Paper published in 1st International Workshop on Edge Migration and Architecture (EdgeWays 2020)

Abstract: The rise of Big Data, Internet of Multimedia Things (IoMT),
and Deep Neural Network (DNN) enabled the growth of DNN-based
Computer Vision solutions to Multimedia Event Processing (MEP) applications. When these are applied to a real-world scenario we notice the
importance of having a system with a satisfactory speed that can fit in
the limited resources of most IoMT devices. However, most solutions for
distributed MEP are dependent on a Cloud architecture, which makes
these applications migration to the Edge more challenging. As a response
to this, we present a microservice architecture for DNN-based distributed
MEP over heterogeneous Cloud-Edge environments. We describe our solution that allows for an easier deployment both on the Edge and on
the Cloud. We show that choosing the proper tools for an Edge-Friendly
solution can lead to 100 times less resource utilisation. Our preliminary
investigation shows promising results, with a reduction in energy consumption by 8% with a minor drawback of 15% in throughput in the
Edge and a negligible increase in energy consumption on the Cloud.