Carbon Intelligent Computing for Hyper-scale Data-centres: Matching Computing and Networking with Global Economy Needs

点击翻译按钮获取中文摘要

Information and Communication Technology (ICT) has significantly changed the global business landscape in the last couple of years and has made it virtually accessible anywhere and anytime. Much of this credit can be attributed to the inception of the Cloud Computing platform that provides virtualised, distributed, and powerful computing infrastructure to cater to the ever-growing economy's needs. The pillars of this powerful platform rest on gigantic industrial facilities called HyperScale Data-centres (DCs), which are characteristic of their colossal setup space, seamless computing, and more than 100 MWs of electricity requirements. However, these DCs consume almost 200 TWh of energy every year to support different business needs, which leads to 0.3% carbon emissions worldwide. Various reports indicate that this trend will sharply rise soon. For instance, by the time a child born today reaches her teens, the energy consumption of these cloud DCs will escalate to almost 20% globally. Furthermore, the growing dependence of the ICT industry on electricity has also changed the way the energy is being produced and used worldwide. During the last two decades, many tech giants have made serious efforts for reduced energy consumption and have adopted clean energy initiatives. Google, one of the biggest Internet and Cloud giants that is carbon neutral since 2007, has introduced one of the latest trends in this direction called "carbon intelligent computing". It supports the notion to shift the computational and network-intensive tasks of the cloud DCs to the periods of high green energy availability in contrast to storing the off-peak power and using it during the peak hours. However, this iconic shift towards 24/7 utilisation of carbon-free energy raises an important research question, i.e., how to match the variable renewable energy availability with the variable digital needs for data processing, storage, and transmission? Another important research challenge is how to shift the DC load in both time and space to meet higher sustainability goals without jeopardising DC's performance. Therefore, in my discovery research program, I will attempt to analyse, elaborate, and address the critical challenges related to the adoption of carbon intelligent computing across the globally distributed HyperScale DCs, and designing optimal scheduling and distribution algorithms for the incoming workloads and traffic across these HyperScale DCs. Thus, the overall idea is to match the computing and networking needs within the HyperScale DCs to the global market in a sustainable and energy-efficient manner. The research program's overall benefits are to further the academic knowledge and industrial advances for efficient Carbon Intelligent Computing since it involves the training of many graduate students. Moreover, the research results will be disseminated using open access preprints and the transfer of intellectual property to Canadian Industries.