A Threshold - based Strategy for DynamicRecon guration of RAID Systems *

Emerging applications make increasing use of shared and distributed data. Workload adaptive storage based on RAID (Redundant Arrays of Independent Disks) is currently the most suitable technology to fulll storage needs and to manage multi-source access with variable characteristics in terms of arrival frequency, mean size and access type (read/write). However, workload adaptive storage suuers from a high data migration cost. The Dynamic Redundancy method improves on workload adaptive storage by reducing data migration signiicantly, and even suppressing it in some cases. This document aids the practical realisation of Dynamic Redundancy by presenting a strategy for identifying the optimal storage layout for a given workload and the thresholds at which the current storage layout should be transformed. This strategy takes account of hardware parameters, as well as the global data access proole of currently executing applications. Simulation results reveal two kinds of Dynamic Reconnguration, one driven by the mean request size and one driven by the percentage of write requests. We present a table of lower bound thresholds showing when each kind of reconnguration should be employed.