Improving the Performance of Batch-Driven Distributed Systems Using Genetic Algorithms, Artificial Neural Networks and Bayesian Networks

Large-scale distributed systems used to run a long, predictable series of operations on a recurring basis (batchdriven systems) are a prime target for optimization to reduce resource usage or time consumption, since even a small improvement may yield large dividends over time. Machine learning techniques are a good way to approach this optimization problem since there are typically many variables that interact in unknown ways. This paper discusses a practical approach to optimizing an existing realworld system using an artificial neural network for simulation, a genetic algorithm for exploration, and a Bayesian network for human presentation and generalization.