Serverless Abstractions for Short-Running, Lightweight Streams

A team of researchers including Natalie Carl and Niklas Kowallik has published a paper proposing 'stream functions,' a novel extension of the Function-as-a-Service (FaaS) model designed to address inefficiencies in processing short-running, lightweight, and unpredictable data streams. The work, accepted for publication at the SESAME '26 workshop, identifies a gap where traditional serverless computing and stream processing both make assumptions that render them inefficient for stateful processing of brief, intermittent data flows. The core innovation treats the short stream itself as the fundamental unit of execution, state, and scaling, rather than individual events or long-lived processes.

Technically, stream functions process data via an iterator-based interface, allowing developers to write seamless logic across events within a stream while maintaining the elasticity and 'scale-to-zero' benefits of serverless platforms. The evaluation is particularly striking: in a video-processing use case, the model demonstrated a ~99% reduction in processing overhead compared to a mature stream processing engine. By providing performance comparable to standard serverless functions but with native stream semantics, this research offers a potentially transformative abstraction for a significant class of real-time, event-driven workloads in IoT, edge computing, and multimedia processing that are currently poorly served.