Environmental, Social, and Governance Co-innovation

Deep Cooling Solution in Data Centers

The speedy growth of data center energy usage is leading the IT industry to look for new sources of energy efficiency. That is especially as energy efficiency gains are not as readily available as they were in the past. In the first half of 2020, the power consumption of data centers in the United States alone reached 340 Mega Watts. And the Information, Communication, and Technology Industry accounts for roughly 7% of global annual energy usage.

The world’s sustainable economy calls for data centers to reduce emissions while boosting their capacities. This requires the wisdom of operators to address a series of questions, such as how to balance these two seemingly contradictory demands for data centers and how to solve the problems faced by the data centers through edge computing, machine learning, and operation management technology of facility/IT converged architecture.

VMware has a long-term commitment to sustainability. VMware announced the goal of carbon neutrality as part of the 2030 Agenda to achieve net zero carbon emissions for VMware’s operations and supply chain, which outlines its commitment to achieving 30 goals by 2030 to create a more fair, sustainable, and resilient world. A vital component of the 2030 Agenda is “Intrinsic Sustainability” innovation, which focuses on supporting customers’ carbon-cutting initiatives through VMware’s virtualization, cloud solutions, and partners’ efforts.

To address these urgent needs, VMware has once again created a software solution to a hardware problem. Deep Cooling uses AI and big data to accurately predict and match the optimal parameter combination of cooling in the data center by modeling with machine learning. Using operation metrics of critical equipment in the cooling chain, monitoring data of the dynamic environment, and energy usage, Deep Cooling reduces energy and costs by dynamically adjusting parameters in each link of the cooling system.

To address these urgent needs, VMware has once again created a software solution to a hardware problem.

Deep Cooling models various physical systems such as power load, heating, and cooling of the whole data center by using historical data and then optimizes the control of the cooling system based on the prediction results of loads and upgrades the traditional feedback-based control of the cooling system to model-based feed-forward control. Energy efficiency optimization of Deep Cooling is a dynamic cycle process.

In China, VMware and Intel are converging ecosystems to promote innovation in edge computing, artificial intelligence, and carbon-neutral computing. For example, Intel, Quarkdata, and VMware are cooperating to develop a joint solution to accelerate the adoption of sustainable carbon neutrality computing technology in data centers worldwide. Deep Cooling, sustainable carbon-neutral computing technology, helps companies optimize the energy consumption performance of the entire cooling system in the data center through continuous edge intelligence intervention and real-time control. Under the joint promotion of the three parties, this solution has been implemented in several sizeable domestic data centers, helping customers to significantly improve power usage effectiveness (PUE) and reduce carbon emissions.

The main steps of each cycle are:

Data collection

Data is collected from the dynamic environment, machine room air conditioners and physical space, manual data, etc. Using intelligent sensors and a gateway device, preliminary to summarize the data and generate intermediate data such as machine room air conditioner redundancy and machine room heat maps.

After connecting with the data center infrastructure management (DCIM) and the cooling system, Deep Cooling collects data on temperature, humidity, energy consumption, air conditioner load, and associated water lines of each machine room. Access to this data provides a robust set of data display and visualization systems. The result is a clear picture of energy usage that intuitively and graphically shows temperature differences and heat maps of the cooling and heating equipment in machine rooms. In addition, the solution visually shows the thermal distribution and energy consumption of the cabinet and air duct, through which users can rapidly discover and deal with abnormal devices and settings.

The system also has abundant alarm and policy auditing capabilities to better support the operation and maintenance of the data center. Generally, the system can analyze the heat load inside the machine room, analyze and visualize the cooling output, and visualize every detail of equipment and sensors:

Heat map example in Aria Operations of cooling system in a data center
Temperature trend examples in Aria Operations from sensors in a data center

The system also has abundant alarm and policy auditing capabilities to better support the operation and maintenance of the data center.

Analysis and modeling

Build various heat balance models for components of the cooling system, such as terminal air conditioner, water-chilling unit, water line, etc., and establish the relationship among heat load, cooling system, and temperature.

Based on the data collected by the overall system, the cooling system in the data center is modeled in two steps. Step one is to model the use-end of the actual cooling capacity, that is, server rooms. Step two is to optimize the control of the computer room air conditioning (CRAC) fans on the end, based on the models established in step one.

Steps to simulate physical model for server rooms and optimize control of CRAC fans

The second step is to model the data center load volume to better control the centralized cooling source:

Steps to model the overall workloads and optimize control of the central cooling source

By modeling the overall physical system of the data center, the system realizes a feed-forward optimization control based on modeling to improve the traditional feedback-based controls for cooling systems. This result is better energy consumption performance and enhanced temperature control.

Optimization control

Solve the parameters, as commanded by the program, of each cooling system at each moment by the trained machine learning model, and then accurately match the system’s output according to the load, to reduce its energy consumption.

The platform digitizes the on-site equipment to build a digital twin and optimizes the control of the equipment in the server room for the whole life cycle.

The main types of equipment controlled by the system include:

  • Terminal precise air conditioner
  • Water-chilling unit
  • Valves and pumps of water line
  • Fans of cooling devices and related equipment

Generally, the system automatically controls the parameters of related equipment to dynamically adapt to the data center’s load and provide more uniform and stable temperature control and better power consumption performance. At the same time, the system also designs a set of tools and processes to facilitate the managers of the data center to intervene in the implementation of Deep Cooling at any time and take over the control of the cooling system in case of emergency. The open-loop control of data center managers is generally introduced through the following rule-based configuration:

Rule-based configurations to control cooling system in data centers

The system achieves the optimal control of the overall cooling system of the data center through a distributed computing system based on edge inference. Each server room has an independent control node to execute the model inference at the edge, calculate optimization control, and control all the air conditioners. Similarly, in the cooling source, there is also a matching edge control node to execute relevant optimization control.

The system saves 18% – 25% of electricity for cooling, which leads to a PUE (power usage effectiveness) improvement of approximately 0.1-0.15. In addition, the system requires minimal changes to data center hardware and does not negatively impact normal production during the implementation process. Therefore, the return on investment (ROI) of Deep Cooling is more than double that of the hardware-based energy consumption modifications.

Project Flowgate, an open-source project initiated and maintained by VMware, empowers enterprises to integrate facilities with data from IT systems to form a unified data platform and create a holistic approach to assist the cloud and data center operation teams in decision-making. By unifying these two different data sets into one layout, together with advanced technologies such as augmented reality, patrol robots, and deep learning, data center administrators can optimize operations better and make wiser choices for application placement.

Overall visual architecture of converged operation management in data centers

The Deep Cooling solution, data visualization, and dashboard components were originally announced together at VMware Explore US 2022 and are currently available via the Aria (vRealize) Operations management pack, found in the VMware Cloud Marketplace.

We are working closely with Intel and Quarkdata to continuously extend the deployment of the Deep Cooling solution. And while the partnership pipeline in China is strong, we are actively looking to expand partnerships to other regions globally, especially in Europe. If you are interested in learning more about Deep Cooling, feel free to reach out to us.


Leave a Reply

Your email address will not be published. Required fields are marked *