Execution Traces of an MNIST Workflow on a Serverless Edge Testbed

For the evaluation of a Serverless Edge Computing platform, we built an Edge Cloud testbed consisting of several machines: A Cloud VM An Nvidia Jetson TX Four Raspberry Pi 3b+ Two Intel NUCs with i5 processors We were interested in profiling these devices with a Machine Learning workflow deployed as a serverless application. To that end, we implemented three functions: Preprocessing, Training, and Serving as OpenFaaS functions. The workflow trains an MNIST model.