Data Replication and Delay Balancing in Heterogeneous Disk Systems

Declustering and replication are well known techniques used to improve response time of queries in parallel disk environments. As data replication incurs a penalty for updates, database designers face the problem of finding which part of the database to load on each disk and how these parts should be replicated. This problem becomes more complicated in heterogeneous environments where disks have different speeds and capacities as intuitively faster disks should be assigned a greater portion of the database to balance the delay in the system. We present analytic results and heuristics providing “near” optimal solutions to the combined problem of finding declustering proportions and replication schemes in such environments. Our model takes into account the characteristics of the disks such as speeds and capacities, as well as, access patterns to the data. Simulation results for various environments comparing different replication strategies are also provided.