Disk arrays with multiple RAID levels

Abstract

We propose Heterogeneous Disk Arrays (HDAs), which allow multiple RAID levels for database applications to coexist in a single disk array accommodating multiple RAID levels. Our main concern is to efficiently utilize disk bandwidth and capacity, while balancing disk loads in a cloud storage environment, however, a small number of disks is considered in this study for illustrative purposes. Individual RAID levels are adjusted to data availability requirements and workload demands. Adopting the most stringent availability requirements for all datasets would incur unnecessarily high bandwidth overhead for updating datasets, which do not have this requirement. Intermixing RAID levels is beneficial from the viewpoint of balancing disk loads, similarly to the striping paradigm in RAID5. The suitability of the RAID levels varies with database applications: RAID5 - reading/writing large datasets for data mining and warehousing, RAID1 -high performance OLTP applications. Several single pass data allocation methods are proposed in this paper and compared using synthetically generated allocation requests in associated papers.