TLDR: Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed ‘Chimera,’ a novel database system that seamlessly integrates relational and graph databases. This innovation addresses critical limitations in processing complex queries for large AI models, demonstrating performance up to 280 times faster than existing systems and enabling real-time reasoning for advanced AI services. The technology is poised for immediate industrial application, particularly in high-performance AI agents based on Retrieval-Augmented Generation (RAG).
The Korea Advanced Institute of Science and Technology (KAIST) has announced a significant leap in database technology with the development of ‘Chimera,’ a next-generation graph-relational database system designed to dramatically enhance the intelligence and efficiency of AI agents. This breakthrough, led by Professor Min-Soo Kim’s research team, directly tackles the growing challenges faced by companies integrating large AI models with traditional data management systems.
For years, relational databases (DBs) have been the backbone of data management. However, the increasing complexity and scale of AI models necessitate the integration of graph databases, which excel at representing and analyzing intricate relationships. This hybrid approach has historically been plagued by issues such as high cost, data inconsistency, and the difficulty of executing complex queries efficiently. ‘Chimera’ offers a fundamental solution to these problems.
Professor Min-Soo Kim’s team revealed on September 8th that their new system fully integrates relational and graph databases, enabling the efficient execution of graph-relational queries. The system has showcased world-class performance, processing queries at least 4 times and up to an astonishing 280 times faster than existing systems in international performance standard benchmarks, specifically the LDBC Social Network Benchmark (SNB).
Graph databases, with their vertex (node) and edge (connection) structure, are particularly adept at analyzing complexly intertwined information, such as relationships between people, events, places, and time. This makes them invaluable for applications in AI agents, social networking services (SNS), finance, and e-commerce. The demand for integrated query processing between these two database types has even led to the proposal of a new standard language, ‘SQL/PGQ,’ which extends SQL with graph query functions.
Chimera’s innovative architecture addresses the limitations of previous methods, which either mimicked graph traversal with join operations (leading to performance degradation with increased depth) or relied on pre-building in-memory graph views (resulting in memory constraints and poor data freshness). The KAIST team redesigned both the storage and query processing layers of the database.
Key to Chimera’s design is a ‘dual-store structure’ that simultaneously operates graph-specific storage and relational data storage. Furthermore, a ‘traversal-join operator’ was developed to process graph traversal and relational operations concurrently within a single system. This makes Chimera the world’s first graph-relational DB system to integrate the entire process from data storage to query processing.
Beyond its speed, Chimera eliminates query failures due to insufficient memory, regardless of graph data size. By not relying on views, it also ensures immediate reflection of changes to original data, guaranteeing excellent data freshness.
Professor Min-Soo Kim emphasized the significance of their work, stating, “As the connections between data become more complex, the need for integrated technology that encompasses both graph and relational DBs is increasing. Chimera is a technology that fundamentally solves this problem, and we expect it to be widely used in various industries such as AI agents, finance, and e-commerce.”
The research paper, titled ‘Chimera: A System Design of Dual Storage and Traversal-Join Unified Query Processing for SQL/PGQ,’ was co-authored by Ph.D. student Geonho Lee (first author) and Jeongho Park, an engineer at Professor Kim’s startup GraphAI Co., Ltd. (second author), with Professor Kim serving as the corresponding author. The findings were presented on September 1st at VLDB, a globally recognized international academic conference in the database field. The paper’s DOI is https://dl.acm.org/doi/10.14778/3705829.3705845.
This technology is expected to have an immediate industrial impact, particularly as a core component for ‘high-performance AI agents based on RAG (Retrieval-Augmented Generation),’ a smart AI assistant with advanced search capabilities. It will be integrated into ‘AkasicDB,’ a vector-graph-relational DB system slated for release by GraphAI Co., Ltd.
Also Read:
- Global Graph Database Market Projected to Soar to $16 Billion by 2032
- Weaviate Unveils 8-bit Rotational Quantization for Enhanced AI Efficiency
The research received support from the Ministry of Science and ICT’s IITP SW Star Lab and the National Research Foundation of Korea’s Mid-Career Researcher Program.
