Transputer System Research Articles

A reusable load balancer for parallel search problems

In this paper, we describe the design and implementation of a reusable load balancer for parallelsearch problems. Parallel search problems are a significant research topic in the parallel processing area: these applications expose irregularities that require dynamic load balancing techniques. Software for dynamic load balancing should be reusable. Our load balancer can be developed independently from the application -and thus be replaced and reused- as it does not directly interact with the application itself. Our prototype implementation runs on a transputer system; we used different load balancing strategies: sender-initiated, receiver-initiated and symmetrically-initiated. The modularity of our load balancer is illustrated by the fact that we could easily replace the load balancer by different alternatives and compare the performance results.

Scaling aspects for the performance of parallel algorithms

We investigate the performance of two-dimensional cellular automata simulations on two parallel computers of the message-passing type, namely a small Transputer system and one SUPRENUM-cluster. The timing results can be fitted with a simple modelling function with parameters directly connected to properties of the algorithm and the computer system. They exhibit scaling behaviour and it can be shown that the resulting scaling function is universal in the sense that is is valid for a wide range of algorithms.

Analysis of simulated triples γ-ray data on a 100-processor transputer system

We present a study of the administration and analysis of simulated nuclear 3-fold coincidences implemented on a 100-processor transputer system, which

“D 2 R”

The task-level dataflow language D 2 R ("Dynamic Dataflow Representation") is suggested (Version 1.0). D 2 R uses dynamic branching, loops with data-dependent numbers of iterations and alternatives. The dataflow information can be expressed either as a program text or as graph. Primarily the model has been developed for cooperation with dynamic task scheduling algorithms. Its suitability is demonstrated by examples. The representation supports environments allowing to simultaneously execute multiple user programs ("space sharing") as well as to restrict the assignments of task and data nodes to special processors (e.g. I/O nodes). The system has been implemented using the multi transputer system "DAMP" /Bauch-91/.

A C++ language interface for parallel programming

This paper presents an object-oriented interface for parallel programming, and an algorithm for automatic translation into parallel programs. The programming interface consists of a restricted subset of the object-oriented language C++. Parallelism is defined explicitly at the abstract level of object definitions and method invocations within a single C++ program. The translator algorithm first generates a machine-independent communication graph and proceeds with the creation of the parallel programs, which are demonstrated for transputer systems with the HELIOS operating system. The necessary communication statements are generated automatically.

Numerical Simulation of Incompressible Flow on Parallel Computer with the Domain Decomposition Technique.

The domain decomposition technique, which is widely used in the solution of a flow field on multiprocessors, requires a treatment of new boundaries appearing between subdomains. Poor convergence tends to be introduced due to these new boundaries where otherwise very frequent communication between neighboring processors is required. We determined how frequently processors should communicate with each other so as to obtain the best efficiency. In this study, steady 2-D incompressible laminar and turbulent flows are solved on the transputer system equipped with distributed-type memory. The SIMPLE method is employed as a numerical algorithm for the incompressible flow solution. The performance of parallel programs, such as the efficiency and increase in speed, is discussed for the cavity and sudden expansion flows.

Parallel processing network for adaptive controllers

The paper presents a top-down methodology for an approach to the systematic design of medium-to-coarse grain multiprocessor systems for real-time computer controllers. The scheme exploits the processing requirements of control applications to simplify software development and maintenance. A design example of a transputer system for the adaptive control of a prototype three degrees of freedom direct drive arm, currently being developed at Oxford University, is given to exhibit the validity of the design philosophy.

A correlation coprocessor for accurate real-time ultrasonic ranging of multiple objects using the transputer

In mobile robotics one often uses ranging sensors because they are cheap and provide relatively accurate ranging data. In their simplest form, ultrasonic sensors provide the distance to the nearest discernible object with an angular uncertainty of 10° to 20°. Although this information is suitable for collision avoidance, it is absolutely inadequate for more complex tasks like building an environment map. One has to consider more sophisticated methods. In this paper we briefly describe a radar-inspired method to extract more information from a single ultrasonic reading than is possible with the traditional approach. In addition, we describe a coprocessor ASIC design that significantly accelerates the required signal processing task on a low-cost transputer system. The coprocessor behaves like an ‘intelligent’ A/D converter in that it performs the actual acquisition of the sensor's analogue signal, as well as a simple correlation operation. It is connected to the transputer by means of a serial link, which permits effective parallelization of the signal processing steps. The resulting system has been evaluated and performs according to expectations.

A SYSTOLIC ARCHITECTURE FOR CAPON'S DIRECTIONS-OF-ARRIVAL (DOA) ESTIMATION METHOD

In this paper, a systolic array architecture is designed for the parallel implementation of Capon's direction-of-arrival estimation method in array signal processing. The architecture uses the QR decomposition on the data matrix to compute Capon's angular spectrum. The systolic array employs N(N + l)/2 + N processing elements (PEs). The sampling data and the steering vector enter into the systolic array while the angular spectra flow out of the systolic array. The parallel algorithm is simulated by the parallel programming language OCCAM 2 in a single transputer system. The architecture is very efficient with a nigh degree of parallelism.

Data acquisition system based on the T800 transputer for the PC-bus

This article describes an extensible transputer-based data acquisition system mounted on a PC card format. Comparisons of this large card format are made with the transputer module standard (TRAM) formats with reference to their application in engineering systems, in particular for real-time control, instrumentation and signal processing. Two examples of the use of a multi-processor transputer system are given: the simulation of a broad loom weaving machine for production testing of electronic jacquards, and as a high performance vision system.

Evaluation of a rational function

Evaluation of a given rational function can be carried out in various ways, such as by Horner's method, directly through a continued fraction representation, or by a subsequent process that uses three-term recurrences. Four such algorithms are summarized, and their relative performances when they are implemented using a parallel architecture (a four-node transputer system) is investigated.

The time‐parallel multigrid method

AbstractWe consider the problem of solving time‐dependent partial differential equations on an MIMD computer. Conventional methods for the solution of this type of equation when discretized by an implicit method such as backward Euler proceed by solving a sequence of problems iteratively. It is shown that, despite the sequential nature of this process, several processors may be employed to solve at several time steps simultaneously, a technique which we shall refer to as time‐parallelism. Details of a time‐parallel multigrid method based on the parabolic multigrid method due to Hackbusch and of a smoothing technique applicable to the two‐dimensional unsteady incompressible Navier‐Stokes equations are given. The parallel complexity of the method is shown to be qualitatively superior to that of previous parallel algorithms. Efficiencies obtained by an implementation of the method on a parallel Transputer system are presented.

A five-diagonal finite-difference method based on mixed-type interpolation for computing eigenvalues of fourth-order two-point boundary-value problems

A modified finite-difference method based on mixed-type interpolation is derived and applied for computing eigenvalues of fourth-order two-point boundary-value problems. A parallelized version of the resulting algorithm is discussed and implemented on a Helios-based parallel processing transputer system. Some numerical experiments are performed.

A parallel approach to the modified Numerov-like eigenvalue determination for sturm-Liouville problems

A modified Numerov-like eigenvalue algorithm, previously introduced, is parallelized. An inplementation of this algorithm on a Helios based parallel processing transputer system is discussed. Time savings with respect to a sequential approach are commented.

Parallel molecular dynamics of biomolecules

The basic principles of a typical sequential Molecular Dynamics (MD) program suitable for the study of solvated biomolecules are described, the inherent parallelism of MD is analysed and strategies for parallelisation are developed. Due to separate treatment of computation and communication a high level of portability is achieved and both tasks can be optimized independently. It is found out that communication trees are highly efficient means for sending, receiving and gathering the computed data, especially for large processor numbers. A current implementation on a transputer system is presented. Due to the tight memory budget slight modifications are necessary. Nevertheless, we get excellent performance with an average degree of parallelization of 82%.

Occam implementation of process-to-processor mapping on the Hathi-2 transputer system

This paper presents a polynomial-time Occam program for automatically mapping parallel programs onto multiprocessor systems. Based on the heuristic strategy of self-adjusting mapping, our program consists of grouping, placement, routing and self-adjusting procedures. Grouping groups the user-defined processes in a parallel program into target tasks with a possible load-balancing. Placement places the target tasks onto the processors in a transputer network. Routing produces edge-disjoint physical communication paths for the logical communication requirements among the placed tasks in the network. Self-adjusting adjusts first the placement scheme when the routing fails and then the grouping scheme when the necessary adjusting for placement is unable to make the routing succeed. These four procedures work co-operatively until a successful process-to-processor mapping has been finally achieved after a series of progressive self-adjustments. For the problem of mapping n processes in an arbitrary task graph onto m processors in a transputer network configured into a torus, the program needs time O(max{ n 2, m 5}) in the worst case on one processor under full adjusting. The time is reduced to O(max{ n 2, m 4}) if the adjusting heuristic is degraded into semi-adjusting, and to O(max{ n 2, m 2}) when the adjusting heuristic is completely eliminated. The latter result holds only for the transputer networks providing message routing and multiplexing. We demonstrate the implementation result and performance evaluation of the program on the Hathi-2 transputer system. The implementation shows that for both regular and irregular task graphs the program works very well and produces satisfactory results.

THE COMBINATION TECHNIQUE FOR THE SPARSE GRID SOLUTION OF PDE'S ON MULTIPROCESSOR MACHINES

We present a new method for the solution of partial differential equations. In contrast to the usual approach which needs in the 2-D case [Formula: see text] grid points, our combination technique works with only [Formula: see text] grid points, where hn denotes the employed grid size. The accuracy of the obtained solution deteriorates only slightly from [Formula: see text] to [Formula: see text] for a sufficiently smooth solution. Additionally, the new method is perfectly suited for parallelization. On a machine with [Formula: see text] processors we get in practice an overall parallel complexity of only [Formula: see text]. The method can be generalized to higher dimensions. Then, the gain is expected to be even more dramatic. For the two-dimensional case, we report the results of numerical experiments obtained on a Transputer system and on the CRAY Y-MP.

A domain splitting algorithm for parabolic problems

In the parallel implementation of solution methods for parabolic problems one has to find a proper balance between the parallel efficiency of a fully explicit scheme and the need for stability and accuracy which requires some degree of implicitness. As a compromise a domain splitting scheme is proposed which is locally implicit on slightly overlapping subdomains but propagates the corresponding boundary data by a simple explicit process. The analysis of this algorithm shows that it has satisfactory stability and approximation properties and can be effectively parallelized. These theoretical results are confirmed by numerical tests on a transputer system.

Adaptive Pole Placement Control Using Multiple Transputers

The paper details the implementation of an adaptive pole placement algorithm on a multiple transputer system. The controller makes use of the extended least squares parameter estimation algorithm and is controlled at the local level by a supervisor task which monitors the behaviour of the control system to ensure robust operation under a variety of environmental conditions. This enhances the integrity of the controller in real-time operations. The adaptive controller is also configured to be used as a node within a distributed control system such that a central network interface can configure a number of individual controllers within the distributed control network. The controller is demonstrated dealing with a plant with changing parameters and disturbances.

Self-adjusting Mapping: A Heuristic Mapping Algorithm for Mapping Parallel Programs on to Transputer Networks

The problem of mapping parallel programs on to multiprocessor systems is a fundamental problem of great significance in parallel processing. In this paper we propose a fast heuristic algorithm to solve this problem on transputer networks. Our mapping algorithm mainly contains three modules: grouping, placement and routeing, where grouping puts processes in the program into tasks which can be one-to-one placed on to processors in the transputer network, placement sets the grouped tasks on to the processors and routeing produces edge-disjoint physical communication paths for logical communication requirements. The algorithm works by combining three modules under a self-adjusting scheme towards a successful mapping result. For mapping n processes in an arbitrary parallel program on to m processors in a transputer network of grid structure, our algorithm has a worst-case time complexity O(max {n 1 , m 5 }) under full adjusting, O(max {n 2 , m 4 }) under semi-adjusting and O(max {n 2 , m 2 }) under no adjusting, where the last holds only for the transputer networks providing message routeing and multiplexing. The algorithm has been implemented in Occam on the Hathi-2 transputer system.