Towards a reliable and efficient distributed storage system

Abstract

This paper presents RDSS, a Resource Area Network (RAN)-based Distributed Storage System, which is designed for high scalability, long-term reliability, and operational efficiency. Using the RAN architectural pattern, RDSS works on the block level instead of the file level, which increases system flexibility. RDSS adopts directory routing model, lazy master-slave replication, and content hash verification. It employs a Node Ranking System (NRS) to optimize node selection in terms of stability, integrity, and workload. SPIN simulation results show that RDSS can significantly suppress data transfer of misbehaving nodes while yielding only a slight deviation in terms of individual load share compared to the random selection. Microbenchmarks show that RDSS retrieves 64KB data within 35ms and updates 64KB data within 40ms. The contribution of RDSS is that it provides reliable storage services while achieving high efficiency.