Databricks has launched 'Lakehouse//RT (Real-time)', which performs real-time analysis directly at Lakehouse. It is characterized by processing millisecond-level analysis on governance-assisted Delta Lake and Apache Iceberg tables without the need for a separate real-time serving layer.

Databricks released a beta version of Lakehouse//RT, based on its new computing engine 'Reyden,' on the 20th. The company stated that the solution is designed to provide low-latency queries to tens of thousands of concurrent users and agents.

It was reported that it recorded a latency of less than 100 milliseconds even when processing 12,000 queries per second based on standard analytics benchmarks, confirming up to 16 times improved performance compared to the existing real-time serving stack.

Companies requiring high concurrency and low latency have been building a real-time serving layer separate from the lakehouse.

Databricks explained that this structure causes limitations such as vendor lock-in, increased infrastructure costs, governance fragmentation, and the persistence of data copies.

Lakehouse//RT stated that it directly queries the Delta and Iceberg tables to access the latest data without moving data.

All queries run within the Unity Catalog governance framework, which includes policies, permissions, and auditing.

The company added that it can reduce the cost and complexity associated with maintaining the serving layer by eliminating the need for a separate authorization layer, proprietary format, synchronization, or CDC pipeline.

Raiden's asynchronous execution model is said to provide response speeds of as little as 10 milliseconds for small datasets and 100 milliseconds for large datasets.

Unlike existing engines optimized for simple queries, performance technology across complex analysis By applying it and specifying an existing table, you can query immediately without a separate data collection process.

Databricks co-founder and CEO Ali Godsh said, “Lakehouse//RT completes the full spectrum of engines, providing the millisecond-level speed layer that people want and agents need.”

Case studies of client applications were also revealed.

Cisco reported that it confirmed a five-fold improvement in response speeds for threat detection queries, and Magnite stated that it achieved performance of less than 200 milliseconds in core dashboard queries.

Lakehouse//RT is currently available in beta version.