Storage systems today are expected to deliver consistently high quality of service, despite facing constant demands to store increasingly large quantities of data. The technology pull to massive storage systems has led to the development and widespread adoption of virtualised storage infrastructures that incorporate intelligent storage fabrics. The physical resources underlying such systems are organised into storage tiers, each of which delivers a different cost per capacity ratio. An effective performance model is vital to ensure performance and reliability demands of these systems can be fullled. Such a model should abstract the physical features of the underlying disk drives and disk arrays, reecting faithfully the structure of a virtualised storage system. Typically, a storage tier will consist of multiple RAID subsystems. Our approach is to initially develop a performance model for an individual disk drive. This can be extended to a RAID model, for various RAID levels, using existing or enhanced techniques. We are specically looking at RAID levels 0, 0-1, 5 and 6, which are most commonly used. Our goal is to develop a hierarchical, multi-class queueing network performance model of the storage tier and virtualised storage system from these initial model components. Quality of service constraints can then be optimised at no extra cost by applying discerning device selection and data placement strategies across the tiers. We hope to autonomously and transparently migrate data across tiers and organise data within tiers according to performance benets. This will contribute towards an overall project ambition of attaining near-optimal performance and reliability over the data lifecycle.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.