STEP: Self-Tuning Energy-safe Predictors James Larkby-Lahet, Ganesh Santhanakrishnan, Ahmed Amer, and Panos K. Chrysanthis   View Paper (PDF)   Return to Session 4. Caching and Replication Abstract Data access prediction has been proposed as a mechanism to overcome latency lag, and more recently as a means of conserving energy in mobile systems. We present a fully adaptive predictor, that can optimize itself for any arbitrary workload, while simultaneously offering simple adjustment of goals between energy conservation and latency reduction. Our algorithm. STEP, achieves power savings on mobile computers by eliminating more data fetches, which would otherwise have caused excess energy to be consumed in accessing local storage devices or using the wireless interface to fetch remote data. We have demonstrated our algorithm to perform as well as some of the best access predictors, while incurring almost none of the associated increase in I/O workloads typical of their use. Our algorithm reduced average response times by approximately 50% compared to an LRU cache, while requiring less than half the I/O operations that traditional predictors would require to achieve the same performance, thereby incurring no energy penalty. ©2006 Association for Computing Machinery