Choosing from redundant designs of power systems using system outage rate and cost

The push for high availability is on. As customers keep telling computer manufacturers, any outage to their operations is unacceptable. For those involved at the design stage, this usually means that functions within a computer's subsystems, like power, must be made fault-tolerant, or redundant. In its simplest form, redundancy replaces one component with two, so that the first failure will have no impact on the customer. More generally, a K-out-of-N design calls for N components to do the job of N − K + 1, so that the first K − 1 failures can be tolerated but the K-th failure causes a system outage. In this paper, we relate our own experience with power subsystems: how we chose from among different shades of redundancy by concentrating on outage rate and cost. To further simulate real life, we also considered the service strategy, i.e. whether (and when) to preventively replace a component whose failure did not cause the system to go down. A key result of the paper shows that a redundant design coupled with a well-chosen service strategy yields advantages beyond what either one alone would provide. The combination not only decreases the design's already low system outage rate, it makes the outage rate essentially constant—so that ease of predictability is ensured. Three power subsystem designs are outlined, as are the reliability and outage rate functions associated with them. Cost and service strategy are then considered, to help decide which is the optimal design based on total program cost projections and reliability objectives.