Skip to Main Content
Numerous redundant disk organizations have been proposed and used to provide increased performance and reliability from the I/O subsystems architecture, and once a disk fails in such a system, different forms of spar ing and reconstruction have also been proposed. In this paper, we offer a comprehensive evaluation of the relation ship between various disk array architectures, various disk reconstruction strategies, and their reliability under various failure rate and repair rate distributions. Specifically, we perform the evaluations of the reliabilities (through a mea sure of system mean time to failure) of different redundant disk organizations with variations in each of the following orthogonal directions. We also consider the effect of the reconstruction strategy on the reliability. A third and im portant dimension to our study involves the use of realistic failure rates and repair rates of disks that are dependent on the workload and the organization of the disks. Finally, we perform our study of scalability of the results to varying sys tem sizes, by specifically addressing disk organizations with 16 to 10&4 disks. The paper's contribution is in presenting a uniform framework for evaluation of these multidimen sional studies, as well as offering an improved model for predicting disk array reliability taking into account system load as well as disk organization.
Parallel Processing, 1993. ICPP 1993. International Conference on (Volume:1 )
Date of Conference: 16-20 Aug. 1993