While the West was dazzled by the inference engines, knowledge engineering and other exotic concepts from Japan’s Fifth Genera tion Project, and has tended to write it off as a failure, over 50 Japanese companies (including Fujitsu, Hitachi, NTT, Matsu shita, Oki and Toshiba) joined up with a number of universities and independent organisations to work on the more down to earth Tron Project. Perhaps because the Fifth Generation pro ject has failed to live up to the the ridiculous hype it received in the West (but not in Japan), the Tron Project has been either ignored or dismissed as being of little interest outside Japan. Total architecture Although the name stands for The Real-time Operating system Nucleus, the project is much more wide-ranging, covering almost every aspect of computer design and use. This total architec ture is open – anyone can use the Tron specifications at no cost, but the implementation must conform to the standards. As different applications may have very different requirements, it would be wasteful and unwieldly to take the Ada language approach and put everything into a single implementation. So the Tron designers decided to develop a family of closely related archi tectures, operating system kernels and VLSI processor chips, but all would have a common man-machine interface. This results in a range of machines tailored to different tasks that all look alike and work the same way for the user. The specifications provide a guideline for a whole range of highperformance ma chines, from personal workstations and embedded systems to power ful mainframes and network file servers, all conforming to a single family of open-systems computer architectures and with a uniform man-machine interface. A lot of attention has been paid to the way the machines and applications are presented to the user, right down to the level of ensuring that, for example, if one program uses a particular icon on a bitmapped display to represent a certain parameter, then every program on the whole range of machines use the same icon to represent that parameter. It provides a general framework for computer system design as well as specifications for specific system components from VSLI chips to a mainframe operating system kernel. There are separate but related versions for embedded industrial systems, worksta tions, business systems, large file servers, networking and for controlling intelligent objects – any machine or appliance under microprocessor control. As originator Ken Sakamura ex plains: After discussing the many shortcomings of existing systems with many designers and users, we concluded that the only way we were going to get satisfactory computer systems in the 1990s was to develop a completely new architecture. In this way we could address all the problems of existing systems at once and fix them now before carrying them into another generation of machines. Tron is not a specific product but a framework of direction for the Japanese computer industry. It is rather like a greatly expanded version of the X/Open Common Applications Environment, which is designed to allow users to mix’n’match components from the different manufacturers in the group. But the Common Appli cations Environment only really concerns hardware and its inter faces, with the different components held together by Unix. Tron’s open specification is layered rather like the Open System Interconnection model for data communications, with separate specifications for each layer and for the interfaces between them. The layers are: * The instruction-set processor or ISP layer; * The operating system kernel layer; * The operating system shell layer; * The application and man-machine interface layer. Only the instruction set processor layer deals with hardware, the other three layers being concerned solely with software. The operating system has been divided into two layers, the kernel layer and the shell layer. The latter is not a command inter preter like the Unix shell, but a collection of extended system service calls that can be regarded as opt
ional extensions to the operating system. For example, in an architecture that includes a bit-mapped display (such as a work-station), the shell layer might include a collection of system calls for the sophisticated control and management of the bit-mapped display. As they are part of the kernel, they would be much faster and more efficient than if they were implemented as part of an application, and would help to ensure a uniform user interface. However, in architectures such as embedded systems that do not include bit mapped displays, the shell layer would not include these system calls, allowing the operating system to be kept small and effi cient. The main objection the Tron team has against Unix is that encourages people to build their own userfriendly shells around applications for non-programmers to use. Anarchy They conceed that it can be a good operating system for worksta tions and that it has been very successful. But they strongly disapprove of the anarchy where almost every application has a different enduser shell or man-machine interface, making it difficult for the programmer to use different programs. While many people in the West find this an annoyance, it is enough to make Unix completely unacceptable to the Japanese designers. As Tron’s inventor, Ken Sakamura (a professor at Tokyo Univer sity) put it: Imagine having to deal with a set of application programs in which each program offers a different user inter face! (Note the exclamation mark!) Because computers are used for such a wide range of applications, the Tron designers have had to define four initial versions of the Tron architecture: * I-Tron for embedded industrial systems; * B-Tron for workstations and business systems; * C-Tron for large file servers and networking environments; and * M-Tron for interconnecting intelligent objects. The Tron designers consider that the most important difference in the demands the four classes of applications make on their architectures is in real-time response (hence the acronym). Industrial systems are considered to require a response time of a millisecond or less, while most business applications only require a response time of the order of a second. Each of these architectures has its own operating system kernel and a family of VLSI processor chips is under development with an instruction set designed to execute the operating system and application code efficiently. Processor architecture, instruction set, operat ing system, compilers, system software were all designed and optimised together. In part two, Geoff Conrad will report on the Tron state of play.
