With the close of the Adept Project this week comes the release of a new open-source suite of energy measurement tools, developed throughout the course of the research project.
Over its three-year lifespan, Adept has investigated energy consumption in parallel hardware and software. The Adept project brought together experts from high-performance computing (HPC) and embedded computing to develop a comprehensive suite of tools that enable the evaluation and optimisation of a system’s energy and power usage.
‘This is a significant step forward in understanding where exactly in a parallel computing system energy is consumed,’ says Dr Michèle Weiland, project coordinator for Adept. ‘Giving both hardware and software developers access to this type of information allows them to make informed choices about new implementations, without guesswork.’
Energy usage has become a primary concern of HPC users. As computers become more powerful, they inevitably consume more energy – unless the becomes more efficient.
Dr Andrew McCormick, director at Alpha Data Parallel Systems, says, ‘Alpha Data is pleased to have been part of Adept as it was a project that has brought energy efficiency in computing into the limelight. We contributed expertise in developing embedded FPGA and System-on-Chip hardware for multi-sensor systems to the project, developing a custom, robust and flexible data capture system for energy and power measurement as part of the project’s multi-platform experimental verification and test lab. The approaches of optimising code to match hardware architecture or selecting hardware to best match code align with an increase in heterogeneity in computing systems to maximise performance per watt.’
The Adept Project ran for three years from September 2013 to August 2016 and was funded under the European Seventh Framework Programme under Grant Agreement No 610490. The project was coordinated by EPCC, the supercomputing centre at the University of Edinburgh; the consortium partners were Uppsala University; Ericsson AB; Alpha Data Parallel Systems Ltd; and Ghent University.
Introducing the Adept Tool Suite
The Adept Tool Suite consists of three parts: a benchmark suite, power measurement infrastructure, and power and performance prediction tool. Both the benchmark suite and the designs for the power measurement infrastructure are publicly available and are open-source. More information on the Adept project tool set can be found on the porject’s website.
Adept Benchmark Suite: a range of tests, from single-instruction benchmarks all the way up to small applications. These are designed to characterise power and energy use on different systems under a variety of scenarios, in conjunction with external power measurement.
Adept Power Measurement Infrastructure: a hardware-based solution for measuring power in all aspects of a system over time. The fine-grained architecture is capable of taking one million samples per second from all components of a system.
Download the Power Measurement Infrastructure designs:
http://www.adept-project.eu/power-measurement-designs.html
Adept Power and Performance Prediction Tool: this tool brings together the software and hardware aspects of the project to predict how a given piece of software will perform on a given hardware configuration.
A more detailed breakdown of the project can be found on the Scientific Computing World website in an article detailing the project and its achievements entitled: ‘Exploring energy efficiency’.
