Alinhamento de seqüências biológicas em arquiteturas com memória distribuída

AUTOR(ES)
DATA DE PUBLICAÇÃO

2005

RESUMO

The use of cluster of computers for solving problems that require a great quantity of computational resources is becoming an interesting alternative. Clusters are economically feasible and of easy maintenance, offering a computational power equivalent to that of supercomputers. However developing applications for this kind of architecture is complex because it involves issues that are not present in the sequential programming such as data communication and concurrent tasks synchronization, problems that usually are handled by specialized software packages in supercomputers. Considering this context, this work presents the development of a mechanism for supporting communication on clusters of computers focused on exploring this hardware platform for high performance processing. The mechanism was created as a library of functions written in C and it is based on the Active Messages model. Its implementation was performed on the applicative level, using light multiprogramming techniques as programming resource. In order to prove the need of such communication mechanism and to measure the performance of the implemented library, an application with real processing needs in clusters is presented and discussed: the alignment of biological sequences. In this specifc case, the needs of the sequences alignment reflect its computational cost and particularly its memory consumption. Besides the real application, the project and implementation of the communication mechanism considered also its introduction in the Anahy environment for high performance processing on clusters of computers. The evaluation of the communication library developed was achieved by implementing a prototype application for alignment of biological sequences using dynamic programming methods and an application for calculating the Fibonacci Number

ASSUNTO(S)

memória distribuída sequence alignment processamento de alto desempenho multithreading multiprogramação leve multiprogramação active aessages high performance processing distributed memory alinhamento de seqüências mensagens ativas sistemas de computacao

Documentos Relacionados