Declustering Databases on Heterogeneous Disk Systems

摘要: Declustering is a well known strategy to achieve maximum I/O parallelism in multi-disk systems. Many declustering methods have been proposed for symmetrical disk systems, i.e., systems which all disks the same speed and capacity. This work deals with problem of adapting such heterogeneous environments. In environments these are many types servers large range speeds capacities. We deal first case perfectly declustered queries, queries retrieve fixed proportion answer from each disk. show that fraction dataset must be allocated affected by both relative capacity Furthermore, hierarchical structure most distributed where groups placed servers, imposes further complications due variations . server network bandwidths may affect actual achievable transfer rates. propose an algorithm determines loaded on The tailored find loading minimal response time given database size, or compute system profile showing optimal possible ranges sizes. Next we look at probabilistic aspects this how optimize expected retrieval when Proportions data retrieved axe random variables. rather surprising result optimality, not simply proportional its but some compensation made bias towards faster disks. here general can used conjunction symmetric methods.