Yasuyuki Mimatsu

A performance comparison between the DR-net and a hierarchical RAID system

When the number of disks rises in disk array systems that contain multiple disk drives, system performance is limited by a bottleneck at a centralized controller and/or at a communication path that uses a bus. We evaluate, through simulation, a scalable architecture called DR-net, in which the controller functions are distributed to all disk drives and each disk has autonomy in processing its tasks. DR-net provided high throughput in proportion to the number of disks. In a conventional system, the influence of the bus setup time and concentration of the parity calculation load caused the throughput to saturate. We also show that DR-net can take advantage of disk autonomy when reconstructing data stored in failed disks. Our results indicate that the distribution of functions and the autonomy of disk drives enable better scalability and more effective utilization of system resources than with a hierarchical system.