A Reference Model for Integrated Energy and Power Management of HPC Systems

Optimizing a computer for highest performance dictates the efficient use of its limited resources. For High Performance Computing (HPC) systems, today, the major limiting resources are energy and power. Efforts to improve energy and power efficiency of HPC systems and the infrastructure of HPC centers achieve perpetual advances. In isolation, these efforts are unable to cope with the rising energy and power demands of large scale systems. A systematic way to integrate multiple optimization strategies, which build on complementary, interacting hardware and software systems is missing.
This work provides a reference model for integrated energy and power management of HPC systems: the Open Integrated Energy and Power (OIEP) reference model. The goal is to enable the implementation, setup, and maintenance of modular system-wide energy and power management solutions. The proposed model goes beyond current practices, which focus on individual HPC centers or implementations, in that it allows to universally describe any hierarchical energy and power management systems with a multitude of requirements. A principal aspect is how the individual components are connected, interface in a hierarchical manner and thus can optimize for the global policy, pursued as a computing center’s operating strategy.
For future research in energy and power management of HPC system, the OIEP reference model forms a cornerstone to realize — plan, develop and integrate — innovative energy and power management solutions.
The OIEP reference model represents a basis for holistic efficient optimization.