Dynamic Power Management Technique for Multicore Based Embedded Mobile Devices

Abstract

As the proliferation of ubiquitous computing environments becomes a reality, the need for high speed data processing and intelligent system management increases rapidly. In particular, the need for low-power designs and power-aware system management is getting stronger. While multicore systems are deployed in many embedded system areas, an effective power management technique for multicores is not available yet. In this paper, we propose a novel power management technique based on a parallel programming model. OpenMP is a well-known programming paradigm for shared memory multicore systems. OpenMP is based on library routines for parallel processing. By identifying the invoked library routines, how many cores will be adequate for a certain application can be determined, and the number of necessary cores for a given task can be determined during run-time. By turning off unnecessary cores, we can reduce power consumption. We implemented this method by adding capabilities in an OpenMP-compliant compiler and conducted experiments with various benchmarks. We were able to reduce the power consumption by 18% on average compared to other conventional power management methods.Index Terms-Dynamic power management, low-power design, multicore, open multiprocessing (OpenMP).

Runtime Performance and Power Optimization of Parallel Disparity Estimation on Many-Core Platforms

“…DCT selects the number of concurrent processing cores and threads during run-time to manage application parallelism and exchange performance for energy [Porterfield et al 2013;. Both DVFS and DCT control have been used in conjunction as run-time control approaches to achieve minimized energy consumption and a required performance target [Curtis-Maury et al 2008;Hwang and Chung 2013]. These approaches are based on offline training to learn the system architecture followed by online performance prediction to guide run-time optimization and adaptation.…”

“…Firstly, existing approaches [Porterfield et al 2013] and [Curtis-Maury et al 2008] ignore energy minimization in the sequential part of the application, which can be significant. Secondly, these approaches [Curtis-Maury et al 2008;Hwang and Chung 2013] use offline training processes to learn the system architecture and control DVFS and/or DCT. As a result, their models are limited to single use-cases and their scalability is poor for different many-core architectural allocations of the same application.…”

Guest Editorial Special Section on Embedded and Reconfigurable Systems

“…A FFT-based digital receiver for Radar is used as application example, illustrating the benefits of the proposed methodology that takes advantage of strong data locality in the processing nodes providing better utilization of the underlying FPGA architecture resources. Power-aware system management is also the topic of the second paper, entitled "Dynamic Power Management Technique for Multi-Core Based Embedded Mobile Devices" [6], by Y.-S. Hwang and K.-S. Chung. The emergence of ubiquitous computing technologies supported by battery-operated embedded mobile devices with limited energy capacity motivates the presented work.…”

Adaptive Energy Minimization of OpenMP Parallel Applications on Many-Core Systems

“…The training is then followed by online performance prediction as a function of the system configuration and events to guide the runtime optimization and adaptation. Among others, Hwang and Chung [8] showed another runtime energy minimization approach considering joint DVFS and DCT control. Their approach is facilitated through statistical offline learning and implemented using runtime code insertion.…”

“…Firstly, existing approaches [5]- [7] ignore energy minimization in the sequential computation part, which constitutes a major performance component in many-core applications. Secondly, these approaches [7], [8] employ DVFS and/or DCT using offline training processes to learn the system architecture and control parameters. As a result, the scalability is poor for applications with different many-core architectural allocations.…”