Instruction Multiple Data Machines Research Articles

A Simulation Tool for the Performance Evaluation of a Class of Mimd Systems

A user-friendly, menu-driven simulation tool for the performance evaluation of a class of Multiple Instruction Multiple Data (MIMD) systems with private caches is presented in this paper. This tool can be efficiently utilized in the preliminary stages of designing such systems. The class of MIMD machines modelled consists of processors, private caches, and a set of interleaved memory modules. Each processor has a local private cache which is connected through a cross-bar interconnection network to shared Main Memory modules. Specifications of elements can be denned individually or globally to accommodate various homogeneous/heterogeneous system configurations. The tool can be used to configure, model, and gimnljfcto the MIMD computer system and its performance. It uses a set of metrics to indicate the model's performance by estimating specified performance indices of the system and its components that include relative speedup, throughput, and utilization as well as others such as average queue length and average waiting time. Simulation runs were performed to validate and assess the tool's functionality.

Steps towards steady-state simulation on MIMD machines: solving almost block diagonal linear systems

We study, in this paper, the feasibility of using preconditioned iterative methods to solve linear systems on MIMD (multiple instruction multiple data) machines when solving steady-state simulation problems. We show that, provided a good parallel preconditioner is available, the proposals in this paper give very good parallel performance. Our proposals are directly applicable to almost block diagonal linear systems, but can be extended to more general flowsheeting problems (Paloschi, 1996). This will be the subject of a future paper. A preconditioning strategy blending an incomplete LU decomposition with a polynomial preconditioner is proposed and tested with the iterative method GMRES of Saad and Schultz (1986). A parallel implementation is tested on two MIMD machines, a Fujitsu AP1000 with 128 processors and a cluster of workstations with PVM (Parallel Virtual Machine) (Geist et al., 1994). Two examples are used, a simple model of a distillation column using Antoine coefficients and the discretization of a PDE. The second example allows the variation in the size of the problem and results are presented for up to 65 536 equations on the AP1000 and 262 144 equations on the cluster. Almost perfect efficiency is observed when solving a problem of size 65 536 using 32 processors on the AP1000 (and size 16 384 using 16). Very good speed-ups also are obtained on the cluster of workstations (a speed-up of 15.3 using 16 processors) when solving a problem of size 65 536. The results on the cluster are presented for different load conditions on the network.

A Parallel Quicksort Algorithm

An optimal parallel version of quicksort, pquicksort, is presented and analyzed. The algorithm is designed for a p-processor MIMD (multiple instruction, multiple data) machine with shared memory and is an exclusive read, exclusive write algorithm. It is shown that in almost all permutations of N values (with probability 1 - o(1/ N)) pquicksort will sort in O( N lg N/ p) for p ≤ N/lg N.

Computational Problems in Orthopaedic Biomechanics

AbstractTotal joint replacements such as the hip and knee result in composite structures. Proper design of these devices and selection of particular devices requires understanding of the structural response of the resulting composite structure of plastic, metal, interface, and human bone. The problem is challenging because of complex geometry, non-homogeneous material properties (dense cortical bone and spongy cancellous bone), and time varying material properties (bone remodeling). Bone is a active material which responds over time to applied load, changing the density and Young's modulus.This paper addresses several computational issues. Remote sensed data such as computerized tomography scans are used to create finite element models. Nonhomogeneous material properties are represented in finite element models using shape functions. This results in a unique material property for each Gauss point. Finally, stochastic finite elements methods are used to model material variation and uncertainty. Both Monte Carlo and First Order Second Moment methods are used. These problems are particularly well suited for high-level parallelization schemes on multiple instruction multiple data machines.

Independent partitioning of algorithms with uniform dependencies

Uniform dependence algorithms with arbitrary index sets are considered, and two computationally inexpensive methods to find their independent partitions are proposed. Each method has advantages over the other one for certain kinds of applications, and they both outperform previously proposed approaches in terms of computational complexity and/or optimality. Also, lower and upper bounds are given for the cardinality of maximal independent partitions. In multiple instruction multiple data (MIMD) systems, if different blocks of an independent partition are assigned to different processors, communications between processors will be minimized to zero. This is significant because the communications usually dominate the overhead in MIMD machines. >

Classical and Quantum Simulations for Large Systems on Parallel Computers

ABSTRACTMolecular dynamics (MD) method is used to investigate structural transformation and the loss of intermediate range order in SiO2 glass at very large positive pressures and the modification of SiO2 glass network at very large negative pressures. The nature of molecular vibrations in solid C60 has been studied with tight binding molecular dynamics (TBMD) method. Implementations of simulation algorithms on parallel computers are also discussed.In a-SiO2 at high pressures, the height of the first sharp diffraction peak in S(q) is considerably diminished and its position shifts to larger wave vectors. At twice the normal density, Si-O bond length increases, Si-O coordination changes from 4 to 6, and O-Si-O band-angle changes from 109° to 90°. This is clearly a tetrahedral to octahedral transformation, which is observed recently by Meade, Hemley, and Mao in their diffraction experiments using synchrotron radiation.MD simulations of porous silica are carried out in the density range 2.2 - 0.1 g/cm3 Internal surface area, pore surface-to-volume ratio, gyration radius, and fractal dimension are studied as a function of density. Simulations are in good agreement with the experimental results obtained by x-ray scattering. The results reveal a crossover of the structural correlations between short- to intermediate-range (< 8 Å) and fractal- to large-scale-regime (10 ~ 100 Å).Dispersion and density of states (DOS) of inter- and intra-molecular phonons are calculated for orientationally ordered and disordered solid C60 using the TBMD method. Inter-molecular phonon DOS extends up to 7.6 meV and shows libron peaks at 2.4 meV and 3.7 meV in the orientationally ordered phase. Orientational disorder softens libron modes. Intra-molecular phonons below 70 meV also show significant dispersion. Our results are in good agreement with the recent inelastic neutron scattering experiments.MD is a numerical approach which involves the solution of Newton's equations for particles in the system. The multiple-time-step (MTS) approach reduces this computation significantly by exploiting the different time scales for short-range and intermediate-range interactions. Using the linked-list scheme, parallel algorithms are designed to implement on the in-house 8-node iPSC/860, a MIMD (multiple instruction multiple data) machine. Our group has also developed a quantum dynamical simulation scheme for Computational Nanoelectronics based on the quantum molecular dynamics (QMD) method. The QMD algorithm has been implemented on the in-house 8,192-node MasPar, a SIMD (single instruction multiple data) architecture.

Highly parallel computers: Perspectives

Highly parallel machines represent a technology capable of providing superior performance for technical and commercial computing applications. By utilizing technical advances in microsystems, such as the high performance RISC processors, and high speed interconnect, together with progress in algorithms and compilers, massively parallel systems will exceed the limits of performance being approached by traditional vector supercomputers. Prototypes have been built at IBM Research over the past 10 years to test various approaches: YSE, the Yorktown Simulation Engine, was a high performance, special purpose machine for logic simulation. Today, descendants of this machine are in use at nearly every IBM Development Laboratory; the RP3, Research Parallel Processor Prototype, was an innovative shared-memory multi-processor of the multiple instruction multiple data (MIMD) type; GF11, is an enhanced single instruction multiple data (SIMD) machine that currently achieves the highest sustained performance on key applications; and Victor-V256, a distributed memory, fixed-grid interconnect MIMD machine with 256 processors. In this paper we present our views on the influence of personal computing and telecommunication technologies on parallel processing. We will describe where the architecture of future parallel computers is heading and identify several network and hardware design issues which affect parallel machine implementations. We will also describe some of the parallel machines recently built at the IBM T. J. Watson Research Center, and conclude by projecting the evolving use of parallel computing and the likely route parallel computing will take into the 21st century.

Mapping electromagnetic field computations to parallel processors

An introduction to parallel processing for the electromagnetic analyst is presented. Scaled speedup is explained, and the use of SIMD (single instruction, multiple data) and MIMD (multiple instruction, multiple data) machines is discussed. Methods of electromagnetic field computation (method of moments, finite-difference time domain, finite-element method, spectral iteration) are broadly surveyed to indicate which characteristics make them amenable to parallel processing and which architectures are best suited to those characteristics. >

Reconfigurable transputer networks: practical concurrent computation

The architecture of a reconfigurable multitransputer machine capable of a gigaflop (10 9 floating point operations per second) performance will be described in some detail. Problems in efficiently coding such a distributed memory MIMD (multiple instruction multiple data) machine are also discussed. These problems will be illustrated with reference to some practical strategies for exploiting different types of parallelism present in many scientific and engineering problems. Some examples will be discussed in detail.

Very-high-performance multiple-instruction multiple-data applications

The MIMD (multiple instruction multiple data) class of multiprocessors offers, in principle, the capability to bring the resources of an arbitrary number of processors to a problem. This confers the dual advantages of very high performance and of faulttolerance. A number of successful applications will be presented, from which some general conclusions are drawn on the practical feasibility of the use of MIMD machines.

Models for MIMD Machines

This paper describes two simple models for estimating the processing times of algorithms run on MIMD (Multiple Instruction Multiple Data) machines. The first model is developed from very elementary queueing concepts. The second uses the critical path through a PERT-diagram-arrangement of the algorithm's tasks. The models require far less effort than either benchmarking or detailed simulations. We envision them being used in all but the very final stages of the search for, and design of, good combinations of algorithms and MIMD machines.