A Tracing Based Model to Identify Bottlenecks in Physically Distributed Applications

International audienceThe Cloud computing paradigm has become the new industry standard way of designing large scale applications. Over the past years, we observe an increased adoption of this technology on numerous IoT-Edge applications. And while this technology comes with its promises and benefits, considering almost infinite scalability, it also comes along with its drawbacks and challenges. Detecting partial failures or bottlenecks are new obstacles that arose with the adoption of Cloud Applications. Distributed Tracing now allows developers to gain insight on the composition of services within a distributed Application. Today we observe an increased adoption of this technology on numerous cloud-native architectures. The project OpenTelemetry proposes a specification for traces that normalizes this new monitoring data format. In this publication we present an approach that leverages these traces to identify bottlenecks at the scale of a physically distributed application. We propose an extension of our model that builds a hierarchical property graph to exhibit bottlenecks in an application that follows the layered Cloud-IoT network model. Based on OpenTelemetry traces we can maintain a model at runtime of the whole application and compute bottlenecks. Their identification relies on the scores provided by centrality algorithms