Provuse: Platform-Side Function Fusion for Performance and Efficiency in FaaS Environments

A research team from TU Berlin and collaborators has published Provuse, a novel platform-side optimization technique for Function-as-a-Service (FaaS) environments that addresses a fundamental inefficiency in serverless computing. When complex applications comprise multiple functions that call each other, traditional FaaS platforms suffer from increased latency and resource overheads due to redundant function instances and 'double billing'—where both the calling and called functions incur execution costs. Provuse solves this by automatically performing function fusion at runtime, transparently consolidating execution without requiring developers to change their code or deployment patterns.

The system targets provider-managed FaaS platforms that control function entry points and deployment artifacts, enabling this optimization entirely at the infrastructure level. The researchers implemented Provuse using both the tinyFaaS platform and Kubernetes, demonstrating compatibility with mainstream container orchestration frameworks. Their evaluation shows consistent improvements across workloads, achieving an average 26.33% reduction in end-to-end latency and a 53.57% reduction in mean RAM usage. These results indicate that automatic function fusion represents an effective strategy for improving both performance and cost-efficiency in composed FaaS applications.

By moving optimization responsibility from developers to the platform, Provuse reduces the complexity barrier for building efficient serverless applications. The approach highlights the significant potential of transparent infrastructure-level optimizations in serverless systems, where providers can implement intelligent resource management without breaking the serverless abstraction. This research, accepted for publication at the SESAME '26 workshop, points toward a future where FaaS platforms automatically optimize application performance behind the scenes, making serverless architectures more practical for latency-sensitive and resource-intensive workloads.