Carbon-Aware Dynamic Data Routing in Cloud Pipelines

Abstract

As energy consumption in data centers and carbon emissions become more and more critical environmental issues for the information technology industry, energy consumption is accelerating. In 2020, global data centres consumed about 200 TWh of electricity, representing a small fraction of global electricity demand of one per cent. While absolute energy consumption has been relatively constant due to improvements in efficiency, the carbon intensity of energy consumption has emerged as the most important consideration as the percentage of renewables in the electricity supplied to the grid fluctuates significantly from one geographic area to another and one time period to the next. In this paper, the authors will present the main research works prior to 2020 on carbon-aware dynamic data routing in cloud pipelines: computational workloads are dynamically moved in real-time to the data centre with the lowest instantaneous carbon intensity. The synthesis considers the basics of energy consumption models, geographical load balancing algorithms, virtual machine consolidation heuristics, renewable energy integration frameworks and multi-cloud carbon management strategies. Literature provides quantitative evidence showing that carbon-aware routing can cut carbon emissions by 30-45% compared to latency-first routing, with only a small latency increase of 3 to 8 milliseconds. A maximum of 69 percent energy savings is possible using combined VM dynamic consolidation and carbon-aware placement. It also reviews incentive structures for colocation facilities and self-adaptive scheduling algorithms, discusses the key architectural aspects, algorithmic bottlenecks, and future challenges in the field.