Oracle has announced the general availability of its Oracle Globally Distributed Exadata Database on Exascale Infrastructure. This new architecture is geared for new-generation agentic AI use cases that need to scale rapidly and unpredictably, said Wei Hu, senior vice president, high availability technologies, Oracle.
“Agentic AI workloads place immense demand on the backend systems with parallel requests,” Hu told AIM in an exclusive interview.
The service will be available across all Oracle Cloud commercial regions, including India.
The new offering combines Oracle’s distributed database technology with its Exascale infrastructure to help manage, scale, and reduce the cost of mission-critical workloads. Moreover, the distributed database allows organisations to store and process data across different geographies while presenting a single logical view to the application.
These workloads are highly bursty. “They have a big peak and then it dies off,” he said. The serverless architecture allows the system to scale from zero to hyperscale and back again without manual intervention.
SQL Support
Wu said that Oracle’s distributed database offers full enterprise-grade SQL support natively, unlike other systems that layer SQL on top of NoSQL architectures, resulting in poor performance.
When asked about alternatives like PostgreSQL and emerging players like CockroachDB and TiDB, Hu was direct. He argued that while many developers gravitate towards PostgreSQL because it’s “theoretically free,” it often ends up being more difficult and costly to operate due to its lack of advanced features.
“Oracle is a superset of Postgres,” he said, adding that many tasks are significantly easier on Oracle because Postgres simply doesn’t have the same capabilities.”
Wu said that CockroachDB started with RocksDB and later switched to Pebble, but the architecture remains the same and is based on a key-value store that does not support SQL. He explained that this architectural mismatch results in performance and functionality limitations.
“We’re more feature-rich and architecturally better. Our customers often find CockroachDB lacking in functionality, which forces them to re-architect their applications. With Oracle, they get full compatibility and don’t have to worry about missing data types or subset SQL.”
He further added that the Oracle offers six distribution methods and three replication methods. This allows for precise control over how data is stored and replicated, based on application and network requirements.
Besides that, Wu said Oracle supports multiple workloads, including AI, OLTP, and analytics within the same distributed database. This eliminates the need for managing separate specialised databases, reducing complexity.
“With the 23ai release, we support this thing called graph-based replication. It implements an active-active-active symmetric configuration… In case of failure, we will automatically fail over to a replica in less than three seconds with zero data loss.”
What about Vector databases?
Regarding the need for specialised vector databases, Wu said vector search will likely become a built-in feature of general-purpose databases rather than a standalone product, reducing the need for developers to manage multiple specialised systems.
“People came with specialised JSON document stores, specialised graph database stores, specialised vector stores, specialised text stores. This makes it very hard for developers… you end up using five different databases,” he said. He added that Oracle promotes a converged database that supports multiple data types, including vectors, SQL, JSON and text, and multiple workloads, from transactions to analytics and AI, in a single system.
How Can it Benefit Customers?
Wu shared the example of a major US bank that needed to comply with Reserve Bank of India regulations, which require that financial data involving Indian citizens must reside within the country.
He explained that using Oracle’s system, the bank now maintains Indian data in Indian regions and the rest of the global data in the US, while operating as one unified database. Wu said that the application involved hundreds of microservices, making cross-country replication unfeasible. Oracle’s distributed architecture allowed the company to isolate Indian data locally while maintaining a unified system.
Wu revealed that similar deployments are currently being implemented in four additional countries. Another customer, a global credit card company, uses the architecture for an active-active-active setup with three replicas per region. “They get automatic three-second failovers with zero data loss using our new replication mechanism,” Wu said, adding that it is essential for use cases like credit card transactions, where downtime is unacceptable.
The post Why Oracle Thinks PostgreSQL Isn’t Enough appeared first on Analytics India Magazine.
