Welcome to this Cost 804 focus group. The topic of this group is how to use different parallel algorithms and implementations to provide the same function, but with varying performance and energy cost.
The different models (algorithm and implementation) researched aim to offer different solutions in the pareto front for the multi-objective optimization problem of energy-efficient performance.
Several tracks have been identified. The models could exploit the possibilities offered by the nature of parallelism in the target computing environment. Examples are the memory organization in GPU co-processors (different memories, variable cache sizes), the communication network in distributed systems, the operating system behavior (scheduler, DVFS). The parallelism can originate from more than one computing element, but from a collection of elements (CPU and GPU, clusters, etc.).
The structure of programs also influences the energy spent, as it influences performance. The decomposition of the program (for example: its granularity), the nature of interaction between components (for example: IPC) play a key role in the aforementioned objectives. The influence of program structure on energy relates to the amenability of the resulting programs to an efficient scheduling onto the resources (for example: software pipelines).
Deliverables can come in the form of best practices, analysis and experiments, and in more formal models of computation.