Energy efficiency & Large Scale computing

“Advanced immersion cooling strategies for edge-computing”

Traditionally, Cloud Computing has been developed in large data centers that collect and process all information in a centralized manner. Until recently it was a viable strategy, but the increase in demand for IoT and Smart Cities applications makes it almost impossible to continue using only this type of infrastructures in terms of latency and bandwidth.

From this situation arises a new paradigm, Edge Computing. In opposition to traditional infrastructures, this computing paradigm aims to distribute the data for processing at the edges of the network, instead of centralizing all the processing in a single megastructure. So, Edge computing is based on the deployment of numerous smaller data centers, which are located closer to the data sources in both urban and rural areas. Edge data centers support users while minimizing the volume of data sent to the Cloud. The use of this architecture has great advantages: it avoids the saturation of the Cloud, achieving a decongestion of the network and decreases the latency associated with Cloud Computing. Other advantages it presents is the greater security it offers as it is a distributed system.

On the other hand, Edge Computing data centers have specific restrictions since they need to be deployed efficiently near the data sources in urban areas. The main requirements are: occupy a small area, present low energy consumption, provide climate independent cooling and be low cost.

Traditional data centers use air cooling systems. These, due to their large size, low efficiency and high deployment cost, do not fulfill the Edge Computing needs of smaller centers. In addition, traditional data centers take advantage of geographical locations with cold climates to improve the efficiency of refrigeration.

Our research focus on the optimization of a cooling system for Edge data centers based on two-phase immersion cooling using a dielectric liquid. Below are its main advantages, which help meet the needs of this new computing paradigm:

  • The two-phase immersion cooling provides a much more efficient heat transfer. It allows to significantly reduce the energy consumption due to cooling. In scenarios with moderate density, it allows complete passive cooling.
  • It increases the computation density significantly, reducing the area of ‚Äč‚Äčthe infrastructure.
  • Immersion cooling is independent of weather and humidity.
  • Significantly reduces the average temperature of the equipment and its variations. Temperatures of about 60¬ļC are not exceeded, which improves the average time between failures.

Our work focuses on the following research lines:

  • Predictive modeling of immersion temperature for heterogeneous architectures.
  • Power modeling for heterogeneous architectures.
  • Predictive optimization of the immersion temperature for booting and steady state of the devices.
  • Energy optimization of Edge’s complete data center solution.

This research work is performed in collaboration with Technical University of Madrid (GreenLSI Group),  3M Spain, and Adam.