In early computers (from Gamma 60 to GE-600) software was delivered
far too early to soon unhappy customers, so it become necessary to perform non regression
and new functionality testing by a Quality Assurance testing unit for several months
before FCS (first customer shipment).
QA organization included some people coming from support and some ex-developers preferring
a less stressing, if not so glamorous, work.
A separate page summarizes the reaction of Bull engineers to computer technology innovations, whatever they contribute, participate or
resist them.
NB: the software developed inside Bull group companies as application or hardware independent software product id described here
Revision : 29 juin 2001