Efficient speed (ES): Adaptive DVFS and clock modulation for energy efficiency

Abstract

Meeting the 20MW power envelope sought for exascale is one of the greatest challenges in designing those class of systems. Addressing this challenge requires over-provisioned and dynamically reconfigurable system with fine-grained control on power and speed of the individual cores. In this paper, we present EfficientSpeed (ES), a library that improves energy efficiency in scientific computing by carefully selecting the speed of the processor. The run-time component of ES adjusts the speed of the processor (via DVFS and clock modulation) dynamically while preserving the desired level of the performance. These adjustments are based on online performance and energy measurements, user-selected policies that dictate the aggressiveness of adjustments, and user-defined performance requirements. Our results quantify the best energy savings that can be achieved by controlling the speed of the processor, with today's technology, at the cost of negligible performance degradation. We then demonstrate that ES is effective in automatically calibrating the speed of execution in real applications, saving energy and meeting the desired performance goal. We evaluate ES on GAMESS, an ab initio quantum chemistry package. We show that ES respects the stipulated 5% performance loss bound and achieves 16% decrease in energy required to complete the execution while running with a power draw that is 18% lower.

Using Per-Loop CPU Clock Modulation for Energy Efficiency in OpenMP Applications

“…They also apply the same technique in point-to-point communications [3] and collective communications [4]. Cicotti et al [30] present Efficient Speed (ES), a library and runtime that control the speed of processor while minimizing the performance impact. They achieved 16% energy decrease with less than 5% performance loss for MPI applications.…”

Performance and energy analysis of OpenMP runtime systems with dense linear algebra algorithms

“…Zhang et al (2009) use duty cycling to efficiently manage on-chip shared resources. Bhalachandra et al (2015) employed this technique in MPI applications, while Cicotti et al (2014) designed a library to save energy with duty cycling and DVFS. Wang et al (2015) used DCT and duty cycling on OpenMP parallel loops.…”

Towards fine-grained dynamic tuning of HPC applications on modern multi-core architectures

“…A selection of MPI+OpenMP applications from the NAS parallel benchmark suite as well as two AMG based proxy applications from the ASC Sequoia benchmark suite were chosen to evaluate their approach. Other runtime libraries, such as the one by Cicotti et al [5], were also evaluated on proxy applications for energy efficiency. In contrast to these approaches, we use dynamic tuning of core and uncore frequencies, as well as threads to find the best energy-performance tradeoff in a real-world HPC application.…”