Repairability Enhancement of Scalable Systems with Locally Shared Spares

Abstract

Future systems based on nano-scale devices will provide great potentials for scaling up in system complexity, yet they will be highly susceptible to operational faults. While spare units can be generally used to enhance reliability, they must be shared in a limited way among functional units to ensure low-cost overheads when systems scale up. Furthermore, the efficiency of achieving reliability using spare units heavily depends on the replacement mechanisms of such spares. While global and chained replacement approaches can take advantage of the entire replacement capabilities in the network, they usually impose some sort of disturbance to all the functional units in the system during the repair process, thus are dreadfully expensive in terms of performance overhead for systems with high fault rates. In this paper, we focus on a low-cost, fast, "immediate" replacement mechanism that can be implemented locally with minimum disturbance to the system. The proposed schemes aim for maintaining the system with high fault rates in such a lowcost, fast repairable status for many faults before invoking the more expensive, yet optimal, approaches. First, we propose an online repair algorithm: as faults occur during the run-time of the system, the proposed algorithm makes a choice of a spare unit (among several candidates), such that the overall impact on system repairability in the future is minimized. Second, we propose a network enhancement approach, which identifies and connects the vulnerable units to the exploitable spares, thus strengthening the entire system at a low cost.