EcoRun: Energy-aware container scheduling and resource allocation in serverless edge computing
Year: 2026
Type: article
Abstract: The sustainability of computational systems has rendered energy efficiency a critical consideration in modern computing infrastructures. Cloud and edge computing are characterized by substantial and rapidly increasing energy footprints. Serverless computing offers a paradigm that abstracts infrastructure management by offloading resource provisioning to cloud providers, thereby streamlining software development, deployment, and operational governance. Despite these advantages, existing serverless platforms predominantly employ static container caching policies, which retain function instances for a fixed duration following invocation to minimize cold-start latency for subsequent requests. However, such static policies are inherently limited in their capacity to support energy-aware system operation, thereby constraining the realization of sustainable serverless deployments. To address this limitation, we propose an adaptive, dynamic algorithm for request dispatching and container caching, specifically designed to optimize energy costs in distributed, heterogeneous edge environments. Our proposed system, EcoRun, is a serverless scheduler that integrates container dispatching and caching mechanisms within extant serverless platforms. By jointly optimizing request scheduling and container lifecycle management, EcoRun reduces overall energy costs. This joint optimization problem is computationally challenging, being NP-hard, and is further complicated by the absence of deterministic future information, such as request arrival patterns. To mitigate these challenges, we first formulate an energy cost minimization problem that includes container running cost, container switching cost and traffic routing cost. Then, we develop an online dispatching algorithm with provable performance guarantees, complemented by a container caching strategy that dynamically manages instance lifetimes. We present the architectural design of EcoRun and evaluate its performance through both simulations and a prototype implementation over the Knative platform. Experimental results on a real-world testbed demonstrate that EcoRun achieves a 24.4% reduction in request latency and a 15.3% decrease in energy cost relative to baseline approaches, underscoring its efficacy in enhancing both performance and sustainability in serverless edge computing systems.
Source: Computer Communications
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,620
OpenAlex ID: W7204630940