Biological sequence analysis on distributed-shared memory multiprocessors

In this paper we present a task-level implementation of several biological sequence analysis algorithms on distributed-shared memory multiprocessors. All the studied algorithms are computationally expensive and its computational patterns ranges from regular to very irregularly structured ones. An initial approach to evaluate the regularity level for this kind of algorithms based on the rate of free-dependent tasks, the data access pattern and the task homogeneity, is presented in this study. Extensive tests have been performed over different architectures using message-passing and thread-based programming models. A comparative analysis centered on performance and portability is also presented. The results give us a general knowledge about software portability and performance for algorithms on this area