Course Project Final Report Disk Access Response Prediction

We present a method for predicting disk access response times given a trace of previous disk activity using a linear regression model. We build two models and corresponding features to de- scribe difierent situations, the request-based model, in which we have all the information about the prior requests, and trace-based model, in which we predict the response time for a new trace of requests(workload). In addition, the fractal nature of the accesses is examined and a power-law established. This fractal distribution is then used to distinguish between idle and busy periods. In the trace-based model, where many ideal features are not available, a KNN(k-nearest neighbor) method is used to select a previously seen trace from the training data which is similar to the testing data. The accuracy of both models is assessed and conclusions drawn as to which features are most important to predicting response times, and why that might be.