Dynamic Storage Cache Partitioning Using Feedback Control Theory

In this paper, we propose a new quality-of-service (QoS) aware storage cache partitioning scheme that dynamically partitions cache space amongst simultaneously running I/O-intensive applications. The QoS specification is given in terms of latency of data access. Apart from data access latency which is perceived by the user, there can be more constraints put forth from the system administrators point of view. One of the more important constraints is the disk utilization level in a storage system. Normally, higher disk utilization levels are preferred by the system administrators for better resource consolidation. In this paper, we focus on these two constraints, namely, data access latency and disk utilization, which are translated into an overall storage cache hit rate requirement. We employ feedback control theory to achieve the required hit rate target per application. Our experimental results indicate that the proposed storage cache partitioning scheme is able to meet the required storage cache hit rate targets and improve overall storage system performance.