A hybrid approach based on regression tree and radial-based function network for dynamic storage device performance prediction

Abstract

Storage device performance prediction is a critical element of self-managed storage systems and application planning tasks, such as data assignment and configuration. We proposed a new hybrid method (RT-RBF), which combines regression tree (RT) and radial-based functions network(RBF), to model storage device performance. In our proposed algorithm, the RT is firstly used to split the large space of knowledge into several small width and disjoint sub-spaces, and an RBF network is then used for training each of these smaller sub-spaces. With this new method, the advantages of the two techniques are completely amalgamated to obtain a more accurate and incremental model without compromising prediction time. In addition, we consider the caching effect as a feature in workload characteristics. Experiments indicate that RT-RBF model as well as workload characteristics used in the storage device modeling can produce more accurate predictions than RT or RBF.