How digital twins can improve datacentre operations

The use of artificial intelligence (AI) to manage the energy consumption and carbon emissions of datacentres has been bandied about for years, but challenges related to data quality and availability have been hampering its full potential.

For one thing, data about datacentre operations tend to be limited to homogenous environments and do not provide enough coverage across different conditions. Also, there might not be enough data available to train machine learning models.

Even if the data issues are addressed, datacentre operators tend to be risk-averse out of fear that things could go wrong after implementing recommendations from AI engines that operate in a black box.

Red Dot Analytics (RDA), a deep tech company spun off from Singapore’s Nanyang Technological University, has been addressing those challenges with a digital twin platform that lets datacentre operators simulate actions aimed at improving operational and energy efficiency – and to understand the costs and risks of doing so. The data generated by the digital twin can also be used to train machine learning models.

But building a digital twin of a datacentre, including its physical assets and underlying processes and operations, is not an easy task, said RDA’s chief scientist and NTU professor Wen Yonggang.

“Different people have different interpretations of what a digital twin should be and that has been a major challenge when we work with industry partners, customers and stakeholders,” Wen told Computer Weekly, noting that most people tend to think of digital twins as just digital representations of physical datacentre infrastructure.

To Wen, that is just the first layer of RDA’s digital twin platform, which includes other capabilities such as overlaying operational data to perform statistical analysis and diagnostics, along with predictive and prescriptive capabilities that turn insights into recommended actions for datacentre operators.

RDA claimed that its “cognitive digital twin” applications can help datacentre operators reduce energy cost by up to 40% without having to change their hardware. In Singapore, where cooling systems account for the bulk of energy consumption in datacentres, the cost savings can be substantial.

Wen said RDA is also working with datacentre operators to help them better understand their carbon emissions and identify opportunities to reduce their carbon footprint, while others are using its digital twin platform in asset management to reduce the downtime of their datacentre assets.

“We’re building different use cases along the way when we work with customers, but the overall objective is to use digital twins as a decision support platform to turn the usual best practices into a more scientific way of conducting datacentre operations,” Wen said.

RDA currently counts major hyperscale cloud providers, co-location datacentre providers and enterprises in Asia-Pacific as clients, and expects more customers to come onboard in the next quarter. For each customer, RDA deploys a common backend engine but builds different front-end interfaces that integrate its platform with customer workflows.

Some datacentre operators, however, may not have sufficient data to ingest into a digital twin platform due to their ageing data collection systems and diverse data sources, Wen said.

“But the good thing about using digital twins is that we can start with a small amount of data to build the digital twin and gradually refine it such that at certain points, we can collect data from the digital twin to train the model,” he added.

On future enhancements, Wen said RDA is looking to strengthen the security architecture of its digital twin platform and make it easier for customers to build their own applications and services on top of the platform through low-code and no-code development.

“We can empower operators to do some things themselves, rather than rely on us or their service providers,” Wen said. “Additionally, we want to build more AI capabilities into our backend system and take advantage of other machine learning techniques like transfer learning to make our models more robust.”