Measurement laboratory manual - Power measurement bench manual

Problem statement
Objectives
Scope and Constraints

Problem statement

The increasing environmental impact of information and communication technology (ICT) systems has created a growing need for methods that enable the systematic measurement and optimization of energy consumption throughout the software lifecycle. While hardware energy efficiency has improved considerably over time, software increasingly influences the overall energy usage of computing systems through resource utilization, deployment architectures, and operating environments. However, reliably measuring the energy impact of software remains a challenging problem.

Unlike traditional performance metrics such as execution time or memory usage, software energy consumption cannot be observed directly. Software itself does not consume energy; instead, it controls hardware components whose operation results in electrical power consumption. Consequently, the energy effects of software must be inferred indirectly through measurements of the underlying hardware. This creates challenges in attributing observed energy consumption to specific software components, workloads, or system interactions.

The problem is further complicated by the diversity of modern ICT systems. Software may execute on heterogeneous hardware platforms ranging from mobile devices and embedded systems to desktop computers and servers. These devices employ different power delivery architectures, including AC-powered, DC-powered, and battery-powered systems, and may expose varying levels of built-in energy monitoring capabilities. Similarly, modern software systems encompass a wide range of application types, execution environments, and deployment models, including cloud-native services, distributed systems, web applications, and locally executed software.

Although numerous commercial power and energy measurement instruments are available, selecting suitable measurement technologies is difficult due to trade-offs involving accuracy, sampling frequency, bandwidth, cost, automation capabilities, and compatibility with different systems under test. Existing measurement solutions are often specialized for particular device classes or operating environments, limiting the comparability and reproducibility of measurement results across heterogeneous systems.

Furthermore, practical integration of energy measurements into software engineering workflows remains limited. Many measurement approaches require specialized equipment, proprietary tools, or extensive expertise in electronics and instrumentation, creating barriers to the adoption of energy-aware software development practices. As a result, software developers and researchers often lack accessible, reproducible, and cost-effective infrastructures for systematically evaluating and improving the energy efficiency of ICT systems.

To address these challenges, this work investigates measurement technologies, measurement processes, and laboratory designs for ICT energy measurements. The objective is to establish a practical, extensible, and economically viable energy measurement laboratory capable of supporting diverse hardware platforms, software systems, and deployment environments. By providing standardized measurement methodologies and reusable infrastructure, the laboratory aims to enable reproducible energy measurements and facilitate the integration of energy efficiency considerations into software development, research, and continuous optimization processes.

Objectives

1. Explore the solution space for measuring the energy consumption of software and hardware

The laboratory aims to systematically explore and characterize the solution space for measuring the energy consumption of modern ICT systems. Energy measurement in computing is inherently challenging due to the diversity of software execution environments, hardware architectures, and operating conditions. Consequently, the laboratory shall investigate novel and existing measurement approaches suitable for a wide range of systems.

From the software perspective, measurements shall cover multiple application classes, including backend services, frontend applications, serverless workloads, distributed systems, and locally executed software. The laboratory shall also support different execution environments such as different operating systems and native binaries, managed runtimes, browser-based applications, and scripting languages. Furthermore, measurement methodologies shall be developed for various interaction models, including command-line tools, graphical user interfaces, web APIs, and long-running system services.

From the hardware perspective, the laboratory shall accommodate diverse device classes, ranging from server-grade systems and desktop computers to laptops, single-board computers, smartphones, and tablets. Measurement capabilities shall consider different power delivery models, including battery-powered devices as well as AC- and DC-powered systems. In addition, the impact of networking technologies, including both wired and wireless communication, shall be incorporated into the measurement framework.

This objective is explored here.

2. Design an abstract and generalizable measurement process

A key objective is to develop an abstract measurement process capable of supporting a broad spectrum of energy measurement scenarios. The process shall define standardized procedures for configuring experiments, controlling workloads, collecting measurements, and analyzing results in a reproducible manner.

The measurement process should be largely hardware-agnostic and extensible, allowing new devices and instruments to be integrated with minimal changes to the overall framework. The primary limitations of the process should arise only from the available measurement hardware and instrumentation. Such abstraction enables comparable experiments across heterogeneous platforms and facilitates reproducible energy research across different computing environments.

A generic framework for the measurement process is presented here.

3. Conduct a systematic study of measurement technologies and instrumentation

The laboratory design shall be informed by a comprehensive study of existing power and energy measurement technologies. This includes the theoretical foundations of electrical measurements, such as accuracy, precision, sampling frequency, synchronization, calibration, and uncertainty analysis.

The study shall cover a wide range of instrumentation, including experimental research devices, commercial measurement equipment, and ICT devices equipped with built-in energy monitoring capabilities. Particular attention shall be given to understanding the strengths and limitations of different measurement techniques, such as external power measurement, battery telemetry, embedded sensors, and on-chip energy counters. The resulting knowledge base shall guide the selection of appropriate measurement methods for different use cases.

The various kinds of measurement technologies are explored here.

4. Systematically map commercially available power and energy meters

Another objective is to perform a systematic survey of commercially available power and energy measurement devices suitable for laboratory use. The survey shall include dedicated power analyzers and energy meters, programmable power supplies with integrated metering functionality, and digital measurement components that can serve as the basis for custom open-source measurement platforms.

To facilitate wide scale adoption of measurement practices, particular emphasis shall be placed on economically viable solutions that provide adequate accuracy, factory calibration, and accessible digital interfaces. Devices exposing high-level communication interfaces such as USB, Ethernet, serial interfaces, or standardized protocols are especially valuable due to their ease of integration and automation. The study shall identify instruments that balance cost, performance, scalability, and ease of deployment in research environments.

We have composed a curated table of commercial energy meters here.

5. Systematically map measurement software and integration capabilities

The laboratory shall investigate and catalog available software solutions for interfacing with power and energy measurement devices. Modern energy laboratories rely heavily on software to automate experiments, configure instruments, collect measurement data, and synchronize measurements across multiple devices.

The study shall evaluate existing libraries, frameworks, and protocols for communicating with diverse meters using commodity hardware. Consideration shall be given to interoperability, platform support, data formats, automation capabilities, and long-term maintainability. The resulting software architecture should enable the integration of heterogeneous instruments into a unified measurement platform and support reproducible, automated experimentation.

The software solutions for energy measurement are discussed here.

6. Produce a comprehensive laboratory design for ICT energy measurements

The final objective is to produce a complete design document for an ICT energy measurement laboratory. The design shall specify an economically viable set of commercial power and energy meters, associated measurement software, and supporting infrastructure necessary for conducting energy experiments across a wide range of computing devices.

The laboratory design shall define representative measurement configurations for different classes of hardware, including servers, personal computers, embedded systems, and mobile devices. It shall also support diverse networking topologies and experimental setups involving wired and wireless communication. The resulting laboratory should provide a flexible, extensible, and reproducible environment for conducting energy efficiency research and benchmarking the energy consumption of software and hardware systems.

The high level design of the laboratory is available here.

Scope and Constraints

The design of the ICT energy measurement laboratory is guided by practical constraints, research objectives, and the need to maximize applicability across a broad range of hardware and software systems. Rather than attempting to replicate every possible real-world deployment environment, the laboratory is intended to provide a representative and extensible measurement infrastructure that supports a wide variety of experimental scenarios.

Representative rather than exhaustive environments

Modern ICT systems operate in highly diverse environments, ranging from embedded devices and mobile platforms to cloud infrastructure and large-scale data centers. Replicating all such environments within a single laboratory would be economically infeasible and would significantly increase the complexity of the measurement infrastructure.

Instead, the laboratory focuses on constructing representative setups that can emulate or simulate different classes of operating environments. This approach enables meaningful energy measurements while maintaining flexibility and cost efficiency. The laboratory shall be designed to allow future extensions and reconfiguration if research interests shift toward specific hardware classes or deployment scenarios. The overall scope of the laboratory is ultimately constrained by the available budget and physical resources.

Focus on open technologies and reproducibility

The laboratory emphasizes the use of free and open-source software (FOSS) and widely adopted deployment technologies. Containerization and orchestration platforms such as Docker and Kubernetes are preferred due to their portability, reproducibility, and broad industry adoption. Nowadays, common deployment technologies are often platform-agnostic, which further mitigates the potential negative effects of this restriction.

Prioritizing open technologies improves the generalizability of research results and reduces dependencies on proprietary ecosystems. Closed-source software and vendor-specific solutions may introduce limitations related to licensing, interoperability, and long-term availability, which can negatively affect the reproducibility and applicability of energy measurement methodologies. Consequently, support for proprietary platforms is considered secondary unless required by a specific use case.

Budgetary constraints

Economic viability is a key design criterion for the laboratory. The total budget allocated for the laboratory infrastructure was limited to less than 6000 euros. A significant portion of the equipment was acquired prior to the Visiiri project, allowing the available resources to be utilized efficiently.

To ensure scalability and reproducibility, the cost of individual devices should generally remain below 500 euros, with a preferred target of less than 100 euros for smaller components and measurement devices. This cost-conscious approach facilitates replication of the laboratory by research institutions and industrial partners with limited resources, especially in the early stages of adopting measurement practices.

Safety and construction constraints

The laboratory shall be designed to be constructable by researchers and engineers with moderate electronics expertise without requiring specialized certifications or licenses for electrical work. User-accessible electrical interfaces should primarily operate at extra-low voltages below 24 V AC/DC whenever possible.

Mains-powered systems shall be limited to standard 230 V / 10 A continuous power operation. The design should avoid unnecessary hazards to human health and rely primarily on off-the-shelf commercial components to simplify deployment, maintenance, and replacement. The rack setup with a good quality rack mounted PDU (power distribution unit) is critical for this goal as cheap extension cords often do not handle large current continuous loads.

Although the laboratory is designed with safety in mind, work involving power electronics inherently carries risks. These include potential fire hazards, electrical faults, and damage to measurement devices or systems under test due to physical access to sensitive electrical interfaces. Appropriate safety procedures and protective equipment should therefore be employed during laboratory operation.

Scope of software and firmware development

The scope of the laboratory extends beyond hardware instrumentation and includes the development and deployment of custom software components. This includes measurement software, data collection frameworks, automation scripts, custom firmware, and operating system modifications necessary for accurate and reproducible energy measurements.

Such flexibility is particularly important for integrating heterogeneous devices and enabling advanced measurement techniques that are not supported by existing commercial solutions.

Modularity and extensibility

The laboratory architecture shall emphasize modularity to accommodate varying research and industrial use cases. Whenever possible, measurement devices and supporting infrastructure should be deployable independently, allowing users to construct subsets of the laboratory tailored to specific requirements. We also applied this philosophy of modularity on the whole concept of a measurement laboratory. The lab is physically implemented as a rack setup on wheels. This has the practical benefit of being movable inside the premises, which is useful as the total weight of the setup is over 100 kg and may require extensive preparations for a specific new physical configuration.

For example, a partner organization interested solely in measuring a single-board computer with a DC power input may deploy only the relevant DC measurement components without requiring AC power instrumentation or networking infrastructure. This modular approach reduces costs and simplifies deployment for specialized applications.

Compatibility between meters and systems under test

The selection of measurement devices shall prioritize broad compatibility with diverse systems under test. ICT devices vary significantly in operating voltages, current consumption, and transient power behavior. Some devices operate at low voltages but exhibit large peak current draws, which may exceed the capabilities of certain measurement instruments. Then again, higher power measurement instruments may offer lower accuracy and sample rate, when combined with low-power devices.

Consequently, the laboratory shall prioritize general-purpose measurement devices that maximize compatibility across different hardware classes while remaining within budget constraints. Selecting versatile instrumentation often provides better cost efficiency and broader applicability than highly specialized devices.

Exclusion of niche measurement devices

At this stage, the laboratory focuses on measurement devices with broad applicability rather than highly specialized instrumentation. Devices designed exclusively for narrow use cases may provide excellent performance within their target domain but contribute little to the overall flexibility of the laboratory.

For example, USB-specific power meters are highly useful for measuring USB-powered devices but are restricted to a limited set of applications. More general-purpose low-voltage meters capable of measuring a wider range of extra-low-voltage devices—typically below 25 V AC/DC—provide greater utility within the laboratory context. While USB-C is gaining popularity as the standard interface for many low power devices, during the development of the laboratory, only a limited selection of high power USB-C PD meters were available, and since many of the devices using the interface were battery-powered, this also posed challenges for the measurements.

Similarly, some ICT systems, such as high-performance GPUs, AI accelerators, and server-class hardware, can exhibit extremely high power consumption. Supporting such devices may significantly increase costs and restrict the selection of compatible measurement equipment. Therefore, the current laboratory design primarily targets low- and medium-power devices. Existing mid-tier GPUs acquired from earlier research projects may be included when suitable, but dedicated support for high-power computing infrastructure is currently outside the scope of the laboratory.