The Internet of Things (IoT) requires applications to deal with a large amount of data - streamed, processed and stored from small devices to analytical systems. Cloud computing offers a hardware solution to this issue, providing on-demand resources to process IoT data. The newer programming paradigms simplify the use of those cloud resources. The Function-as-a-Service (FaaS) and the Serverless paradigm transform the conception of microservices applications to the definition and the composition of several callable functions. Although defined as distributed architectures - mostly publicly available solutions rely on either a gateway or an internal messaging middleware. These architectures create a \textit{single point of failure} in exchange for more straightforward service to service communication.

In this article, we present Dyninka, a framework to rapidly prototype FaaS-based distributed dataflow applications. Its programming model gathers the definition and the composition of services within a single file using the \textit{multitier programming} paradigm and compiles them into a multitude of services deployable on cloud computing infrastructure. Dyninka is built without a gateway or a messaging platform, and services communicate directly with each other or with the cloud abstracted infrastructure. As a result, we reduce the network and the computation overheads introduced by commercial FaaS frameworks such as OpenFaaS.

We validate Dyninka on a Fog computing scenario with limited resources and several load profiles. For all scenarios, Dyninka shows better stability, throughput and a reduced overhead compared to OpenFaaS.