Developing a radically new thermal-aware design approach for next generation energy-efficient datacenters
David Atienza ~ Project Website
Efficient management of huge volumes of data is a key consideration for data centers. However, with the advent of big data, data management has acquired a new dimension. The requirement for data centers has multiplied several times over, bringing a stream of new challenges. One of those is to ensure optimum use of energy.
But that’s easier said than done. The “power” factor comprises a major chunk of the outgoings at many data centers, making the entire model unsustainable in the long run. According to The Economist, data centers and knowledge management systems comprise 50% of the economic value in developed countries. That calls for heavy power consumption across industries.
This dilemma is the main driver for the YINS initiative. In developing the YINS approach, the researchers have included a consortium of six world-renowned academic partners from EPFL and ETHZ, three industrial partners in Switzerland (Credit Suisse, Eaton and BrainServe), and three third-party industrial partners (Constellium, Friotherm, and Osmoblue).
The main objective of YINS is to optimize energy consumption at data centers at various levels: the chip level, which includes the fundamental principles of computer operations and technology; the system level comprising servers, racks, and IT infrastructure; and the cooling level, where processes are implemented to reduce the heating-up process of computers in operation.
At the chip level, YINS has successfully developed a new ultra-adaptive multi-core server that ensures optimization of energy use, saving up to 50% of energy used by a standard server. At the server level, the proposed architecture (termed Scale-Out Non-Uniform Memory Architecture or SO_NUMA) improves the memory placement to make the interconnected servers operate efficiently with minimum energy consumption. With low network latency, there is practically no energy consumption when the systems are idle. At the cooling level, the YINS approach doesn’t rely on traditional cooling systems based on air or fans, but on an active cooling system called thermosyphon, which uses condensation. This is a critical aspect of energy management because 50% of energy in a typical data center is spent in cooling servers down.
Thus, YINS has successfully developed a suite of tools that can optimize energy consumption at each operational level, ultimately resulting in cost savings for data centers.