Data Centre Optimisation: Strategies for Enhancing Performance and Efficiency

Demand for data centre services is exploding with AI and digitalisation, however facing more resource constraints than ever, optimisation in data centres has never been so important. This article explores how and why data centres optimise their build and operations.

What is data centre optimisation?

Data centre optimisation refers to the improvement of efficiency, resiliency, and effectiveness of data centres through enhancing its physical infrastructure, cooling, power systems, and software platforms. While the reliance of businesses on data centres is long-standing (for base services such as cloud computing), the recent explosion in AI usage and the drastically accelerating pace of digital transformation is driving an unprecedented demand for their services. However, this surge comes at a time when data centres are increasingly facing significant resource constraints – across both power provision, physical space, and skilled talent. As a result, the optimisation of existing facilities has become more important than ever to keep infrastructural pace with these transformative technologies and the rapid growth of the digital economy.

In 2022, data centres consumed approximately 1-2% of global electricity production, a figure expected to drastically rise as the world’s digital economy expands. For example, hyperscale data centres (which power services such as Google, Microsoft and Amazon) are projected to reach over 1,000 facilities globally by 2025. With this, comes exceptionally high electricity consumption – the cost of powering and cooling such data centres represents a significant portion of data centre operator’s operational expenditure, making energy efficiency a central target within optimisation efforts.

Within data centres, short design cycles have often led to missed opportunities around improving efficiency. Many data centre designs still rely on legacy strategies, despite newer, more efficient alternatives. The average data centre PUE (power usage effectiveness) – a measure of data centre energy efficiency – has remained around 1.58 globally, indicating significant room for improvement. By comparison, the most advanced data centres are able to achieve a PUE closer to 1.1, showing the benefit of targeted optimisations and optimised design strategies.

The benefits of data centre optimisation

Cost savings

One of the most immediate and compelling benefits associated with optimisation is the opportunity for tangible cost savings. Efficient use of resources within the data centre environment leads to a reduction in hardware requirements, energy consumption, and cooling costs. Additionally, optimised systems can reduce downtime, avoiding associated revenue losses and unplanned maintenance costs. All of this targets the core operational expenditures of data centre operators, enabling enhanced profitability.

Increased resource utilisation

Optimisation enables better resource utilisation, with workloads being consolidated onto fewer servers, maximising the performance of existing hardware. This in turn supports the growth of workloads such as general-purpose computing, as well as emerging technologies such as artificial intelligence and machine learning. With the global high-performance computing market projected to grow from $39.1 billion in 2023 to $55.8 billion by 2028, demand on data centre resources will be significantly increased – making optimised resource utilisation essential.

Prolonged equipment lifespan

Through the adoption of optimisation techniques such as advanced cooling systems and workload distribution, data centres are able to extend to lifespan of equipment. Improved temperature management reduces thermal stress on hardware, while more balanced workloads (preventing servers from operating at maximum capacity for prolonged periods) reduce wear and tear leading to fewer breakdowns and a longer operational lifespan. Thus, IT hardware replacement costs, which can amount to $1 million or more per year for larger facilities, can be reduced significantly.

Environmental sustainability

Optimised data centres consume significantly less energy. With data centres globally indirectly contributing around 1% of total GHG emissions (according to the International Energy Agency) through their respective electricity requirements , targeting internal efficiencies inevitably holds substantial environmental benefits. Less energy usage results in fewer carbon emissions, helping companies meet their sustainability goals and adhere to environmental regulations.

Best practices for data centre optimisation

AI and machine learning

Recently, many of the most promising new developments in data centre optimisation are related to using artificial intelligence and machine learning to accelerate efficiency gains.

• Deploy AI-enabled predictive analytics: Predictive analytics powered by AI/ML can predictively forecast power needs and optimise energy consumption dynamically. Using this, data centres can automatically adjust power consumption, shitting workloads to off-peak hours or reducing operations during low-demand period.
• Use AI to optimise cooling systems: In forecasting power needs, and using real-time data from temperature sensors, AI-driven systems can also implement targeted plans for effective use of data centre cooling systems. Machine learning algorithms can be deployed to optimise airflow, enabling further efficiency gains. In deploying AI, Google has been able to achieve a 40% reduction in its cooling energy usage across its data centres, while Microsoft has been able to achieve a 30% reduction.
• AI-monitored energy usage efficiency measures: AI can continuously monitor various systems in real-time, providing valuable insights across energy consumption patterns and inefficiencies. For example, AWS has used Ai to actively optimise server load distribution to reduce energy waste.

Airflow management

Efficient airflow management is a core component of an optimised data centre and can cut cooling costs by as much as 40%.

• Optimise hot/cold aisle configuration: Using a data centre layout which incorporates both ‘hot’ and ‘cold’ aisles acts to hot and cold air in the data centre. Racks are aligned to ensure that the intake sides face ‘cold’ aisles while the exhaust sides face ‘hot’ aisles. This prevents hot air from recirculating ensuring the data centre environment remains cooler.
• Use blanking panels: Installing a blanking panel – a typically plastic or metal piece used to cover empty spaces – in empty rack spaces helps to prevent air mixing and ensures that cold air reaches equipment intakes, improving cooling efficiency.
• Seal gaps: Using plastic curtains, underfloor seals, and grommets to seal any gaps acts to prevent the mixing of hot and cold air, improving cooling system performance.

Cooling system optimisation

Optimising cooling systems is another way to ensure maximum data centre efficiency.

• Improving cooling efficiency: Deploying variable-speed fans and liquid cooling systems which adapt to real-time server loads can be highly effective. According to ASHRAE (American Society of Heating, Refrigerating, and Air-Conditioning Engineers), allowing server inlet temperatures to rise up to 27°C can reduce cooling costs without impacting server performance.
• Leverage free cooling: Free cooling systems that use air or water from outside the facility to cool the data centre can reduce reliance on compressor-based cooling, significantly reducing energy costs. This methodology can lower cooling energy consumption by up to 70%, depending on the climate and setup.
• Heat reuse systems: Instead of releasing waste heat, data centres are able to use it to provide heating for nearby buildings – thus transforming a waste product into a useful product. For example, in Denmark, a Microsoft data centre is set to provide surplus heating to warm upwards of 6,000 local households. While heat reuse significantly improves environmental outcomes, this is not accounted for in the traditional efficiency measurement of PUE. Consequently, many data centre operators find themselves lacking the proper incentives to invest in heat reuse iniatives, given efficiency is commonly judged purely in terms of PUE. This misalignment of incentives in a critical issue which regulators and the industry as whole should take steps to address.

Effective power management

Adopting the most efficient power management systems is central to data centre optimisation

• Use high-efficiency power supplies: Modern power supplies can have efficiencies of up to 92%, while older models tend to be approximately 60-70% efficient. Implementing modernised power supplies can thus enable tangible efficiency gains.
• Introduce high-efficiency IT equipment: Using servers, storage devices, and network hardware that are ‘Energy Star’ rated or specifically designed for energy efficiency can result in significant savings. Energy-efficiency servers can use up to 30% less power than standard equipment.
• Adopt dynamic power management: Advanced power management features, such as placing servers in low-power states when underutilised can reduce power consumption by as much as 30%. Additionally, introducing idle-state management in networking devices within data centres can also enable energy savings.
• Use LED lighting with motion sensors: Replacing traditional lighting with LEDs and installing motion sensors can significantly reduce energy used for lighting – by up to 80%. As lighting typically accounts for approximately 2% of data centre energy, this reduction can further boost overall energy savings.

Virtualisation and consolidation

Introducing virtualisation and consolidating data loads can result in substantial efficiency gains.

• Increase server utilisation: Virtualising workloads enabled data centres to consolidate multiple tasks onto fewer physical servers. This reduces the total number of servers, and by extension, energy use. Virtualised servers often operate at 60-70% CPU utilisation, compared to only 10-30% for non-virtualised ones.
• Improve resource allocation: Software-defined data centres (SDDC) allow for the flexible allocation of resources like storage, computing, and networking, according to intelligent assessment of data centre needs. This leads to better overall resource efficiency. This Is reliant on effective monitoring systems which can continuously measure power consumption, cooling efficiency, and hardware performance.

Data centre optimisation framework: A step-by-step guide

Step 1: Implement monitoring systems

To begin the optimisation process, it is key to have oversight over the core metrics relevant in assessing your data centre’s current efficiency. To do this, one must set up comprehensive monitoring systems to track key data centre metrics, such as energy consumption, cooling efficiency, and resource utilisation. This accurate, real-time data is essential for identifying opportunities for improvement.

Step 2: Benchmark current performance

The data collected by monitoring systems can then be used to establish performance benchmarks. This includes setting baseline metrics for power efficiency, system utilisation, and operational costs to help identify areas where optimisation is most crucial.

Step 3: Identify areas for improvement

Analyse benchmark data to identify the most prominent inefficiencies in power, cooling and resource usage. Comparison against industry-leading benchmarks for these metrics can be a useful device in understanding relative performance and setting targets accordingly. It is important to look for areas where systems are underperforming or consuming more resources than necessary, and prioritise them for optimisation.

Step 4: Develop and implement optimisation strategies

Create a tailored plan to ringfence and address the identified inefficiencies. This could involve reconfiguring existing systems, redistributing workloads, or adopting new technologies. Changes which are implemented should be aimed to improve overall efficiency, resilience, or sustainability efforts.

Step 5: Review and refine continuously

Data centre optimisation is a continuous process. After planned optimisation strategies are implemented, it is necessary to continuously monitor performance. Data should regularly be reviewed in order to inform and adjust strategies which ensure ongoing efficiency and adaptions to any changes in demand or infrastructure needs.

Download this article as a PDF