Oracle expands vector support in latest cloud database release – Business

Oracle Corp.’s flagship database management system is now available as a cloud service. Oracle Database 23ai features vector search and more than 300 additional major features, with many focused on simplifying the development and use of artificial intelligence models.

Vector search is used in information retrieval and machine learning to find and rank items based on their similarity to a query. Instead of searching text directly, vector search represents both the items in the database and the query as vectors in a high-dimensional space that captures the semantic or contextual meaning of the items. The technique is used to train large language models and works with various unstructured data types.

Oracle said today that vector search can now be used to securely combine searches for documents, images, and other unstructured data with searches on private business data without the need for duplication or data movement. Business and vector data are stored and processed on the same database.

Developers can add semantic search capabilities to new and existing applications. Oracle said “Smart Storage” capabilities in its Exadata System Software 24ai significantly accelerate vector search and enable its use across large volumes of data and users.

Vector search is useful in settings such as retail, where a seller can turn a product catalog into vectors that can provide recommendations if an item is out of stock, said Jenny Tsai-Smith, vice president of overall database product management at Oracle. “ChatGPT has caught everyone’s imagination, and a lot of companies have come to us saying they have ideas about how to use these [vector] capabilities,” she said.

Steve Zivanic, global vice president of database and autonomous services product marketing, said vector capabilities are rapidly becoming a standard feature on all database platforms. “Vector databases are a feature, not a market,” he said.

Real-time replication

Also new in this release is an updated version of GoldenGate, a replication component used for high-availability scenarios and heterogeneous data integration across clouds. GoldenGate 23ai’s new capabilities enable users to create a real-time data fabric for retrieval-augmented generation, a type of AI fine-tuning.

 “Because generative AI is trained on data at a snapshot in time, it becomes outdated,” Tsai-Smith said. “With this feature, you can have real-time data converted into vectors for more timely responses.”

Developers can now read and write relational data using JavaScript Object Notation as well as SQL.  JSON queries can use Representational State Transfer or native APIs without compromising data consistency, Oracle said.

“Developers tend to want to forget that they’re using relational DBs,” Tsai-Smith said. “They take a JSON document that has everything they need to make changes and put it back in the DB. With JSON duality, we are storing data as relational tables to reduce data duplication.”

Developers can now also build applications that navigate connections between and within data using property graph queries. A property graph is a type of data structure used primarily in graph databases and graph processing frameworks that represents relationships between data elements and nodes in ways that can’t easily be expressed in conventional relational tables.

Property graph queries can run on top of all types of data supported by Oracle Database, including relational data, JSON data and spatial data. Developers can define a graph model directly on operational data and query it using SQL/PGQ, a graph query language built on top of SQL. Oracle also intends to increase support for the new GQL standard over time.

Raft, a data replication capability that integrates with transaction execution in a sharded database, now enables automatic failover with zero data loss in under 10 seconds. The feature is critical when building cloud-scale distributed databases with high scalability and availability requirements. Oracle Globally Distributed Database allows data in the cloud to be stored across multiple physical databases in multiple locations while exposing a single database image to applications.

Oracle Database 23ai is available in Oracle Cloud Infrastructure on Oracle Exadata Database Service, Oracle Exadata Cloud@Customer and Oracle Base Database Service, and on Oracle Database@Azure.