Parallel Computing Technologies(PaCT –91) by Prof. N. N. Mirenkov

USSR Academy of Science Computing Center, Novosibirsk

We are currently entering an era of developed parallelism. Parallelism provides the high performance needed in science and engineering, the responsiveness required in real- time control, the fault tolerance necessary for high-reliability systems etc. These capabilities become available for use in an increasing number of applications, but offer new challenges to the system designer and programmer.

The advanced research in parallel computing technologies as well as the development of high industrial quality products are now especially important for the USSR as the country attempts to become tightly integrated with the world community. We believe that increasing the research contacts and scientific cooperation is one of the first steps to take. So, in spite of the considerable number of international conferences and workshops on parallel systems, we decided to establish a Soviet series on parallel computing technologies (the PaCT series). As a result, the PaCT-91 Conference was organized by the Computing Centre of the Siberian Division of the USSR Academy of Science in co-operation with the Soviet Parallel Computing Association and held in Novosibirsk on the 7th – 11th September 1991. This was the first time that such an international conference on parallel computing technologies had taken place in the USSR. 100 participants from 11 countries were present at the Conference.

The Conference began only a few weeks after the 'putch'. So, we feared lest the delegates should fail to arrive. But the participants showed that they were brave people. In spite of great travel troubles, practically all of them arrived at Novosibirsk in time. We would like once more to thank them for coming to Novosibirsk and we are deeply grateful to those who sent us letters of support.

Keynote introduction

The Conference Keynote Introduction was presented by Prof. W Handler (Erlangen, FRG) and Prof. Ju. Mitropolski (Moscow, USSR). W. Handler in his presentation 'Nature needed billions of years...' noted that: The teraflop multiprocessors will represent a means to an end-to numerical simulation, to huge databases, to artificial intelligence, to expert systems and to many other things. The main question will nevertheless be whether we will wisely utilise these tools to preserve our world and its natural resources from destruction. conflicts or wars.

Ju. Mitropolski in his presentation gave his own interpretation of the keynote remarks prepared by Prof. A. Melnikov, in which he commented: 'I think that the social aspect of supercomputing will be very important during the next few years. It will be necessary to incorporate a special subject on supercomputing into the programmes of the universities and technical colleges. Also, it is very important to provide a wide access to modern supercomputing for young researchers and engineers. In the USSR we have now begun to manufacture computers of the CRAY type to support advanced work.

Invited papers

Invited papers at PaCT-91 included:

• R. H. Perrott (UK): 'Parallel programming language developments'. This paper was concerned with the topic of parallel programming language design and implementation. An outline of the main Issues affecting the spread and acceptance of parallel computing was given followed by consideration of specific approaches to parallel programming.
• Ch. Lengauer (UK): 'A view of systolic design'. The presentation was concerned with an overview and bibliography of recent results and current issues in the mechanical synthesis of systolic arrays.
• H. Zima (Austria): 'Automatic parallelisation for distributed-memory systems'. In current automatic parallelisation systems for distributed-memory machines, the user must explicitly specify how the data domain of the sequential program is to be partitioned and mapped to the processors. In this presentation, the author outlined the salient features of a new knowledge-based software tool that provides automatic support for this task.
• D. R. Wilson (UK): The next generation of computer systems'. The presentation was concerned with the state of the art and future trends in parallel machines as well as with the penetration of parallel computers into industrial and commercial computing systems. Prof.Wilson's opinion was that among the firms manufacturing the chips those would be the first who would create the teraflop performance computers It must be noted that this presentation made the strongest impression on the audience, especially on the Soviet participants.

Other presentations

I would like to pay special attention I to some results of Soviet B contributors.

B. A.Babajan in his paper asserted that their architecture made It possible to ensure high logic performance, which was three times as high as the performance of any other modem architecture, including superscalar (IBM RS/GOOO, Intel 80860 and 80960 etc.) and vector pipeline (Cray Y-MP. NEC SX-5 etc.) architectures.

N. N. Mirenkov and S. A. Simonov suggested linear algorithms for the static analysis of deadlock freedom and uniqueness of parallel programs.

V.Kutepov and V. Falk described the project for the integration of the functional, logical and data-flow parallel programming. An attempt to integrate two techniques, declarative and procedural, was 31% considered by S. P. Corlatch and G. N. Corlatch.

A formalism for changing the depth of cellular computation parallelism at the expense of its slowdown was presented by 0. L. Bandman. A method was created to get from the computation within a k-dimensional cellular array of finite size to result- equivalent computation within another array of smaller size.

A. P. Vazhenin, S. G. Sedukhin and Ya. I. Fet suggested a new type of parallel architecture. Tile principal idea was to combine mutually supplementing useful features of systolic/wavefront arrays and STARAN-like system.

N. N. Mirenkov and L. A. Bulysheva considered an approach of multifunctional co-operation over multiple data within the framework of one procedure. They suggested a new type of library procedure as well as sequential program integrators that executed transformations inverse to those of parallelisers.

N. A. Anisimov presented a formalism that was intended to implement a modular approach to the specification of complex distributed and concurrent systems. A new notion of an entity was introduced. It was defined as a Petri net together with a set of labelling, each labelling being treated as an access point

Intended for communication with other entities.

A class of n-dimensional regular graphs was considered by E. A. Monakhova. A brief survey of the works of Russian researchers studying the graphs and new results on the analytical representation of optimal structures and on the existence of optimal two-dimensional networks were presented.

V. A. Anisimov and V. E. Malyshkin described in their paper the parallel programming system INYA. It enables one to develop the applied parallel program which can be effectively executed on specially defined large- block hierarchical multicomputer systems.

B. D. Korneev, N. N. Mirenkov and A. Sh. Nepomniashaya considered a very-high-level language PARIS related to AI applications. One of the principal target applications was the efficient interpretation of other wellknown very-high-level languages. Among the aims of the project PARIS is the formulation of requirements for future parallel machines.

And last but not least I would like to mention a problem of mapping of the information graph of a complex algorithm into the graph of the interprocessor network of the parallel system, which was considered by 0. G. Manakhov. He presented the parallel distributed algorithm for optimal or suboptimal solution of the mapping problem as well as the objective functions and necessary conditions of optimal mapping.

Conclusion

It was pleasing to loam during the PaCT-91 Conference that an I eminent scientist and participant, Prof. W. Handler. was named doctor honoris causa of Novosibirsk University.

IhePaGr-91 International Conference was a successful scientific meeting. The PaCT-93 conference will be held in the European part of the USSR. We are considering three possible venues for the conference: Sochi on the Black Sea, Solnechnogorsk on the Baltic Sea or a touring ship which travels from Moscow to Astrachan and back by the river Volga. In the new political situation in the Soviet Union it is very important for Soviet scientists and engineers to make contact with their Western colleagues. If you would like to contact any PaCT-91 author. please contact either myself or the Chairman of this Journal's Editorial Advisory Panel, Prof. Derek Wilson.

The Conference Proceedings were published by World Scientific Publishing Co. Pte. Ltd. (Singapore).

COMPUTING & CONTROL ENGINEERING JOURNAL JANUARY 1992