Adaptive Management of Multimodel Data and Heterogeneous Workloads

Data management systems are facing a growing demand for a tighter integration of heterogeneous data from different applications and sources for both operational and analytical purposes in real-time. However, the vast diversification of the data management landscape has led to a situation where there is a trade-off between high operational performance and a tight integration of data. The difference between the growth of data volume and the growth of computational power demands a new approach for managing multimodel data and handling heterogeneous workloads.
With PolyDBMS we present a novel class of database management systems, bridging the gap between multimodel database and polystore systems. This new kind of database system combines the operational capabilities of traditional database systems with the flexibility of polystore systems. This includes support for data modifications, transactions, and schema changes at runtime. With native support for multiple data models and query languages, a PolyDBMS presents a holistic solution for the management of heterogeneous data. This does not only enable a tight integration of data across different applications, it also allows a more efficient usage of resources. By leveraging and combining highly optimized database systems as storage and execution engines, this novel class of database system takes advantage of decades of database systems research and development.
In this thesis, we present the conceptual foundations and models for building a PolyDBMS. This includes a holistic model for maintaining and querying multiple data models in one logical schema that enables cross-model queries. With the PolyAlgebra, we present a solution for representing queries based on one or multiple data models while preserving their semantics. Furthermore, we introduce a concept for the adaptive planning and decomposition of queries across heterogeneous database systems with different capabilities and features.
The conceptual contributions presented in this thesis materialize in Polypheny-DB, the first implementation of a PolyDBMS. Supporting the relational, document, and labeled property graph data model, Polypheny-DB is a suitable solution for structured, semi-structured, and unstructured data. This is complemented by an extensive type system that includes support for binary large objects. With support for multiple query languages, industry standard query interfaces, and a rich set of domain-specific data stores and data sources, Polypheny-DB offers a flexibility unmatched by existing data management solutions.