GraphScale's Abstraktion Refinement Approach

The innovation of GraphScale is based on a so called "abstraction" that is used for reasoning and querying. This abstraction is build form the original data and represents a kind of compressed summary that is updated alongside reasoning. After computation of all conclusions these are - similar to a de-compression step - written back to the original data. The result is a completely materialized data repository able to include derived facts to queries. This procedure is provable sound and complete as described scientifically. For querying via the integrated SPARQL interface of GraphScale the abstraction provides an optimized index for efficient answer set computation.

GraphScale Architecture

GraphScale is a bridging technology that adds high-performance and expressive reasoning to triple and databases systems. This task is not trivial both conceptually as well as practically if real-world requirements such as scalability and flexibility are take serious. GraphScale is the first system that effectively connects reasoning about OWL 2 with existing data storages at industry scale. The accompanying figure shows the general architecture of a GraphScale solution. The system relies on well-established interface standards in order to support a broad range of reasoning as well as database systems directly. Via OWL-API and OWLlink practically all OWL 2 engines such as Konclude can be used to reason with the abstraction. For the data back-end a SPARQL or SQL interface is sufficient. For SPARQL querying the resulting solution there are two options. Firstly, it is possible to to query the materialized data within the data store or secondly via the built-in, optimized SPARQL engine of GraphScale.<br/> As a consequence of the component architecture an existing semantic application is easily extended with GraphScale technology since the technical interfaces of the central processing store (such as SPARQL) remain stable.

GraphScale Facts

  • Injects expressive reasoning to triple and data stores
  • Massive parallel architecture for high-performance processing
  • In-memory option for maximal performance
  • Interfaces to RDBMs, NoSQL DBs, tripels stores
  • Sound and complete for high expressivity (OWL 2 RL +)
  • Optimized SPARQL interface