ProgramaÃÃo paralela de alto nÃvel e eficiente sobre arquiteturas distribuÃdas

AUTOR(ES)
DATA DE PUBLICAÇÃO

2003

RESUMO

New technological paradigms have emerged in high-performance computing since last decade. The consolidation of distributed architectures and the recent advances in the state-of-the-art performance of processor and interconnection networks gave birth to clusters, conventional computer networks that can compete with supercomputers in its niche of applications, but with costs of an order of maginitude smaller. Recently, the advances in the state-of-the-art of interconnection of wide area networks interconnection technologies have suggested the use of them as virtually infinite scalable supercomputers, called grids. In the last years, great research efforts have been spent world wide to make possible the use of internet for super-computing applications, with promising results. Cluster and Grid computing are now key concepts in high-performance computing, dictating its further directions. Besides widening the number of potential users, a myriad of new applications have been induced in this area, extrapolating merely scientific computing, traditionally the most important niche of applications for high-performance computing. However, these new applications have been characterised by higher structural complexity and scale, requiring higher level tools to support their development. The parallel programming paradigm is the essential tool for the development of applications that take advantage of performance of the clusters and grids. However, its inherent complexity and the lack of eÂcient higher level tools for the development of applications claim for efforts to be conducted providing more advanced models for developing effcient parallel programs, support their validation by formal methods and adapted to the modern software engineering techniques. The # parallel programming model, a product of this thesis, emerges as an alternative to the conventional mechanisms to support development of parallel programs over distributed architectures. Targeted at emerging cluster and grid technologies. The # model has solid foundations in formal models based on Petri nets, allowing property analysis and performance evaluation of parallel programs, using existing tools based on this formalism. The Haskell# language has been implemented, adhering the # model, but using Haskell to describe sequential computations. The use of Haskell allows to make transparent the separation between coordination and computation media of # programs

ASSUNTO(S)

parallel programming functional programming redes de petri ciencia da computacao processamento de alto desempenho high performance computing petri nets programaÃÃo funcional programaÃÃo paralela software engineering engenharia de programas

ACESSO AO ARTIGO

Documentos Relacionados