The next 700 CPU power models

Abstract

Software power estimation of CPUs is a central concern for energy efficiency and resource management in data centers. Over the last few years, a dozen of ad hoc power models have been proposed to cope with the wide diversity and the growing complexity of modern CPU architectures. However, most of these CPU power models rely on a thorough expertise of the targeted architectures, thus leading to the design of hardware-specific solutions that can hardly be ported beyond the initial settings. In this article, we rather propose a novel toolkit that uses a configurable/interchangeable learning technique to automatically learn the power model of a CPU, independently of the features and the complexity it exhibits. In particular, our learning approach automatically explores the space of hardware performance counters made available by a given CPU to isolate the ones that are best correlated to the power consumption of the host, and then infers a power model from the selected counters. Based on a middleware toolkit devoted to the implementation of software-defined power meters, we implement the proposed approach to generate CPU power models for a wide diversity of CPU architectures (including Intel, ARM, and AMD processors), and using a large variety of both CPU and memoryintensive workloads. We show that the CPU power models generated by our middleware toolkit estimate the power consumption of the whole CPU or individual processes with an accuracy of 98.5% on average, thus competing with the state-of-the-art power models.

Taming Energy Consumption Variations In Systems Benchmarking

“…To study the energy consumption of nodes, we considered Intel Running Average Power Limit (RAPL) [16], which is one of the most accurate tools to report the CPU/DRAM global energy consumption. We also used PowerAPI [8], which is a power monitoring toolkit that builds a model over RAPL to compute the energy consumption at process-level when we needed to isolate energy consumption of a single process. Our clusters are provisioned with a minimal version of Debian 9 (4.9.0 kernel version) where we install Docker (version 18.09.5), which will be used to run the RAPL sensor and Figure 3: Comparing the variation of binary and Docker versions of aggregated LU, CG and EP benchmarks the benchmark itself.…”

A Comparison of Parallel Profiling Tools for Programs utilizing the FFT

“…The developers can therefore make a tradeoff on configurability by the user, for ease of use and can take advantage of special hardware features. As an example, at present, there is no standard interface to measure energy use or input and output utilization (although there are efforts towards this such as Power API [20], the Global Extensible Open Power Manager [14], Energy Aware Runtime systems [11] and PowerAPI [10]) but Craypat can provide this information to the user while the open source tools that can do this, TAU and Extrae, need to be configured to do so.…”

On Reducing the Energy Consumption of Software

“…(6) How do you describe perfect tooling that suits your coding requirements for green software design-you can go deep into technical details? (7) How do you think we should inform about energy software consumption for a better awareness? (8) Do you think that getting a better software energy consumption consideration is the responsibility of the developer or the company?…”

“…The last decade witnessed several attempts to consider green software design as a core development concern to improve the energy efficiency of software systems at large [2,3,18,23,26]. However, despite previous studies that have contributed to establish guidelines and tools to analyze and reduce the energy consumption [1,7,12,16,17,25,32], these contributions fail to be adopted by practitioners till date [14,28].…”