Energy-efficient phase-based cache tuning for multimedia applications in embedded systems

Abstract

The proliferation of multimedia applications in embedded systems has led to a research focus on optimizing the energy consumption of these applications without significantly degrading the execution time and adhering to data processing deadline constraints. To maximize optimization potential, phasebased tuning methodologies specialize system configurations to different phases of application execution with respect to design constraints. Multimedia applications are ideal candidates for phase-based tuning since these applications exhibit variable execution characteristics. In this paper, we propose a phasebased tuning methodology for multimedia applications that leverages application characteristics to determine the best cache configurations for different phases of execution. Results reveal that phase-based tuning for multimedia applications determines cache configurations within 1% of the optimal on average and yields an average energy delay product savings of 29%.

Scheduling and Tuning for Low Energy in Heterogeneous and Configurable Multicore Systems

“…However, for systems with hard real-time constraints, SaT must utilize a policy that prioritizes performance as well as energy. We reiterate that, in general, SaT achieved substantial energy savings as compared to the base configuration and prior work, and, for future work, we intend to explore techniques for improving SaT's performance such as phase-based tuning [43], and/or incorporating quality-of-service monitoring policy [41] into the decision-making of SaT, in order to make SaT suitable for performance centric systems.…”

“…Since SaT profiled the applications during runtime, SaT executed the applications on different cache configurations with unknown performance expectations, and the unknown configurations could degrade performance for the instruction cache more than data cache. One method to alleviate unknown performance impact while tuning is to use phase-based tuning [43], and/or incorporating quality-of-service threshold policy [40] into the decision-making of SaT, in order to make SaT suitable for performance centric systems. We plan to investigate these possibilities in future work.…”

Evolutionary Design of the Memory Subsystem

“…Adegbija and Gordon [14] designed a phase-based cache tuning algorithm for multimedia applications to determine the best cache configuration for each execution phase. Phase classification breaks applications execution using a fixed tuning interval.…”

Quality of Service-Aware, Scalable Cache Tuning Algorithm in Consumer-based Embedded Devices

“…For a quad core system, this tuner imposes 322 cycles overhead [1], which negligible considering the number of cycles to execute 1 million instructions. Prior work shows that this tuning interval is long enough to warm up and stabilize the caches [2].…”

“…We assume the performance threshold can be relayed to the tuner by the operating system using memory mapped I/O, special instructions, etc. depending on the implementation details [2].…”