Pattern Classi cation Using Hidden Markov Models

Input/output performance on current parallel le systems is sensitive to a good match of application access pattern to le system capabilities. Automatic input/output access classiication can determine application access patterns at execution time, guiding adaptive le system policies. In this paper we examine a new method for access pattern classiication that uses hidden Markov models, trained on access patterns from previous executions, to create a prob-abilistic model of input/output accesses. We compare this approach to a neural network classiication framework, presenting performance results from parallel and sequential benchmarks and applications.