In support of push-based streaming for the computing continuum

Real-time data architectures are core tools for implementing the edge-to-cloud computing continuum since streams are a natural abstraction for representing and predicting the needs of such applications. Over the past decade, Big Data architectures evolved into specialized layers for handling real-time storage and stream processing. Open-source streaming architectures efficiently decouple fast storage and processing engines by implementing stream reads through a pull-based interface exposed by storage. However, how much data the stream source operators have to pull from storage continuously and how often to issue pull-based requests are configurations left to the application and can result in increased system resources and overall reduced application performance. To tackle these issues, this paper proposes a unified streaming architecture that integrates co-located fast storage and streaming engines through push-based source integrations, making the data available for processing as soon as storage has them. We empirically evaluate pull-based versus push-based design alternatives of the streaming source reader and discuss the advantages of both approaches.