When Do Microservices Save Energy? Evidence from Environmental Simulation Workflows

A comparative analysis of monolithic, polling-based, and event-driven orchestration for containerised environmental models

Joshua Rowley; Abdessalam Elhabbash· 2026· DOI 10.48550/arXiv.2608.18376

The core problem

Environmental simulation models are essential for scenario analysis, calibration, and decision-making. However, repeated execution of these models can incur significant energy costs. Microservices architecture offers modularity and scalability, but its low-carbon impact remains unclear because decomposition introduces overheads from orchestration, communication, persistence, and idle services. This paper evaluates four environmental models as containerised microservice workflows, comparing monolithic execution with polling-based and event-driven orchestration. The central research question is: under what conditions do microservices save energy in environmental simulation workflows?

The results show that microservices increase energy consumption for smaller or tightly coupled models, where coordination overhead dominates. For a larger workflow, event-driven orchestration reduces energy use despite longer runtime. Specifically, selective downstream re-execution achieves a 41% reduction during repeated parameter exploration. This indicates that the benefits of microservices are context-dependent: they are not universally energy-efficient, but can be advantageous for larger, loosely coupled workflows when event-driven orchestration and selective re-execution are employed.

Environmental simulation models are essential for scenario analysis, calibration, and decision-making. However, repeated execution of these models can incur significant energy costs. Microservices architecture offers modularity and scalability, but its low-carbon impact remains unclear because decomposition introduces overheads from orchestration, communication, persistence, and idle services. This paper evaluates four environmental models as containerised microservice workflows, comparing monolithic execution with polling-based and event-driven orchestration. The central research question is: under what conditions do microservices save energy in environmental simulation workflows?

The study evaluates four environmental models, each containerised as a microservice workflow. Three execution strategies are compared:

The findings highlight a trade-off between modularity and energy efficiency. For small or tightly coupled models, the overhead of orchestration, communication, and persistence outweighs any benefits, leading to higher energy consumption. For larger workflows, event-driven orchestration can reduce energy use by avoiding unnecessary computations, even if runtime increases. Selective downstream re-execution further enhances energy savings by only recomputing affected components during parameter exploration. This suggests that microservices should be adopted judiciously, with careful consideration of model size, coupling, and orchestration strategy. Future work could explore adaptive orchestration that dynamically switches between monolithic and microservice execution based on workload characteristics.

CS practitioners and researchers

Opening member content…