Efficient Disk Drive Performance Model for Realistic Workloads

Accurate models for hard disk drive performance are needed to ensure that computer systems, both large and small, can meet the performance requirements for their intended application. There are a few tools available for predicting disk drive performance when one only has access to the published drive specifications, which is often the case when there is a need to select a drive from multiple candidates. There are several existing models for predicting hard disk performance, but they either require a large number of often inaccessible parameters, or they do not provide sufficient accuracy. In this paper, we present a hard drive performance model that requires only 23 input parameters and, on average, deviates from benchmarks less than 3.5% for random workloads, 9% for sequential workloads, and 9% for mixed workloads across multiple drive products. The proposed model's accuracy and ease of use can facilitate rapid and efficient decision making for information technology professionals.