Energy-Efficient and Energy-Aware Computing
Minimising the the energy consumption of computing
Highly parallel architectures such as graphics processing units (GPUs), clusters and multi-core systems have gained significant prominence in several domains of science where accelerated computation is vital.
Although modern systems are increasingly energy efficient compared to earlier systems, their increased utilisation leads to increased overall energy consumption. Carbon emissions, constantly increasing energy prices and recent changes in emissions legislations are some of the reasons that drive our desire to minimise the overall energy consumption of computing.
A common belief is that faster programs should consume less energy and thus are energy efficient. Our initial investigations show that this is not necessarily true.
The aim of this research is to investigate routes for performing energy efficient computing. Primarily, we are interested in energy-efficient and energy-aware algorithms which are yet efficient algorithms (performance-wise). With this in focus, we are working on the following:
- Understanding the energy consumption of software / algorithms
- Establishing metrics for assessing the energy efficiency of algorithms/software
- Creating models for estimating the energy consumption of software / algorithms
- Developing benchmark suite for assessing systems for their energy efficiency (EPACK)
- Developing a software framework (EMPACK) for energy profiling and optimisation of applications.
The EMPACK is under development and it has already enabled us in understanding the energy consumption of software applications. EMPACK will enable energy profiling of applications and should enable static energy/runtime optimisations. However, EMPACK will also facilitate feedback-directed self optimisation (both energy and runtime).
Systems Under Test
We utilise all modern multi-processor/many-core systems to all modern GPUs to evaluate our benchmarks, applications and algorithms. We will be publishing these findings in this page very soon.
MRSC 2011, Bristol, UK, Towards Developing Energy-Aware Algorithms, Jeyan Thiyagalingam and Anne Trefethen