Performance analysis of media-based cache via analytical and simulation model

Abstract

Storage has been a performance bottleneck in IO intensive applications for over one decade. To alleviate it, many cost-effective designs of disk subsystems are proposed. Media-based cache (MBC) is one of the recent implementations. This paper addresses the performance model of MBC using analytical approaches and Monte Carlo techniques. In particular, we give a detailed model to capture the characteristics of the general distribution of service times (disk access times) of MBC and we use a modified queuing model of M/G/1/K with multiple vacations to formulate the MBC dynamics. The system throughput and blocking probability under different workload characteristics and system parameters are studied. Essentially, we illustrate how the system parameters affect the MBC performance in different typical scenarios, which may be used as design guidelines for disks.