Energy-efficient computer system for large scale computing

Financing Narodowe Centrum Nauki under the Opus programme.

The project is carried out Zespół Złożonych Systemów IAIS, Politechnika Warszawska.

The amount of the grant awarded:  320 200 PLN.
Implementation period: 2016–2019
The result of the research carried out under the project is a computer system including energy-saving algorithms for task scheduling, resource allocation and server work regulators in the Linux kernel. Current analyses indicate that the load of application servers in the cloud computing is at the level of 10-50% of their capacity. Maintenance of computing power reserves is associated with the need to provide a guarantee of quality and resilience of services to sudden changes in the speed of tasks inflow. Such a control policy enforces excessive power consumption and thus contributes to increased carbon dioxide emissions to the atmosphere. Energy demand is also one of the dominant constraints on the rate of development of HPC systems worldwide. The aim of the project is to gain new knowledge about the dynamics and random nature of computing processes in application server clusters and to develop a proposal to solve the problem of energy-efficient control of data center. The innovative character of the proposed research results from the use of currently available tools (hardware and software) to identify models of computational processes and to apply the methodology of designing control systems for complex systems. The essence of the project is a comprehensive approach, in which all mechanisms to reduce energy demand are considered. The research is interdisciplinary in nature. The tools of control theory and optimization, machine learning, statistical methods and computer simulation are used. Due to the scale of complexity of the problem, new tools of theoretical analysis are developed.