Sustainability-Aware Serverless Orchestration for Multi-Cloud Enterprise Systems: A Reference Architecture and Implementation Framework

Abstract

Serverless computing has become a default deployment model for event-driven, bursty workloads, prized for eliminating idle infrastructure. Yet the same properties that make Function-as-a-Service (FaaS) systems attractive also make them hard to reason about from a sustainability standpoint: short-lived invocations run on shared infrastructure with little visibility into their carbon cost, and existing platforms optimize almost exclusively for latency and price. This paper proposes a reference architecture for cost- and carbon-aware serverless orchestration in multi-cloud enterprise environments, introducing a scheduling and observability layer above native provider runtimes that weighs carbon intensity and cost alongside latency servicelevel objectives. We formalize the placement decision as an ordered set of hard constraints followed by a weighted cost/carbon objective, and validate the mechanism through a trace-informed simulation spanning nine provider/region combinations and roughly 100,000 synthetic invocations. The simulated fleet shows an estimated 35.9% carbon reduction and a 5.1% cost reduction for latency-tolerant workloads relative to a single-provider baseline, without measurably affecting latency-critical traffic. We relate the contribution to UN Sustainable Development Goals 9, 12, and 13, and close with a field-evaluation framework and a discussion of the trade-offs, including cold-start latency, vendor portability, and data gravity, that adopting organizations will need to weigh.