Carbon-Aware Workload Shaping for Event-Driven Microservices Across Multi-Region Clouds

Abstract

Cloud infrastructure that power event-driven microservices is increasingly deployed across geographic regions and uses electricity on a scale that is a significant source of greenhouse gas emissions in the world. This paper integrates the results of 20 peer-reviewed and industry sources on carbon-aware computing, sustainable data-center design, and event-driven microservice architecture to create a unified framework for shaping workloads in a carbon-aware fashion for multi-region clouds. The synthesis brings together recent work on carbon-aware scheduling and scaling systems with temporal workload-shifting and existing microservice communication patterns to shed light on how the event-driven pipelines can be reorganized in terms of time, space, and elasticity while maintaining the latency service-level agreements. The quantitative results presented are based on the reported carbon reduction from elastic carbon scaling up to fifty-one per cent, carbon reduction from latency-aware geographic provisioning up to seventy per cent and carbon savings from temporal shifting up to more than twenty per cent during weekends in several national grids. The paper also places these gains in perspective against the overall information and communication technology (ICT) emissions curve, which saw global data-center electricity consumption grow to around two hundred five terawatt-hours in 2018, or nearly one per cent of the world's total electricity consumption – despite a five-hundred-fifty-percent increase in computing workloads. A suggested reference architecture combines carbon-intensity forecasting, event-broker-level routing and elastic autoscaling to form a closed loop that works for Kafka-based and serverless event-driven applications.