– KAIST Department of Computer Science Professor Minsoo Kim’s Research Team Develops ‘Chimera’ Technology
[Herald Economy=Reporter Koo Bonhyuk] A domestic research team has successfully developed a next-generation graph-relational DB system that can solve limitations such as cost burden, data inconsistency, and the complexity of processing composite queries all at once. With the application of this technology, AI is expected to be capable of real-time inference of complex connections beyond simple searches, enabling the implementation of smarter AI services.
KAIST announced on the 8th that Professor Minsoo Kim’s research team from the Department of Computer Science developed a new DB system, ‘Chimera’, which completely integrates relational DB with graph DB to execute graph-relational queries more efficiently. Chimera has demonstrated query processing performance at least 4 to 280 times faster than existing systems in international performance standard benchmarks, setting a world-class record.
Unlike traditional relational DBs, graph DBs represent data using nodes and edges, making them strong in analyzing and inferring information complicatedly intertwined like person, event, place, and time. Thanks to this feature, its use is swiftly expanding in various fields such as AI agents, SNS, finance, and e-commerce.
Alongside this, as the demand for processing composite queries between relational DB and graph DB grows, a new standard language ‘SQL/PGQ’, extending graph query functions to relational query language (SQL), has been proposed.
SQL/PGQ is a new standard language that adds graph traversal functions to the existing database language (SQL), designed to query (search) table-formatted data as well as connection information of people, events, and places all at once. With this, complex relationships such as “Which company does the friend of a friend of this person work for?” can be searched much more simply than before.
The problem is that previous approaches relied on mimicking graph traversal with cumbersome join operations or pre-constructing graph views in memory. With the former, performance drastically decreases as the exploration depth increases and, with the latter, execution fails with even slightly larger data sizes due to memory limitations.
‘Chimera’ fundamentally resolved these limitations. The research team newly designed both the storage layer and the query processing layer of the database.
The team introduced a ‘dual store structure’ that runs a graph-specific storage along with a relational data storage. They applied a ‘search-join operator’ for simultaneously processing graph traversal and relational operations, enabling complex operations to be efficiently executed within a single framework. As a result, Chimera positioned itself as the world’s first completely integrated graph-relational DB system, combining all processes from data storage to query processing.
Professor Minsoo Kim stated, “As data interconnections become increasingly complex, the need for integrated technology that encompasses graph and relational DB grows.” He added, “Chimera is a technology that fundamentally addresses this issue and is expected to be widely used in various industries such as AI agents, finance, and e-commerce.”