In his keynote speech at the start of Transputer Applications 90 in Southampton last week, Tony Hoare set out the conference theme, by identifying what he saw as the important question can Transputer architecture evolve into a general-purpose computing surface? Hoare has been involved with Transputers in various ways since the beginning. In 1977 he was brought in as a consultant at Inmos International Ltd, the Bristol-based company that designed the chip. Now at Oxford University, Hoare is optimistic about the future for Transputers, but he also recognised the current sticking point preventing Transputer architecture reaching that general-purpose status. Programmers, it seems, preferred to continue programming in Fortran and C, rather than switch to Occam, the programmimg language designed by Inmos for Transputers. Some two years ago, parallel versions of such languages as Fortran began to appear.
But if programmers are not persuaded to use these new languages to develop applications, the evolution of Transputers into a general surface will surely be tardy. Hoare spent the bulk of his speech explaining to the assembled programmers and engineers the basic theory of Occam in algebraic terms, going through the formulae in the hope of being able to provide a few programming tips. The conference seemed to be academically rather than commercially oriented, with the 300 attendees were mostly from academic institutions. It is the second in a series, hosted by the Southampton centre for Transputers, one of six around the UK. The centres were set up by the Transputer Initiative, jointly funded by the Department of Trade and Industry and the Science & Engineering Research Council. Its aim is to increase the profile of Transputing technology and encourage the development of a viable software base for that technology. Technology transfer to British industry is one of its ultimate goals. Practical assistance is provided to would-be programmers at academic sites, by loaning them personal computer and Sun workstation-based Transputer add-on boards with supporting software. Any software that is then developed may be incorporated into the Initiative’s Software Exchange Library which academics can also use for further research and development and, it its hoped, moving towards application software along the way. In the commercial world, which was represented at the conference by an accompanying exhibition, the Transputer is generally used as a fast processing chip, to speed up existing applications.
By Sonya McGilchrist
It has done very well in this respect, and many claim that it is the best selling RISC chip worldwide. But large-scale application software, specifically designed to take advantage of the new parallel possibilities offered by the Transputer is was not much in evidence. Simon Holdship of Transputer Technology Solutions, the Transputer Initiative’s Southampton centre and the conference host, says that large scale commercial application software will not be long in coming, but that it requires huge investment in research. Converting existing software is cheaper. Besides which, companies are reluctant to move to new software because the existing software was probably expensive to install in the first place and it has also been tried and tested. As well as running training and consultancy services, TTS converts existing code for Transputer architecture, reducing the time to run applications and increasing the turnaround time for processor-intensive applications. Converting software for the oil company Haliburton-Tesel cost around UKP5,000 according to Holdship. Developing a large-scale application for Transputer hardware is something TTS would like to do, but the money required makes it infeasible at the moment. Steven Doyle at Sension reckons that one reason why software is lagging behind hardware development is that the Transputer mystique – that Transputers are high-tech academic things – is only just beginning to be dispelled. Even now, there are still no applications, only the development tools for spee
ding up existing ones. But with the introduction of popular programming languages in parallel versions, Doyle reckons the change will come, probably in the financial area. However, not all applications are suited to parallelism and the problems of splitting some applications are, according to Steven, non-trivial. One company that has made use of the increased computing power of Transputers is Volkswagen AG. Peter Zimmermann from Volkswagen explained how the company has used Transputers for its visual simulation program, which enables designers to test drive a car on-screen without going through the expensive process of building a prototype. Using Transputers, the simulation program can be run cost-effectively in real-time, with the required features such as weather variations, texturing and hard and soft shading. Aircraft simulators cost between Dm10m and Dm100 but the Transputer system at Volkswagen costs around three times the price of a new car – which is obviously more cost effective. Looking at the Transputer in a European context, David Talbot of the European Commission stressed that the key to future success is building up confidence in the Transputer community, which includes customers.
Makes one’s head ache
He said software was an essential underpinning to this, software to help build applications and thus illustrate to end users the potential of this new approach to computing, but also, application software addressing an ever increasing range of problem areas. To help widen the skill base so essential to the real success of parallel systems, the Esprit programme last year launched Parallel Computing Action. Its aim was to set up Europe-wide centres with interests in the application of parallel computing systems, the development of software for these systems and the introduction of the technology to future parallel systems programmers. And 55 Transputer-based parallel systems are now deployed throughout the Community, supporting 200 reserchers and producing 300 students each year with an understanding of the applications that this class of system can tackle, as well as practical experience in programming them. As with the Transputer Initiative, users are required to share any software that results. Building parallel systems still makes one’s head ache, according to David Talbot and therefore better tools do need to be built, but he agreed that getting applications onto existing boxes is also vital to the future development of the Transputer industry.