The sparse matrix-vector multiplication is an important computational kernel, but is hard to efficiently execute even in the sequential case. The problems--namely low arithmetic intensity, inefficient cache use, and limited memory bandwidth--are magnified as the core count on shared-memory parallel architectures increases. Existing techniques are discussed in detail, and categorized chiefly based on their distribution types. Based on this, new parallelization techniques are proposed. The theoretical scalability and memory usage of the various strategies are analyzed, and experiments on multiple NUMA architectures confirm the validity of the results. One of the newly proposed methods attains the best average result in experiments on a large set of matrices. In one of the experiments it obtains a parallel efficiency of 90 percent, while on average it performs close to 60 percent.

We describe DDE-BIFTOOL, a Matlab package for numerical bifurcation analysis of systems of delay differential equations with several fixed, discrete delays. The package implements continuation of steady state solutions and periodic solutions and their stability analysis. It also computes and continues steady state fold and Hopf bifurcations and, from the latter, it can switch to the emanating branch of periodic solutions. We describe the numerical methods upon which the package is based and illustrate its usage and capabilities through analysing three examples: two models of coupled neurons with delayed feedback and a model of two oscillators coupled with delay.

This paper describes a new method for the suppression of noise in images via the wavelet transform. The method relies on two measures. The first is a classic measure of smoothness of the image and is based on an approximation of the local Holder exponent via the wavelet coefficients. The second, novel measure takes into account geometrical constraints, which are generally valid for natural images. The smoothness measure and the constraints are combined in a Bayesian probabilistic formulation, and are implemented as a Markov random field (MRF) image model. The manipulation of the wavelet coefficients is consequently based on the obtained probabilities. A comparison of quantitative and qualitative results for test images demonstrates the improved noise suppression performance with respect to previous wavelet-based image denoising methods.

We examine the parallel execution of a class of stochastic algorithms called Markov chain Monte-Carlo (MCMC) algorithms. We focus on MCMC algorithms in the context of image processing, using Markov random field models. Our parallelisation approach is based on several, concurrently running, instances of the same stochastic algorithm that deal with the whole data set. Firstly we show that the speed-up of the parallel algorithm is limited because of the statistical properties of the MCMC algorithm. We examine coupled MCMC as a remedy for this problem. Secondly, we exploit the parallel execution to monitor the convergence of the stochastic algorithms in a statistically reliable manner. This new convergence measure for MCMC algorithms performs well, and is an improvement on known convergence measures. We also link our findings with recent work in the statistical theory of MCMC.

Abstract In this paper the performance of the Intel iPSC/2 hypercube multiprocessor is analyzed. Computation and communication performance for a number of benchmarks are presented. We derive some fundamental performance parameters of the machine. Further, we investigate the difference between several communication schemes. Using the results of our measurements we can highlight some features and peculiarities in the iPSC/2 hardware and software. Where possible we make a comparison with the iPSC/1 and Ncube hypercubes.