Energy Optimization by Using Last Level Cache Partitioning for Multicore Platforms

Abstract

Last Level Cache (LLC) partitioning has been ex-tensively examined to optimize energy and improve performance. Cache partitioning strategies proposed to reduce inter-core in-terference for the shared cache in multicore platforms. However, these approaches start focusing on performance in multicores and energy consummation. Current energy-efficient solutions in this field minimize power consumption by employing dynamic voltage and frequency scaling (DVFS). This paper has proposed an algorithm for improvements to existing models for energy consummation in multicore platforms. The proposed algorithm enhances the energy consummation resulting from DVFS by using LLC partitioning. The algorithm consists of two phases. The first phase sets the level of DVFS to the task, while the second phase gets the optimum partition factor of LLC to this task. Extensive simulations illustrate the efficiency and influence of the proposed method on the total energy consumption of multicore platforms by each stage of the suggested algorithm. Comparison by DVFS only and DVFS with cache partitioning show that the proposed algorithm achieves 15% improvements in energy-saving and reduces 25% L2 cache misses rate.