Control of Energy Efficiency in Modern Data Centers
Pre-study Project (August, 2014 -September, 2015)
Energimyndigheten , Swedish Energy Agency
Luleå University of Technology, Luleå Sweden
ABB corporate research, Västerås, Sweden
The purpose of this 1 year pre-study project was to develop a roadmap for research and development activities aiming at the improvement of energy efficiency in datacenters. This topic is especially important in Norrbotten region that aims at becoming a globally attractive hosting area for datacenters.
Traditionally, improvements in the energy consumption of servers in data centers have been mainly sought to be achieved by scheduling of computational processes and workflows running in servers, which is done by the dedicated software. It has been also well understood that substantial energy gains can be achieved by better design of the utilization of heating, ventilation, and cooling systems (HVAC), which is currently done using traditional Building Management Systems (BMS) and Building Automation Systems (BAS). Locating servers in places with cooler climate (such as Norrbotten region) is a simple way of reducing spending on cooling, but it is believed that a lot of potential is hidden in improvements of BMS operation algorithms.
However, in current large data centers, cooling systems operate based on maximum capacity of fans and load of servers. In this approach excessive energy is used in case of changes in environment conditions, such as volatile computing load, changing weather conditions, or reduced power supply. A more flexible control of ventilation could bring energy savings.
During our preliminary research and literature review, we realized that a robust technique was required to dynamically model the cooling system, simulate its behavior and optimize its performance with regard to its energy consumption. The ultimate goal then is to propose adaptive and energy efficient cyber-physical architecture, to control the data center’s cooling system in order to reduce its unnecessary energy consumption and waste.
After a comprehensive investigation, we found out that modularity and object-oriented design of automation software can effectively increase the flexibility of the BAS systems. Moreover, as the contemporary BAS has distributed and heterogeneous nature, distributed intelligence has to be incorporated into the software solutions for BAS. Furthermore, In order to validate the efficiency of distributed control, a thermal model of a server room is very important and crucial for creating energy efficient cooling systems. Finally a simulation environment which can incorporate all of the aforementioned concepts is an essential part of the solution. Therefore, our solution comprises of the following modules:
- Modular Function Block (FB) architecture defined under IEC 61499, as an empowering technology to provide modularity of automation software and maintain the heterogeneity of the BAS cooling systems.
- Utilizing Multi-agent systems as an enabler technology for applying distributed artificial intelligence.
- Utilizing Formal modeling techniques in order to create real-time and accurate thermal model of the server room cooling system.
- An adaptive simulation environment for examining and evaluating different control strategies in order to select the most energy efficient strategy at any given period of system’s (cooling system) execution