Revisiting LP-NUCA Energy Consumption

• DOI: 10.1145/2632217
• Publication Date: 2014-07-01T14:23:02Z
• Authors: Darío Suárez Gracia; Alexandra Ferrerón; Luis Montesano De...; Teresa Monreal Arnal; Víctor Viñals Yúfera
• Keywords: Locality of reference; Working set; Cache pollution

Cache working-set adaptation is key as embedded systems move to multiprocessor and Simultaneous Multithreaded Architectures (SMT) because interthread pollution harms system performance battery life. Light-Power NUCA (LP-NUCA) a adaptive cache that depends on temporal-locality save energy. This work identifies the sources of energy waste in LP-NUCAs: parallel access tag data arrays tiles low locality phases with useless block migration. To counteract both issues, we prove switching serial reduces without harming propose machine learning Adaptive Drop Rate (ADR) controller minimizes amount replacement migration when low. demonstrates these techniques efficiently adapt drop policies They reduce LP-NUCA consumption 22.7% for 1SMT. With contention 2SMT, savings rise 29%. Versus conventional organization, energy--delay improves 20.8% 25% 1- 2SMT benchmarks, and, 65% mixes, gains are larger than 20%.