Introduction
Modern organizations, particularly in the healthcare sector, face a "perfect storm" of data management challenges: massive datasets, thousands of concurrent users, and highly complex, heterogeneous security policies. Traditional Role-Based Access Control (RBAC) models often prove too static for dynamic real-world scenarios, such as emergency medical access. While Semantic Web technologies like OWL and RDF offer the interoperability and reasoning capabilities necessary for complex rule management, they frequently encounter scalability bottlenecks. As the number of individuals (the ABox) grows, standard reasoners often suffer from memory exhaustion or prohibitive latency. This paper presents a distributed, ontology-based framework for Temporal Role-Based Access Control (TRBAC) designed to achieve near-constant response times for security policy enforcement at scale.
Methods
The proposed architecture shifts the computational burden away from a single global reasoner through a three-stage process:
- Partitioning: The global Knowledge Base (KB) is divided into $n$ smaller KBs. These partitions share a common schema (TBox) but contain distributed data instances (ABox).
- Modular Inference: A dedicated inference KB ($KB_{inf}$) stores Semantic Web Rule Language (SWRL) rules defining organizational security policies.
- Distributed Reasoning & P-Links: To facilitate cross-partition navigation without merging, the authors introduced "P-links" (partition links). Upon an access request, the system utilizes SPARQL queries and indexing (Lucene/LARQ) to retrieve only the relevant triples from disk into memory for localized reasoning.
This approach isolates reasoning to a minimal subset of data relevant to a specific session, bypassing the need to process the entire organizational database simultaneously.
Results
The framework was evaluated using synthetic healthcare data mimicking large-scale environments with thousands of users and complex individual properties. Key performance metrics focused on reasoning time as a function of ABox size and rule complexity.
As shown in Table 1, the system demonstrates efficient scaling. While a naive in-memory reasoner typically fails due to memory exceptions when reaching 1,000 individuals, the partitioned approach maintains response times between 152ms and 552ms for loads up to 1,008 individuals. Notably, the time taken for policy decisions remains relatively stable even as the population of patients and doctors increases. Furthermore, increasing the number of security rules up to 1,000 had a negligible impact on performance, as the system selectively fires only those rules pertinent to the active session.
Discussion and Conclusion
The primary insight of this work is that the ABox constitutes the main bottleneck in security reasoning. By partitioning data while maintaining a shared TBox, the system achieves significant scalability. The use of "local knowledge" allows for rapid decision-making without global context. Additionally, the framework adheres to Open World Assumption (OWA) safety by requiring explicit proof for access grants, thereby preventing security vulnerabilities inherent in incomplete knowledge bases.
Future work will address TBox scaling, as the current system still faces challenges when the schema itself becomes excessively complex. The authors also plan to validate the framework across diverse domains beyond healthcare to demonstrate its general applicability.
Summary updated