Green Queue: Customized Large-Scale Clock Frequency Scaling

Abstract

Abstract-We examine the scalability of a set of techniques related to Dynamic Voltage-Frequency Scaling (DVFS) on HPC systems to reduce the energy consumption of scientific applications through an application-aware analysis and runtime framework, Green Queue. Green Queue supports making CPU clock frequency changes in response to intra-node and internode observations about application behavior. Our intra-node approach reduces CPU clock frequencies and therefore power consumption while CPUs lacks computational work due to inefficient data movement. Our inter-node approach reduces clock frequencies for MPI ranks that lack computational work.We investigate these techniques on a set of large scientific applications on 1024 cores of Gordon, an Intel Sandybridgebased supercomputer at the San Diego Supercomputer Center. Our optimal intra-node technique showed an average measured energy savings of 10.6% and a maximum of 21.0% over regular application runs. Our optimal inter-node technique showed an average 17.4% and a maximum of 31.7% energy savings.

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

“…In Adagio, Rountree et al [25] present a runtime system that accurately predict slack. Tiwari et al, in Green Queue [6], combine inter-node DVFS approach and intra-node DVFS approach and scale the DVFS technique to 1024 cores. We focus on CPU clock modulation technique, which has low power switching overhead when used in fine granularity.…”

“…One scenario of applying DVFS takes advantage of the work-load imbalance (slack) in MPI applications. Such systems include Jitter [24], Adagio [25], and Green Queue [6]. Kappiah et al [24] use a runtime system Jitter to predict the appropriate clock frequencies for upcoming iterations based on observations of previous iterations.…”

“…Ge et al divided applications into fixed intervals and determined a suitable frequency for each interval by using performance counters [1]. Other options include applying DVFS during communication phases or non-critical execution paths of MPI applications to lower power consumption without affecting the performance [2], [3], [4], [5], [6]. All of these approaches apply DVFS for coarse-grain phases of the application.…”

Metrics for Evaluating Energy Saving Techniques for Resilient HPC Systems

“…Large scale studies of failure rates [2] and the causes of these failures have been performed. Energy consumption is also a well studied area, with several energy saving runtime techniques [3]- [8] having been proposed and implemented. Energy saving techniques such as DVFS [9] and clock throttling [10] have been used to provide power state governors for operating systems.…”

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

“…This method is prone to making incorrect decisions for computations with properties that are not specifically accounted for in the construction of microbenchmarks. Green Queue [23] uses sophisticated binary analysis and tracing tools to demarcate computational phases in an application. Based on modeled power and measured performance, each phase is assigned a static frequency to run on.…”