Enhancing Data Availability through Background Activities

Latent sector errors in disk drives affect only a few data sectors and are often not detected till the affected data is accessed again. They may cause data loss if the storage system is operating under reduced redundancy, because of previous failures. In this paper, we evaluate effectiveness of two known techniques to detect and/or recover from latent sector errors, namely scrubbing and intra-disk data redundancy. These two techniques are treated as background activities that complete without affecting the otherwise normal operation of the storage system. We focus on how disk idle times can be managed to effectively complete these background tasks without affecting foreground task performance, while reducing the window of vulnerability for data loss. We show via detailed trace-driven simulations that scheduling policies for background jobs that are based on careful monitoring of the stochastic characteristics of idle times in disk drive, have a minimal effect on foreground task performance while dramatically improving storage system reliability.