A New Approach to Disk-Based Mass Storage Systems

We present a new approach to create large-scale cost-effective mass storage systems using high-capacity disks. The architecture is optimized across multiple dimensions to support streaming and large-block data access. Because I/O requests in these applications are to a small fraction of data, we power-cycle drives as needed to create a very high-density storage system. Data protection is provided through a new variant of RAID, termed power-managed RAID™, which generates parity without requiring all drives in the RAID set to be powered on. To increase bandwidth, the system employs several concurrency mechanisms including load balancing I/O streams across many parallel RAID controllers. A number of optimizations are made to the interconnection architecture and caching schemes that lowers the cost of storage to that of typical automated tape library systems while exploiting the performance and reliability advantages of disk systems. Recent results at COPAN Systems have proven the viability of this new storage category and product. 1. Background Growing reference information, regulatory data and rich media archives [1] are providing ne