Transputer System Research Articles

Parallel implementations of numerical simulation of wind flow around buildings

In this work, two parallel implementation strategies for the numerical simulation of wind flow around buildings were developed based on a four-processor transputer system. The first parallel strategy is based on the functional decomposition of the problem. It is easily implementable, but the degree of parallelism is low. In the second strategy, both the functional and the domain parallelisms of the problem were utilized, so the degree of parallelism is highly increased. Numerical examples indicate that the parallel speed-ups and efficiencies achievable by the second strategy are significantly greater than those of the first one.

Power system parallel computation by a transputer network

Abstract The load flow and transient stability analysis are two of the most elementary computation problems that planners and operators must face to ensure power system security. With the growth in size and complexity of interconnected power systems, the electric power system computation, especially the time domain transient stability simulation for security assessment, is becoming more time-consuming or too expensive. There is always a considerable incentive to find better computation schemes and to speed-up electric power system computation. In this paper, we present an efficient and unified parallel algorithm with the coincident parallel Transputer system for both load flow and transient stability studies. The implementation results on an actual Chinese power system are discussed and showed that, the whole developed parallel processing system is very cost-effective to speed-up the power system computation.

Evaluating computation/communication balance in an object recognition task

Abstract I present performance measurements for some compute- and communications-intensive image processing tasks required by an object recognition application. The application was implemented as a task farm on a network of transputers, which still are one of the cleanest and best-balanced building blocks for parallel systems. I report on relevant properties of the basic algorithms the application is comprised of; some relevant details of the task farm software; performance measurements on a range of system sizes for tasks of very different granularity and characteristics (including broadcast of global data) with measured loads on processors and communications links; and the detrimental effects on performance of hardware variants that increase communications latency and reduce available bandwidth. The measurements show that for some of the image processing tasks, the transputer system is `on the edge' with respect to available bandwidth and latency. Some possible improvements to the communications infrastructure are discussed in light of these results. Finally, current approaches to affordable parallel computing, such as networks or clusters of workstations, are put into perspective by comparing them with the transputer system, using computation/communication balance as a figure of merit for the comparison. This shows that for current microprocessors, even their pin bandwidth is not sufficient to sustain equivalent (scaled) performance for this typical image processing application; any multiprocessor system built with current networking hardware is so unbalanced compared to the transputer system that it will be severely limited by communications bandwidth and latency.

Experiences with parallel job scheduling on a transputer system

Both time and space sharing strategies have been proposed for job scheduling in multiprogrammed parallel systems. This paper summarizes the major observations gained from an experimental investigation of these two partition sharing strategies on a Transputer system. A number of factors such as the applications and their software architectures in the multiprogramming mix, the partition sharing strategy, and the partition size are varied and the resulting insights into system performance and scheduling are presented. Space sharing is observed to produce a superior performance in comparison to time sharing for a number of multiprogrammed workloads. Time sharing showed a better performance for workloads with high variability in process execution times, and with high rates of interprocess communication. The relationships between system performance and a number of workload and system characteristics are presented.

Stochastic and neural models of an induction motor

The input/output representations of non-linear discrete time MISO-systems of an induction motor are discussed. To acquire the system data for estimation the motor was stimulated by a random binary speed reference signal. Different stochastic model structures (ARMAX and LARMAX) based on local models have been tested. Furthermore, two kinds of artificial Neural Networks, the Multilayer Perceptron Network (MLP) and the Radial Basis Function Network (RBF) have been used to model the process dynamics. Finally the model structures are compared with regard to the modeling validity and the computational expense on a transputer system.

On the parallelization of artificial neural networks and genetic algorithms

Simulating an ANN or a genetic algorithm on a parallel processing system one can use several techniques. This paper presents two methods on implementing parallel simulations of Artificial Neural Networks (ANNs) on Transputer Based Systems, using the C programming language under Helios O.S. and Component Distribution Language (CDL) or, alternatively, the 3L Parallel C language. The Processor Farm topology is used for the parallel implementation of Back-Propagation and Multi-Layered Feed-Forward ANNs. A transputer system was also used to implement a simulation of an island parallel genetic algorithm (PGA) and a new optimization method based on PGAs. The method, called Co-operating Populations with Different Evolution Behaviours (CoPDEB), is independent of the machine architecture. It allows the populations to exhibit different evolution behaviours by using a variety of selection mechanisms, operators, communication methods, and parameters as explained in the sequel.

Parallel algorithm for solving a problem of convection in porous medium

The solution of a steady-state problem of free convection in an inclined enclosure bounded by four rigid walls of constant temperature, filled with porous medium, is studied. The Gauss-Seidel iterative method implemented on a Transputer System is used. Several graphics showing the solution and performance of the parallel algorithm are given. © 1997 Elsevier Science Limited.

Parallel computation of super-/hypersonic flows on workstation network and Transputer arrays

A numerical method for solving the Euler equations based on the method of lines has been implemented on networked engineering workstations and a multi Transputer system using a domain decomposition approach. The message passing is handled by PVM (Parallel Virtual Machine) software. The present approach is tested for both 2D and 3D test problems governed by the Euler equations. For dynamic load balancing and reduction of computational work a strategy of selective solutions of subdomains based on local residuals has been employed. For steady supersonic and hypersonic flow problems, a reduction of 40–50% computational work and 12-1% of CPU time has been achieved.

A parallel implementation of a modified Richardson-Lucy algorithm for image de-blurring

Passive imaging in the millimetre wave region of the spectrum is attractive compared with imaging in the visible and infrared because of its better penetration through cloud and rain. However, the diffraction limited spatial resolution is far worse and computer restoration of millimetre wave images is necessary. The Richardson-Lucy algorithm is frequently used for image restoration because compared with linear algorithms it has reduced sensitivity to error in a priori knowledge, and fewer restoration artefacts. However, it requires many iterations to converge and is very slow compared with linear methods. This paper investigates modifications of the Richardson-Lucy algorithm which speed up the execution of the algorthm. Software techniques investigated include incorporating a speedup parameter, using dual point spread functions and implementation in the Fourier domain. A parallel implementation of the algorithm has also been investigated using a tree topology network of between 1 and 10 transputers. Finally an application of the transputer system to the de-blurring of a millimetre-wave image obtained from a mechanical scanning system is described in the paper.

A UNIFIED ADAPTIVE ACTIVE CONTROL MECHANISM FOR NOISE CANCELLATION AND VIBRATION SUPPRESSION

This paper presents the design and implementation of an active mechanism for noise cancellation and vibration suppression within an adaptive control framework. A control mechanism is designed within a feedforward control structure on the basis of optimum cancellation at an observation point. The design relations are formulated such that to allow on-line design and implementation and thus result in a self-tuning control algorithm. The algorithm is implemented on an integrated digital signal processing (DSP) and transputer system. Simulation results verifying the performance of the algorithm are presented and discussed.

Dynamics simulation of multibody chains on a transputer system

Dynamics simulation of multibody chains on a transputer system

Design and implementation of an adapative single pole autoreclosure technique for transmission lines using artificial neural networks

Adaptive single pole autoreclosure (SPAR) offers many advantages over conventional techniques. In the case of transient faults, the secondary arc extinction time can be accurately determined and in the case of a permanent fault, breaker reclosure can be avoided. This paper describes, in some detail, the design and implementation of a SPAR technique using artificial neural networks (ANNs). The design described includes special methods for extracting features from post-circuit breaker opening fault data, which is a prerequisite for setting up training data sets. The technique is then implemented in hardware based on a high performance T800 transputer system and some results obtained from laboratory tests of this equipment are presented.

A CASE STUDY OF PARALLEL PROCESSING: INFORMATICS VS SCIENTIFIC COMPUTING

Informatics and Scientific Computing approach parallel processing in a different way. We briefly describe the different points of view of both camps. Next we concentrate on a case study in the area of scientific computing. The problem chosen is from Physical Chemistry (self-consistent field computation). We describe the problem, the sequential solution, the parallelization strategy and present the performance values we have achieved. Our implementation is based on a 60-node transputer system, available at the Parallel Processing Laboratory in Basel.

Transputer implementation of a multiple agent model for object tracking

This paper describes a transputer system in which multiprocessors emulate a set of cooperative agents tracking moving objects in the scene. Agents have access to a pyramid image structure to process multi-resolution image segments containing objects of interest. At the same time, agents communicate through a blackboard structure which also provides an instantaneous description of the dynamic scene.

Fuzzy PD Control Applied to a Flexible Link Manipulator

This paper describes the application of a fuzzy PD control system to a manipulator with a single flexible link. Fuzzy systems which are inherently parallel lend themselves easily to implementation on a transputer system and should be able to take advantage of the parallel processing capabilities of transputer implementation. The system is evaluated in terms of control performance and speed. Simulation results show that the Fuzzy PD system performs better than the more conventional P, PD, PID and Pseudo Derivative Feedback (PDF) control schemes. Comparisons in performance between the transputer implementation and a simulation model written in advanced Continuous Simulation Language (ACSL) are also presented.

An experimental comparison of two distributed single-source shortest path algorithms

We experimentally compare two distributed algorithms based on Dijkstra's algorithm for the single-source shortest path problem. The algorithms are intended for asynchronous distributed systems with a small number of processors without shared memory, and specifically address the situation where communication is costly. Variations of the algorithms have been implemented in occam, and results from experiments on a 16 processor transputer system with randomly generated graphs of different types and densities with up to 20000 vertices and 250000 edges are reported. The distributed algorithms exploit the two obvious sources of parallelism in Dijkstra's algorithm. In the update-driven algorithm several vertices may be selected and scanned simultaneously. Since not all selected vertices can be guaranteed to be correct, the achievable speed-up depends on the problem instance, and no worst-case guarantee better than the sequential algorithm can be given. In contrast the minimum-driven algorithm performs the scanning in parallel but from only one selected vertex at a time. Ideally this algorithm has linear speed-up for dense graphs. Implemented naively both algorithms perform poorly, but can be improved by relaxations and approximations which reduce communication volume. In all cases the theoretically less attractive update-driven algorithm turns out to perform much better than the apparently preferable minimum-driven algorithm.

Performance of a QR algorithm implementation on a multicluster of transputers

Some results of an implementation of the QR factorization by Householder reflectors, on a multicluster transputer system with distributed memory are presented, that show how important is the communication time between processor in the performance of the algorithm. The QR factorization was chosen as test method because it is required for many real life applications, for instance in least squares problems. We use a version of Householder transformation that is the basis for numerically stable QR factorization. The machine used was the MultiCluster 2 model of Parsytec which is distributed memory system with 16 Inmos T800 processors. The Helios operating system was chosen because it provides transparency in CPU management. However it limits the sets of connecting topologies to be used. The results are presented in terms of speedup and efficiency, showing the importance of the communication time on the total elapsed time.

FMS Scheduling Using Branch-and-Bound with Heuristics on a Transputer System

Flexible production systems require large investments which implies that their optimal utilization is mandatory for their economic success. We deal with the optimal on-line allocation of the most costly resources, the machines, in such systems. Optimality is defined with respect to given due dates for the production of certain parts or batches of parts. The system may have machines with overlapping or identical functionality. The on-line scheduling refers to the fact that decisions cannot be made aforehand but have to be based on the actual situation of the system which may change at any instant due to the introduction of new jobs, changes of due dates, or disturbances or deviations from the nominal operationSuch scheduling problems are known to be NP-hard and thus optimization is in general not possible exect of very simple cases. Therefore, we developed an algorithm based on the branch-and-bound principle and uses heuristics to cope with the complexity. For improving the run time behaviour of our algorithm we parallelized our modified branch-and-bound algorithm and realized it on a transputer system. Our transputer system has only four transputers and we reached a speed up over three and an efficiency of nearly 0,8

Re‐working of two standard moment‐method codes for execution on parallel processors

AbstractTwo separate exercises on a similar theme are reported. In the first, the moment‐method program NEC was extensively modified in order to optimize its performance on a four‐processor Cray X‐MP computer. In the second exercise, the MININEC moment‐method program was completely rewritten into the occam programming language so that it could be executed directly by a transputer system. The NEC exercise exploited judicious use of vectorization and alternative parallelization techniques and achieved a reduction of run times by factors of about five or more. The MININEC exercise has so far achieved only a modest reduction in run‐time but it has the potential for a much greater improvement if an optimum strategy for configuring the transputer network is implemented.

Performance prediction of large MIMD systems for parallel neural network simulations

In this paper, we present a performance prediction model for indicating the performance range of MIMD parallel processor systems for neural network simulations. The model expresses the total execution time of a simulation as a function of the execution times of a small number of kernel functions, which have to be measured on only one processor and one physical communication link. The functions depend on the type of neural network, its geometry, decomposition and the connection structure of the MIMD machine. Using the model, the execution time, speedup, scalability and efficiency of large MIMD systems can be predicted. The model is validated quantitatively by applying it to two popular neural networks, backpropagation and the Kohonen self-organizing feature map, decomposed on a GCel-512, a 512 transputer system. Measurements are taken from network simulations decomposed via dataset and network decomposition techniques. Agreement of the model with the measurements is within 1–14%. Estimates are given for the performances that can be expected for the new T9000 transputer systems. The presented method can also be used for other application areas such as image processing.