Parallel Materialization of Large ABoxes

Area of Science:

• Knowledge representation and reasoning
• Database systems
• Parallel computing

Background:

• Efficient materialization is crucial for large knowledge bases.
• Existing systems face scalability challenges with large ABoxes.
• Parallel processing offers a potential solution for computational bottlenecks.

Purpose of the Study:

• To develop and evaluate a framework for efficient knowledge base materialization on shared-nothing parallel machines.
• To address the computational demands of large ABoxes.
• To leverage TBox axiom parallelism for performance gains.

Main Methods:

• A framework for parallel materialization on shared-nothing systems.
• Partitioning of TBox and ABox axioms using a min-min strategy.
• Integration with existing inference systems (e.g., SwiftOWLIM) for local computations.
• Coordination of information exchange between processors.

Main Results:

• The framework demonstrates speedup in a cluster environment by exploiting TBox axiom parallelism.
• Experimental evaluation using Lehigh University Benchmark (LUBM) datasets.
• The approach's performance is influenced by the complexity of the TBox.

Conclusions:

• The proposed framework offers an efficient approach to knowledge base materialization in parallel environments.
• Parallelism in TBox axioms can be effectively exploited for performance improvements.
• TBox complexity remains a limiting factor for this parallel materialization strategy.