With the increased usage of digital technologies, the analysis of data in enterprises becomes more important. As big enterprises usually are made up of multiple autonomous divisions, multiple data landscapes evolve, which hinders activities related to metadata across them. Knowledge graphs are suitable for integrating metadata from different sources. While current work discusses different approaches and the motivation for metadata management based on knowledge graphs, there is currently a lack of discussion of the architecture of the implemented knowledge graph. In this work, different architectures for a metadata management system based on knowledge graphs are identified, discussed, and compared. Therefore, based on potential use cases criteria has been derived. To complement the qualitative comparison with experimental results and test a potential implementation, the architectures have been implemented in GraphQL, and experiments have been conducted. Combined, this results in an overview of the different architectures and for which use cases they fit best.