Stochastic modeling for performance and availability evaluation of hybrid storage systems

Improvements in computational systems may be constrained by the efficiency of storage drives. Therefore, replacing hard disk drives (HDD) with solid-state drives (SSD) may also be an effective way to improve system performance, but the higher cost per gigabyte and reduced lifetime of SSDs hinder a thorough replacement. To mitigate these issues, several architectures have been conceived based on hybrid storage systems, but performance and availability models have not been proposed to better assess such different architectures. This paper presents an approach based on stochastic models (i.e., stochastic Petri nets and reliability block diagrams) for performance and availability evaluation of hybrid storage systems. The proposed models can represent write and read operations, and they may estimate response time, throughput and availability. A case study based on OpenStack Swift platform is adopted to demonstrate the feasibility of the proposed approach.