A Cloud-Native Data Lakehouse Architecture for Big Knowledge Graph OLAP

Knowledge graphs (KGs) represent real objects and the relationships to each other. KGs are often bound to specific contexts. This is the case in air traffic management (ATM) where knowledge is inherently coupled to a context consisting of dimensions such as a location, time or topic. This fact led to the development of a generally applicable technique called KG-OLAP (online analytical processing). KG-OLAP provides a multidimensional view on contextualized knowledge graphs and enables contextual and graph operations on the resulting KG-OLAP cube. The proof-of-concept prototype with GraphDB published on GitHub demonstrates the functionality of KG-OLAP. However, the prototype is not feasible for big data which makes is unsuitable for data-intensive applications. For example, in Europe alone, over ten billion RDF triples of ATM knowledge are generated annually. With that not only the volume is a problem but also the speed of data generation as well as the semi- and unstructured nature of ATM data types. The goal of this thesis is to propose and implement a generic architecture that meets both KG-OLAP and big data requirements. The first contribution of this thesis is the Big KG-OLAP reference architecture including process definitions for the following main functionalities: data ingestion and contextual operations Slice’n’Dice and Merge. A prototypical cloud-native implementation of the proposed architecture deployed on Amazon Web Services and demonstrated using an ATM use case. The third and last contribution is a performance evaluation of the main functionalities testing their scalability.