For many storage providers, the cost of providing service calls exceeds the costs of the hardware being serviced. In this paper, we show that zeromaintenance, small disk arrays are too expensive, but that low-maintenance arrays are feasible and describe a possible implementation. Our evaluation technique replaces Mean Time to Data Loss with the lifespan. Our results also show the impact of the assumption of constant failure rates on the results of modeling.
Jehan-François Pâris, Thomas J. E. Sc