As a company specialising in the design, construction and operation of state-of-the-art data centers, Data4 places particular emphasis on energy efficiency. In a world of increasing demand for IT resources, data centers must continually innovate to meet these needs while reducing their environmental footprint. Artificial intelligence (AI) is emerging as a promising solution for optimising data center management, thus offering significant gains in many areas, including energy efficiency. Below we take stock of the current situation.
The areas of application of AI in data centers
AI is used in data centers to optimise various aspects of their operation. Firstly, energy management is a major challenge. AI can analyse data on power consumption and cooling systems to dynamically adjust the corresponding parameters as needed. It thus contributes to reducing energy consumption while maintaining optimal conditions for the proper functioning of equipment.
Predictive maintenance is another area where AI is bringing considerable benefits. By monitoring equipment in real time and analysing data from sensors, AI can detect early warning signs of potential failures. This allows for the optimal planning of maintenance operations by minimising the risk of breakdowns and the associated costs.
Good to know. As highlighted in the France Datacenter white paper, AI and predictions help define optimal operational conditions with regard to future needs and can thus assist the user in determining the best technical configuration (equipment to be commissioned, cascades, operation of recyclers, etc.) and settings (chilled water temperature, etc.), while respecting the security of the facility through redundancy. By extension, models can simulate the impact of performance actions such as setting changes and equipment replacements and upgrades. AI can be used as a lever for optimisation, reducing the environmental impacts of using certain machines or products in specific industries.
Using AI to improve energy efficiency in data centers
Artificial intelligence offers considerable potential for addressing the challenge of improving energy efficiency in data centers by leveraging complex data models and facilitating real-time analysis and decision-making.
Training AI in real-life conditions: a long but necessary process
Training artificial intelligence with real data is a crucial step to ensure the relevance and efficiency of the models developed. Real data allow AI to understand and learn the complex characteristics, patterns and relationships that exist in real-life environments. Based on these data, AI is able to make informed decisions and accurate predictions when faced with similar situations. However, training AI with real data can be a process that can take several months or even years.
There are several reasons for this length of time :
Data collection: to train an AI model, it is necessary to have a large and diverse data set that is representative of reality. Collecting these data can be time-consuming, especially if the events or phenomena studied are rare or seasonal. For example, in a data center, weather data in August or January have a direct impact on the building’s cooling capacity. Having varied data is therefore essential.
Data quality: real data are often noisy, incomplete or inaccurate. It is therefore necessary to clean, pre-process and supplement them before using them for AI training. This preliminary work can be long and tedious, but it is essential to obtain a high-quality model.
Model complexity: AI models, especially deep neural networks, are composed of millions of parameters that need to be adjusted to minimise the error between the model’s predictions and real data. Training such models requires significant computational resources and can be time-consuming, depending on the size and complexity of the model and the amount of data available.
Model fitting and validation: once the model has been trained, it has to be validated with new data to ensure its performance and ability to generalise. This step may require additional adjustments to the model and re-training, further extending the process.
The data models on which AI is based
To optimise energy efficiency in Data4’s data centers, AI relies on several data models, including :
- The customer’s energy requirements, which vary depending on what services are used and when.
- The environmental context, such as weather conditions and humidity, to adapt the cooling of equipment.
- Maintenance data, such as the history of interventions and the condition of equipment before and after maintenance, allowing AI to assess the impact of operations on energy performance.
- Data from the equipment itself, such as temperature, fan speed and airflow sensors, to adjust energy parameters in real time
The three main phases of AI deployment
There are three main phases to implementing AI to improve data center energy efficiency.
1 – The first phase consists of data collection, which must extend over a sufficiently long period (at least one year) to obtain a high-quality data set that is representative of reality. This step is crucial to ensure the relevance of the decisions made by AI.
2 – The second phase is the training of AI, which relies on machine learning techniques to learn from the data collected. AI is thus able to recognise trends, establish correlations between variables, and model the energy behaviour of the data center’s equipment.
3 – Lastly, the third phase involves leveraging AI to perform analyses and make decisions in real time. AI is then able to identify energy optimisation opportunities, detect anomalies and propose corrective actions to improve the data center’s energy performance.
As Olivier de Nomazy, Product & Innovation Manager at Data4, confirms, “by the end of 2023, we will have collected the essential data allowing us to run artificial intelligence models in 15 Data4 data centers in France and abroad. This work, which began one year ago, will enable us to further refine our clear positioning in favour of greater energy efficiency in order to combine cutting-edge technology with climate responsibility”.
Security & the role of AI in data center management
The increasing adoption of artificial intelligence in data centers is also raising important security and liability issues. Data center operators are facing the challenge of striking the right balance between the automation offered by AI and maintaining human control over critical decisions.
AI as a guidance aid
As security is a major issue for data centers, the question arises as to whether AI should be allowed to make autonomous decisions or whether its role should be limited to providing guidance. Data4’s position is clear: AI should not intervene directly on equipment without human supervision, in order to avoid risks related to deviant behaviour or errors of judgement. AI should therefore be seen as a valuable tool for guiding operators’ decisions, but under no circumstances should it replace them.
Artificial intelligence therefore offers considerable potential for improving the energy management of data centers. By leveraging various data models and being able to analyse and act in real time, AI helps optimise energy performance while reducing maintenance and operating costs. The prospects for the growth of artificial intelligence in the data center industry are promising, and Data4 will continue to explore and integrate this technology to deliver ever more efficient and sustainable solutions to its customers.
