Performance Prediction of Distributed RAID Storage System

Since information is growing explosively, the requirements for high performance storage systems are critical in practical applications. The research on performance prediction has become increasingly important in the area of distributed storage systems. This paper presents the architecture of iSCSI-based distributed RAID storage system. At the basis of analyzing the system data flow, the stochastic Petri net model of the system is constructed. And then an analytical method for computing system I/O response time is proposed utilizing equivalent conversion of serial or parallel transitions. Theoretical analyses show that bandwidth is the bottleneck of distributed RAID storage system in low-speed network; cache hit rate and CPU processing power at central server are the main factors affecting the I/O response time in high-speed network. The performance of distributed RAID system can be improved through increasing the network bandwidth and the performance of the centre node.