Fogify: A Fog Computing Emulation Framework

Topology Editing

A typical workflow starts with the user editing the docker-compose file of an IoT application, and extend it to encapsulate Fogify’s model. The Fogify model is composed of: (i) Fog Templates, allowing the description of Services, Nodes and Networks; and (ii) the Fog Topology, enabling users to specify Blueprints. A Blueprint represents an emulated device and is a combination of a Node, Service, Networks, replicas and a label. Services are inherited from docker-compose while the x-fogify section provides all Fogify primitives. Thus, users still develop their application using familiar docker constructs with the added functionality of Fogify not affecting portability. This means that a Fogify enhanced description will run in any docker runtime environment without any alterations, however users will lose the functionality offered by Fogify.

Deployment

When the description is ready, the user deploys the application using the FogifySDK through a Jupyter notebook, with the description received by the Fogify Controller. If no error is detected by the Controller, it spawns the emulated devices and creates the overlay mesh networks between them, instantiates the services, and broadcasts (any) network constraints to Fogify Agents. Specifically, the Controller translates the model specification to underlying orchestration primitives and deploys them via the Cluster Orchestrator, ensuring the instantiation of the containerized services on the emulated environment. Located on every cluster node, Fogify Agents expose an API to accept requests from the Controller, apply network QoS primitives, and monitor the emulated devices.

Testing

On a running emulated deployment, Fogify enables developers to apply Actions and “what-if” Scenarios (sequences of timestamped actions) on their IoT services, such as ad-hoc faults and topology changes. Actions and Scenarios are written by following the Fogify Runtime Evaluation Model. When an action or a scenario is submitted, the Fogify Controller coordinates its execution with the Cluster Orchestrator and the respective Fogify Agents. Furthermore, Fogify captures performance and app-level metrics via the Fogify Agent monitoring module. All metrics are stored at the Agent’s local storage and can be retrieved from the FogifySDK.

Analysis

To create an end-to-end interactive analytic tool for emulated deployments, we exploit the FogifySDK capabilities in Jupyter Notebook stack. Specifically, we pre-installed the FogifySDK library on Jupyter thus the user can (un-) deploy a Fog Topology, apply ad-hoc changes and scenarios, and, especially, retrieve runtime performance metrics. For the latter, FogifySDK stores metrics to an in-memory data structure, namely panda’s dataframe, providing exploratory analysis methods that produce plots and summary statistics. Except of out-of-the-box plots, provided by Pandas, we extended FogifySDK with tailored functions that provide a set of plots illustrating and explaining the effects of actions and scenarios in application performance. With the wide range of analytic methods provided by FogifySDK, users extract useful insights about QoS, cost, and predictive analytics. Finally, users may integrate other libraries, like scikit-learn, to endrose their analysis with ML and AI models.