A Taxonomy and Survey on Energy-Aware Scientific Workflows Scheduling in Large-Scale Heterogeneous Architecture

Power and energy consumption are the primary concerns for large-scale heterogeneous computing systems such as HPC systems and clouds. Power is an identified limiter in the viability and sustainability of exascale systems. To achieve this, we need to improve energy efficiency at all levels of the HPC ecosystem. Scientific workflows are well-known computing models for executing computational and data-intensive workloads on parallel and distributed systems. Energy consumption of scientific workflows on the heterogeneous computing platform is of paramount concern. The workflow management system needs to consider various multiobjective optimization parameters while scheduling and executing scientific workflows. There is a need for a comprehensive and efficient energy-aware workflows runtime system to incorporate energy-aware mechanisms at all levels. This paper has outlined scientific workflows and discussed the distinct challenges in the path of its energy-aware execution. We have discussed the multiobjective optimization problem and presented a survey on the state-of-the-art workflow scheduling algorithms. We have also outlined the need for energy-aware runtime systems and proposed a reference architecture and runtime for energy-aware scientific workflows.