ICL Distributed Array Processor Research Articles

Chemical database processing using parallel computer hardware

Research activities in the fine chemicals industry involve the generation and analysis of huge amounts of chemical and biological data, and this has led to the development of highly sophisticated, computer-based, chemical information systems to support the research [ 11. Most chemical databases contain two-dimensional structural data, i.e. planar chemical structure diagrams, but there is also increasing interest in the processing of three-dimensional structural data, e.g. from X-ray crystallography and molecular mechanics studies. In this paper, we summarize work in our laboratory over the last 3 years which seeks to determine the utility of parallel hardware for the efficient processing of the structural data that are contained in chemical information systems. Two of the very many different types of parallel computer [2] are discussed, these being the ICL Distributed Array Processor and the INMOS transputer.

Textual and chemical information processing using parallel computer hardware

This paper discusses the use of parallel computer hardware to increase the efficiency of processing in databases of text and chemical structures. After a general introduction to parallelism, two types of parallel computer are described: the ICL Distrib uted Array Processor and the INMOS Transputer. Experimen tal results are presented of the use of the DAP for cluster analysis, of the transputer for chemical substructure and maxi mal common substructure searching, and of both machines for text retrieval.

Parallel computational techniques for simulating dopant diffusion in silicon

Parallel computational techniques have been used to simulate dopant diffusion in Si (e.g. source/drain fabrication of a CMOS transistor) on an ICL distributed array processor (DAP), a single-instruction multiple-data-stream (SIMD) type machine. The finite element and conjugate gradient methods were used for solution. An order of magnitude reduction in computation time has been achieved.

EFFICIENCY OF HIERARCHIC AGGLOMERATIVE CLUSTERING USING THE ICL DISTRIBUTED ARRAY PROCESSOR

The implementation of hierarchic agglomerative methods of cluster anlaysis for large datasets is very demanding of computational resources when implemented on conventional computers. The ICL Distributed Array Processor (DAP) allows many of the scanning and matching operations required in clustering to be carried out in parallel. Experiments are described using the single linkage and Ward's hierarchical agglomerative clustering methods on both real and simulated datasets. Clustering runs on the DAP are compared with the most efficient algorithms currently available implemented on an IBM 3083 BX. The DAP is found to be 2.9–7.9 times as fast as the IBM, the exact degree of speed‐up depending on the size of the dataset, the clustering method, and the serial clustering algorithm that is used. An analysis of the cycle times of the two machines is presented which suggests that further, very substantial speed‐ups could be obtained from array processors of this type if they were to be based on more powerful processing elements.

Parallel Algorithms for Fluid Dynamics Problems

The recent trend toward increased parallelism in supercomputer architectures, together with the need to use existing supercomputer resources more efficiently, has motivated this study of parallel algorithms designed specifically for implementation on parallel machines. Users of parallel machines like the ICL Distributed Array Processor (DAP) are all too familiar with the problem of having to reformulate their algorithms to incorporate the inherent parallelism of the DAP architecture. The individual processing elements of the DAP are comparatively slow, and the full potential of such a machine is only realized if the algorithm used is highly parallel. The very fast processing capabilities of pipeline machines like the Cray-1 or the Cyber 205 tend to obscure this point somewhat, and in fact most users of these machines are content with using existing codes, vectorized where possible. Such a strategy, however, is not appropriate for the parallel processors like the ICL DAP, AMT DAP, the Goodyear MPP, GEC GRID, etc.The application of parallel algorithms to the solution of fluid dynamics problems is considered. Most of the work concerns the solution of the two- and three-dimensional incompressible Navier-Stokes equations, steady and unsteady. The results are relevant for the above-mentioned parallel machines, but the methods can be adapted for use on vector machines.Concerning parallel algorithms, the easiest and possibly those with the widest potential application are relaxation techniques based on red-black (R-B) ordering. The use of R-B successive overrelaxation, R-B LSOR, etc., on two- and three-dimensional cavity flows, unsteady channel, and boundary layer flows at high Reynolds numbers is considered. While these methods are extremely problem dependent, in the favorable cases they can be as competitive as more sophisticated serial “vectorized” algorithms. As far as parallel algorithms are concerned, the most important criteria are what is the cost per iteration, and how fast do they converge. For application on serial machines, the latter factor seems to dominate.

Bibliographic pattern matching using the ICL Distributed Array Processor

This article discusses the use of the ICL Distributed Array Processor (DAP), a highly parallel array processor containing 4096 processing elements, for pattern-matching operations in a bibliographic retrieval system. The hardware and software features of the DAP are described, together with a pattern-matching algorithm that makes full use of the DAP's parallelism. This algorithm can be used to search for exact patterns, right-hand or left-hand truncated patterns, embedded patterns, patterns containing fixed-length (FLDC) or variable-length (VLDC) don't care characters, and patterns that specify adjacent words. Experiments with a set of 898 query patterns and 2472 titles and abstracts from the INSPEC database show that right-hand truncation searches can be carried out on the DAP about ten times faster than searches using the Aho-Corasick pattern-matching algorithm on a Prime 550 minicomputer, the relative advantage of the DAP increasing linearly with a decrease in the number of query patterns. Simulated FLDC and VLDC searches on the DAP take about two and three times as long as right-hand truncation searches, respectively. © 1988 John Wiley & Sons, Inc.

Searching and clustering of databases using the ICL Distributed Array Processor

The ICL Distributed Array Processor (DAP) is an SIMD array processor containing a large, 2-D array of bit serial processing elements. The architecture of the DAP makes it well suited to data processing applications where searching operations must be carried out on large numbers of data records. This paper discusses the use of the DAP for two such applications, these being the scanning of serial text files and the clustering of a range of types of database. The processing efficiency of the DAP, when compared with a serial processor, is greatest when fixed length records are processed.

Neural network models

The remarkable processing capabilities of the nervous system must derive at least in part from the large numbers of neurons participating (roughly 10 10), since the timescales involved are of the order of milliseconds, rather than the nanoseconds of modern computers. We summarise common features of the neural network models which attempt to capture this behaviour and describe the many levels of parallelism which they exhibit. A range of models has been implemented on the SIMD (ICL Distributed Array Processor) and MIMD (Meiko Computing Surface) hardware at Edinburgh. Examples include: (i) training algorithms in the context of the Hopfield net, with specific application to the storage of words and text with content-addressable memory; (ii) the back-propagation training algorithm for the multi-layer perception; (iii) image restoration with Hopfield and Tank analogue neurons, and (iv) the Durbin and Willshaw elastic net, as applied to the travelling salesman problem.

A spectral meteorological model on the ICL DAP

The key points to the implementation of a meteorological spectral model on the ICL Distributed Array Processor (DAP) are presented. Spectral models involve transforming the variables between spectral and gridpoint space and these spectral transforms comprise of Legendre and fast Fourier transforms. The storage format of the data is discussed and the algorithms used for the Legendre transforms presented. Timings of these algorithms are compared with those from a serial machine.

Scientific computation on SIMD and MIMD machines

The ICL Distributed Array Processor and the Meiko Computing Surface have been successfully applied to a wide range of scientific problems. I give an overview of selected applications from experimental data analysis, molecular dynamics and Monte Carlo simulation, cellular automata for fluid flow, neural network models, protein sequencing and NMR imaging. I expose the problems and advantages of implementations on the two architectures, and discuss the general conclusions which one can draw from experience so far.

Automatic classification of chemical structure databases using a highly parallel array processor

AbstractThis article describes the use of the ICL Distributed Array Processor (DAP) for the automatic classification of chemical structure databases using the Jarvis‐Patrick clustering method. This method is based upon the calculation of a table containing the nearest neighbors for each of the molecules in the database which is to be clustered. These nearest neighbors can be identified very efficiently using the DAP since it allows up to 4096 molecules to be compared with a specified molecule in parallel. Experiments with files of 4096 and 8192 structures from the Fine Chemicals Database show that clustering with the DAP is up to 6.7 times as fast as using a highly efficient, inverted file algorithm on an IBM 3083 mainframe.

Content-addressability and learning in neural networks

The content-addressability of patterns stored in Ising-spin neural network models with symmetric interactions is studied. Numerical results from simulations on the ICL distributed array processor (DAP) involving systems with up to 2048 neurons are presented. Behaviour consistent with finite-size scaling, characteristic of a first-order phase transition, is shown to be exhibited by the basins of attraction of the stored patterns both in the case of the Hopfield model and for systems using a local iterative learning algorithm designed to optimise the basins of attraction. Estimates are obtained for the critical minimum overlaps which an input pattern must have with a stored pattern in order to successfully retrieve it.

Software and hardware techniques for string searching in serial document databases

This paper considers the use of text signatures and of parallel computer hardware to increase the efficiency of text scanning in serial document files. After a review of software and hardware techniques for text scanning, the characteristics of text signature systems for document retrieval are discussed, in particular the relationship between text signature length and both the efficiency and the effectiveness of searching. A parallel pattern matching algorithm is then described for use with the ICL Distributed Array Processor, a highly parallel SIMD array processor. The efficiency of this algorithm is compared with that of the Aho-Corasick pattern matching algorithm when implemented on a conventional minicomputer. Microprocessor-based multiprocessing networks are becoming widely available as a means of implementing MIMD parallelism. Experiments with the Boyer-Moore pattern matching algorithm on a network of INMOS transputers demonstrate near-linear speed-ups for networks containing up to 11 transputers.

Implementing Neural Network Models on Parallel Computers

The remarkable processing capabilities of the nervous system must derive from the large numbers of neurons participating (roughly 10 10 ), since the time-scales involved are of the order of a millisecond, rather than the nanoseconds of modern computers. The neural network models which attempt to capture this behaviour are inherently parallel. We review the implementation of a range of neural network models on SIMD and MIMD computers. On the ICL Distributed Array Processor (DAP), a 4096-processor SIMD machine, we have studied training algorithms in the context of the Hopfield net, with specific applications including the storage of words and continuous text in contentaddressable memory. The Hopfield and Tank analogue neural net has been used for image restoration with the Geman and Geman algorithm. We compare the performance of this scheme on the DAP and on a Meiko Computing Surface, a reconfigurable MIMD array of transputers. We describe also the strategies which we have used to implement the Durbin and Willshaw elastic net model on the Computing Surface. Received June 1987

A review of parallel finite element methods on the DAP

This paper reviews the research work that has been done to implement the finite element method for solving partial differential equations on the ICL distributed array processor (DAP). A brief outline of the principle features of the method is given, followed by details of the novel techniques required for implementation on the highly parallel architecture. Various methods of solution of the finite element equations are discussed; both direct and iterative techniques are included. The current state-of-the-art favours the use of the preconditioned conjugate gradient method. Some suggestions for future research work on parallel finite element methods are made.

Use of text signatures for document retrieval in a highly parallel environment

This paper considers the use of text signatures, fixed-length bit string representations of document content, in an experimental information retrieval system: such signatures may be generated from the list of keywords characterising a document or a query. A file of documents may be searched in a bit-serial parallel computer, such as the ICL Distributed Array Processor, using a two-level retrieval strategy in which a comparison of a query signature with the file of document signatures provides a simple and efficient means of identifying those few documents that need to undergo a computationally demanding, character matching search. Text retrieval experiments using three large collections of documents and queries demonstrate the efficiency of the suggested approach.

The design and implementation of a pascal-based language for array processor architectures

This paper describes the programming language Actus II, which has evolved from the Pascal-based parallel language Actus, and has also been influenced by the architecture of array processors. This language facilitates the construction of parallel algorithms in a notation which is independent of the underlying architecture. Work on the implementation of a compiler for the ICL Distributed Array Processor (DAP) is currently under way and some aspects of this implementation are described.

Parallel computation of cluster properties: application to 2D percolation

The authors discuss various parallel algorithms for the enumeration of two-dimensional cluster properties in such problems as percolation and the Ising model. As an application, in a Monte Carlo simulation performed on the ICL distributed array processor (DAP), they re-examine the numerical evidence for a recent conjecture by Jug (1986) concerning percolation singularities in two dimensions.

Adapting a Navier-Stokes Code to the ICL-DAP

This is a report of the results of an experiment: to adapt a Navier-Stokes code, originally developed on a serial computer, to concurrent processing on the ICL Distributed Array Processor (DAP). In this paper the algorithm used in solving the Navier-Stokes equations is briefly described. The architecture of the DAP and DAP Fortran is also described. The modifications of the algorithm so as to fit the DAP are given and discussed. Finally, performance results are given and conclusions are drawn.

A parallel approach to the picture restoration algorithm of Geman and Geman on an SIMD machine

This paper considers how the image restoration technique of Geman and Geman, which involves searching for the maximum a posteriori distribution of an image modelled as bounded Markov random fields using simulated annealing, can be approximated on a parallel SIMD processor array, the ICL Distributed Array Processor (DAP). For the version implemented, the potential speed-up over an equivalent serial processor is equal to half the number of processors in the array, or 2048 for the 64×64 DAP. The time taken by the DAP for one updating cycle is about 30 ms, and the typical complete picture restoration consisting of 1000 annealing cycles takes around 30 s. A method of increasing the efficiency of the present algorithm is suggested and the possibility of making the algorithm work with several frames of data collected over time is discussed.