Facilitating distributed data-flow programming with Eclipse Zenoh

Abstract

Data-flow programming is the computational model of choice for a large class of application domains, such as, real-time data processing, robotics platforms, and big-data analytics. Traditionally, dataflows are deployed and executed within well-defined system boundaries, such as robots, radars, or data-centers. These boundaries however are expected to blur with the advent of Edge Computing, which provides a multi-tier infrastructure spanning from the cloud to the things and enables for the distribution of applications across this continuum. In this paper we make a step towards the design of an Edge-native data-flow by mixing technologies coming from both worlds: ERDOS, a novel data-flow framework, and Eclipse Zenoh, a Named-Data-Networking built for the Edge Computing. More specifically, we (i) investigate how ERDOS can be expanded to cover Edge deployments by leveraging Zenoh, (ii) analyze the advantages provided by this integration, and (iii) evaluate the performance of a Zenoh-powered ERDOS. Our results show that ERDOS experiences a higher throughput and bounded latency when operating over Zenoh. Moreover, Zenoh enhances ERDOS with full location transparency, allowing developers and system designers to focus on the logic of their application as opposed to the topology deployment. Finally, our integration of Zenoh and ERDOS is available as open source at https://github.com/atolab/erdos-on-zenoh.