Using AI to Reduce Energy Consumption, Cost and Carbon Emissions in Data Centers
In this ATC Insight
Cameron Conn, AI Technical Manager of QiO Technologies, was a co-contributor to the paper.
As the pace of technology accelerates, energy consumption and its associated costs impose growing concerns. While companies wrestle with net zero plans, ensuring sustainable solutions that do not reduce performance or incur higher costs is crucial. AI and machine learning are transforming everything, and World Wide Technology (WWT) expects that they will transform the way in which data centers are operated.
WWT has identified a new and AI-driven energy-saving product, Foresight Optima DC+™ which claims that it can reduce data center energy consumption without compromising performance.
This paper outlines the testing undertaken by WWT in the Advanced Technology Center (ATC) during Q1 and Q2 2023, covering the background, approach, and methodology used, as well as the test results of the Foresight Optima DC+™ product. We conclude with the potential contribution of Foresight Optima DC+™ to a more sustainable path for technology growth.
Background – Data centers are indispensable in technological advancement
Growing demand for computing, networking and storage to deliver public and private cloud services such as artificial intelligence, Software as a Service (SaaS) and Application Service Providers (ASP) has resulted in a significant increase in the use of data centers for efficient execution of such intensive workloads. Data centers consume a large quantity of energy. By some estimates, they contribute up to 2 percent of global carbon emissions. Lighting, utilities, cooling, heating/ventilation, and IT equipment require abundant financial and natural resources such as water and energy. Increasing pressures surrounding climate change, resource depletion and financial strains call for sustainable solutions to optimize energy efficiency and reduce emissions.
The amount of energy consumed by a data center depends on each server's load. Data centers consume energy to run, maintain and cool IT servers; Data center cooling can consume up to 40 percent of total power.
This paper focuses on testing the effectiveness of an AI software solution to optimize power consumption while servers are running at a predetermined load.
WWT technical engineers are available to assist with any inquiries relating to liquid cooling technologies.
Leveraging the ESG-dedicated Advanced Technology Center Lab
Testing and analysis of the solution took place in WWT's ATC, which has a lab dedicated to sustainability testing and research. The ATC is a platform of physical and virtual labs created to foster a collaborative innovation ecosystem to design, build, educate and deploy innovative technology products and integrated architectural solutions.
The Solution - Foresight Optima DC+™
QiO Technologies is a leading Industrial IoT AI-based software products company that targets energy-intensive industrial companies, data centers and telecom providers. In partnership with Intel, QiO Technologies has created Foresight Optima DC+™ to address the energy efficiency and Scope 3 emissions within the IT equipment of data centers. The Foresight Optima DC+™product optimizes data center performance at the device level – maximizing production and quality while concurrently minimizing energy consumption, operating costs and CO2 emissions.
Figure 1.1 shows a schematic overview of Foresight Optima DC+™
Foresight Optima DC+™ with its inbuilt AI models is trained to monitor and learn server power usage and CPU utilization patterns intelligently. Consequently, Foresight Optima DC+™ proposes (open loop) or automatically executes (closed loop) actions to adjust power consumption to meet the needs of the server workloads more efficiently, without compromising operations or quality of service.
"As server usage has historically been managed conservatively to guarantee uptime and Service Level Agreements (SLAs), sleep states have not been utilized effectively Exploiting this fact with a data-driven optimization approach allows for significant energy consumption savings to be achieved without impacting QoS."
Gary Chandler, CTO (Chief Technology Officer) of QiO Technologies
Methodology
To ensure that the efficacy of the Foresight Optima DC+™ could be seen across servers with varying utilization rates, the team installed the product on four WWT servers – 2x R650 and 2x R750 – each with a different pre-defined agreed load.
Each server model was tested with one flat and one varying load. This was done to prove that the product would have a positive effect in reducing power consumption across varying loads and load types.
Power draw data was rolled up every hour during the period of each experiment, where each row in the dataset was an average of the power draw for the hour. The QiO team directly read results from the motherboard, while the WWT team used the Power Distribution Unit (PDU) to validate the findings independently.
Steps:
- Configure hardware for the testing
- Determine pre-agreed loads for each server to run tests on
- Data collection with Foresight Optima DC+™ inactive - 3rd April 9AM UK to 11th April 9AM UK
- Data collection with Foresight Optima DC+™ active – 11th April 9AM UK to 19th April 9AM UK
Promising results highlighting proof of value in saving energy use, costs and emissions
This experiment showed a general reduction in energy consumption across the four servers. Servers with flat loads (650A, 750A) saw an average power draw reduction of 19-23 percent, and servers with varying loads (650B, 750B) saw a higher average power draw reduction of 27-29 percent. The full results are as follows:
Server | Agreed Load | Avg. power draw Foresight Optima DC+™ off (W) | Avg. power draw Foresight Optima DC+™ on (W) | % Reduction | Remarks |
021474 - R650B | 20% Flat | 312 | 240 | 23 |
|
021262 - R750B | 80% Flat | 480 | 390 | 19 |
|
021473 - R650A | 20%/40%/60% changeable load | 408 | 390 | 29 |
|
021261 - R750A | 50%/70%/90% changeable load (rotated every hour) | 480 | 350 | 27 |
|
As demonstrated by the charts above in Figure 1.2, starting 11th April, when QiO's Foresight Optima DC+™ was activated, there was an immediate drop in power draw across all four servers, regardless of agreed loads or whether they were varying or flat.
Independent Validation by WWT
To thoroughly vet the data, WWT ATC engineers also independently validated the impact of the product by collecting PDU data during the experiment.
Results captured in Figures 1.3 and 1.4 closely replicate the power draw drops shown in Figure 1.2; the power measurements taken at the PDU correlate strongly with the data collected from the motherboard, indicating that all four servers are running with considerably less power after implementing the product.
"Although tested using synthetic loads, these results are very positive with potential upsides to energy reductions noted."
Jeff Gargac, WWT Technical Solutions Architect
Additionally, the WWT team gathered other secondary data and observed changes before and after Foresight Optima DC+™ was activated. For example, the temperature was collected as a key variable, and there was an average drop in the inlet temperature across all four servers when the Foresight Optima DC+™ product was active.
Before the product was activated, the inlet and exhaust temperatures across all four servers averaged 19.2 and 34.3 degrees Celsius, respectively. Starting 11th April – the average temperatures dropped to 18.9 degrees Celsius (inlet) and 31.1 degrees Celsius (exhaust), a 1.6 percent and 9.2 percent reduction.
The fall in temperature indicates another area where energy can be saved. WWT's prior research into using cooling liquid as a more sustainable and energy-efficient method of cooling servers investigated fuel and cost-saving benefits from better cooling technologies. Future research can further analyze the cascading effect of power draw reductions, reducing cooling requirements and consequently the energy required to drive cooling systems.
A future for more sustainable data centers
"Having tested this solution in WWT's Advanced Technology Center (ATC) Sustainability lab, where we are able to independently and objectively test hardware and software permutations, in our technical opinion, this solution merits serious consideration in the march towards making data centers more sustainable."
Chris Braun, WWT Technical Solutions Architect
Results from this Proof of Value at the ATC Sustainability Suite demonstrate that the Foresight Optima DC+™ product can reduce energy consumption in servers with both fixed and dynamic loads. Furthermore, prior research has revealed that the Foresight Optima DC+™ product also saves energy when the servers are idle, with Intel servers potentially seeing a 52 percent decline in energy usage while not being used.
WWT will continue to invest in and research ways to reduce energy consumption across data centers, helping our customers bring down energy draw, energy costs, and related carbon emissions and helping these customers accelerate their environmental and sustainability goals.