Adaptive Resource Management in Scalable Network – Attached Storage Systems 1 , 2

The growth in the commercial use of the Internet and the proliferation of dataintensive network services, have heightened the demand for large–scale storage systems. Over the last few years, Network–Attached Storage has emerged as a basic tenet for simplified storage management and improved scalability, reliability and performance of storage systems. We propose a novel technique for efficient resource management in such systems, with the objective of minimizing response time while maximizing throughput. The salient features of our approach are: (a) intelligent replication of data objects across multiple disks based on client access patterns, (b) load balancing among the disks using hashing, and (c) adaptability of the system with minimal overhead. Our approach is easy to implement in a decentralized manner. Further, we present simulation results of the proposed scheme using synthetic workloads. Our approach is shown to significantly outperform various alternatives with respect to response time and throughput as the load on the system increases, while creating only few replicas. Results show that our approach is adaptive when the number of disks change. The resulting system is scalable and self–configurable, and adapts to changes in the operating environment quickly.