CASE STUDY
How AI helps Huawei build green data centres
Data centres are being overhauled , revamped and modernised to support Digital Transformation and the powerhungry cloud deployments they contain .
Artificial Intelligence ( AI ) is a key part of this overhaul , as energy conservation – to address both cost and environmental concerns – becomes a top consideration for data centres . AI solutions are being deployed in the push to get data centres to become more energy efficient as well as safer and more reliable .
The growth in data traffic volumes shows no sign of letting up , nor does the global demand for High Performance Computing ( HPC ). This comes with an increase in high-density servers , the use of graphics processing units and specialised AI processing chips , and all of these generate significantly more heat than traditional CPUs . It leads data centres then to require even more power to dissipate the heat they generate . It ’ s therefore no surprise that heat dissipation and energy management have become a vital topic for the data centre industry .
A data centre ’ s Power Usage Effectiveness , or PUE , is seen as a key indicator for evaluating its energy efficiency . The PUE value rises as the data centre becomes less efficient . Huawei ’ s cloud data centre in Ulanqab , for example , achieves an annual PUE as low as 1.15 , and is seen as a good example of how green data centres should be built .
Huawei has leveraged Machine Learning ( ML ) to develop its iCooling intelligent thermal management solution for data centre infrastructure and to address the energy consumption issue . The iCooling system incorporates deep learning to analyse historical data and identify key factors which affect energy consumption and create a PUE prediction . An optimisation algorithm then establishes the ideal parameters which are transmitted to various control systems .
At Huawei ’ s cloud data centre in Langfang , northern China , the deployment of iCooling has resulted in a PUE that is 8 % lower than it previously was , meaning a big saving in annual power costs . And at a China Mobile data centre in Ningxia , the introduction of iCooling technology has reduced the data centre ’ s total energy consumption by 3.2 %, saving more than 400,000 kWh of electricity a year . According to Huawei , as data centre loads increase and AI learning capability improves , 6 million kWh of electricity will be saved in the data centre every year , the equivalent to a reduction of about 3 million kg of carbon dioxide emissions .
While improved power management is a striking example of the value of
AI to the data centre industry , AI is also proving hugely valuable in the management of data centres in other ways . It ’ s used to improve the safety and reliability of data centres .
Prior to the introduction of AI , fault detection would need to be done on a case-by-case basis , isolating the faulty component . By collecting information from the power supply and distribution systems , AI can predict impending device and component failure , warn operations and maintenance personnel , and furnish additional information to assist decision-making .
Huawei ’ s iPower intelligent power supply and distribution technology , for example , can detect and isolate faults and then recover from them at subsecond speeds , improving the reliability of the data centre ’ s power infrastructure and reducing the risk of fire . iPower also accurately predicts battery life span and health , allowing preventative maintenance before any failure occurs .
AI can also be used in the data centre to reduce repetitive work and to monitor and manage fragmented resources which might otherwise easily be overlooked . Huawei ’ s iManager data centre infrastructure management system uses intelligent hardware and IoT sensors to manage power , cooling and space to optimise utilisation , with AI managing the allocation and operation of assets . ◊
OUR DIGITALISATION JOURNEY WITH HUAWEI HAS TRANSFORMED OUR OPERATIONS , BUILDING MORE RESILIENCE AND CONVENIENCE THROUGHOUT THE MUNICIPALITY .
www . intelligentdatacentres . com Issue 22
51