Dynamic Provisioning of Storage Workloads

Due to lack of generic, accurate, dynamic and comprehensive models for performance estimation, customers typically tend to underprovision or over-provision storage systems today. With multi-tenancy, virtualization, scale and unified storage becoming norms in the industry, it is highly desirable to strike an optimum balance between utilization and performance. However, performance prediction for enterprise storage systems is a tricky problem, considering that there are multiple hardware and software layers cascaded in complex ways that affect behavior of the system. Configuration factors such as CPU, cache size, RAM size, capacity, storage backend (HDD/Flash) and network cards etc. are known to have significant effect on the number of IOPS that can be pushed to the system. However, apart from system characteristics as these, storage workloads vary reasonably and therefore, IOPS numbers depend heavily on types of workloads provisioned on storage systems. In this work, we treat storage system as a hybrid of black-box and white-box models, and propose a solution that will enable administrators to make decisions in the presence of multiple workloads dynamically. Our worst-case prediction is within 15% error margin.