A Graph Based Framework for the Manipulation of Sparse and Irregular Distributed Data-structures 1
نویسنده
چکیده
R esum e : Les applications r eelles utilisent des structures de donn ees classiques telles que des matrices, mais fournissent la plupart du temps une impl ementation sp eciique au probl eme, comme Compressed Sparse Column (CSC) { voir par exemple SPARSKITT17]. Ceci est en particulier vrai quand on travaille avec de grandes structures de donn ees creuses, comme des matrices provenant du domaine des el ements niss19]. La dii erence entre l'impl ementation et la structure de donn ees qu'elle impl emente eeectivement est encore plus importante quand on consid ere les applications data-parallel dans lesquelles les structures de donn ees sont distribu ees sur un r eseau de processeurs. Dans ce cadre, il y a un besoin agrant d'outils permettant au d eveloppeur de visualiser ses donn ees, leur structure, et les op erations qui leurs sont appliqu ees, quelle que soit la faa con dont elles sont cod ees et distribu ees. De tels outils doivent proposer des repr esentations de haut niveau, c'est a dire faire abstraction de l'impl ementation physique des donn ees pour atteindre l'image que s'en fait le d eveloppeur. Ils doivent int egrer la s emantique des applications tout en fournissant des m ecanismes permettant de synth etiser ou de ltrer les informations aan de pouvoir se concentrer sur un aspect particulier du probl eme. Dans cet article, nous pr esentons les bases d'un mod ele que nous avons mis en place pour permettre une telle abstraction, tout en prenant en compte les probl emes d'eecacit e, a la fois en terme de m emoire et de vitesse. Un prototype utilisant ce mod ele a et e impl ement e dans l'outil graphique Visitt7] qui fait partie du projet HPFIT 1 7, 8, 9]. HPFIT est un projet qui implique 3 equipes de recherche situ ees a Bordeaux au LaBRI, a Lyon au LIP, et a Bonn (Allemagne) a GMD/SCAI. Son objectif est de proposer un environnement de d eveloppement HPF int egr e, supportant les structures de donn ees creuses et irr eguli eres. Abstract: Real world applications use well-known data structures such as matrices, but most of the time provide a speciic problem-oriented implementation, e.g. Compressed Sparse Column (CSC) { see for instance SPARSKITT17]. This is particularly true when dealing with large sparse data-structures, such as matrices coming from the domain of nite elementss19]. The gap between the implementation …
منابع مشابه
Expressing Massively Parallel Algorithms for Irregular Data Expressing Massively Parallel Algorithms for Irregular Data
In this paper we present a data-parallel language called MPL, for which the basic data structure is the graph. The purpose of this language is to program numerical algorithms that use irregular data, and to allow them to execute on distributed memory MIMD machines. We deene the graph data structure and give some examples of how irregular data structures can be described from graphs with our app...
متن کاملA Graph-Based Framework for the Definition of Tools Dealing with Sparse and Irregular Distributed Data Structures
Industrial applications use specific problem-oriented implementations of large sparse and irregular data structures. Hence there is a need for tools that make it possible for developers to visualize their applications in terms of these data, their structure, and the operations that are applied to them, whatever their effective implementation and their distribution are. In this paper, we present...
متن کاملSpeech enhancement based on hidden Markov model using sparse code shrinkage
This paper presents a new hidden Markov model-based (HMM-based) speech enhancement framework based on the independent component analysis (ICA). We propose analytical procedures for training clean speech and noise models by the Baum re-estimation algorithm and present a Maximum a posterior (MAP) estimator based on Laplace-Gaussian (for clean speech and noise respectively) combination in the HMM ...
متن کاملFace Recognition using an Affine Sparse Coding approach
Sparse coding is an unsupervised method which learns a set of over-complete bases to represent data such as image and video. Sparse coding has increasing attraction for image classification applications in recent years. But in the cases where we have some similar images from different classes, such as face recognition applications, different images may be classified into the same class, and hen...
متن کاملParallelizing Irregular Applications through the YAPPA Compilation Framework
Modern High Performance Computing (HPC) clusters are composed of hundred of nodes integrating multicore processors with advanced cache hierarchies. These systems can reach several petaflops of peak performance, but are optimized for floating point intensive applications, and regular, localizable data structures. The network interconnection of these systems is optimized for bulk, synchronous tra...
متن کاملA note on vague graphs
In this paper, we introduce the notions of product vague graph, balanced product vague graph, irregularity and total irregularity of any irregular vague graphs and some results are presented. Also, density and balanced irregular vague graphs are discussed and some of their properties are established. Finally we give an application of vague digraphs.
متن کامل