Getty Images/iStockphoto

Oracle HeatWave ML fuses machine learning with database

The tech giant is continuing to build out the capabilities of its MySQL HeatWave cloud database with real-time elasticity, performance boosts and machine learning integration.

Oracle today released new machine learning integrations with its MySQL HeatWave cloud service.

The tech giant originally launched the MySQL HeatWave service in December 2020 as the Oracle MySQL Database Service with MySQL Analytics Engine. The company rebranded it as HeatWave in 2021.

The service runs the MySQL database as a managed offering on Oracle Cloud Infrastructure (OCI) and enables users to run both online transaction processing (OLTP) and OLAP workloads. Oracle has steadily added new features to the HeatWave platform, including automation features the company branded as "Autopilot" in Aug. 2021.

As part of the new update, generally available now, Oracle integrated machine learning (ML) capabilities into the MySQL HeatWave platform. With HeatWave ML, users can directly run ML workloads on the database service.

The MySQL HeatWave platform contains many "elegant" capabilities, said Holger Mueller, an analyst at Constellation Research.

The ability to support both OLTP and OLAP data workloads is one such set of capabilities, and Mueller noted that Oracle has expanded that by adding machine learning inside the MySQL HeatWave database service.

Screenshot of Oracle HeatWave ML database
With HeatWave ML, Oracle is directly integrating machine learning capabilities with its MySQL cloud database service.

MySQL HeatWave ML goes beyond Autopilot

Customer feedback first led to the release of the MySQL Autopilot features in 2021 and has now led to the integration of machine learning with HeatWave ML, said Nipun Agarwal, senior vice president, MySQL HeatWave development at Oracle.

With the introduction of HeatWave, customers are now storing more data into the MySQL database; and for many of these customers, they need to run machine learning on this data.
Nipun AgarwalSenior vice president, MySQL HeatWave development at Oracle

"With the introduction of HeatWave, customers are now storing more data into the MySQL database; and for many of these customers, they need to run machine learning on their data," Agarwal said.

He explained that without HeatWave ML, users had to first extract data out of MySQL into a machine learning-capable platform to run ML training or inference workloads. That's no longer the case, due to the HeatWave ML capability.

HeatWave ML uses AutoML technology that is already supported in the Oracle Autonomous Database platform.

With Oracle's AutoML, users can build out machine learning pipelines. The ML technology is now stored inside the MySQL database and the machine learning processing is also executed directly within the database.

By handling machine learning operations within the database, Agarwal said users deal with less complexity for users, as they don't need to export data.

There are also security benefits to handling machine learning inside the database, as the access controls for the MySQL database still apply. By contrast, when the data had to be exported into a separate ML tool, users needed to set up and configure access control again.

Real-time elasticity and data processing advances for MySQL HeatWave

In addition to the HeatWave ML capabilities, Oracle is added new real-time elasticity and data processing improvements.

The MySQL HeatWave service runs on OCI, which can scale resources up or down as demand warrants. Agarwal explained that the new real-time elasticity feature in MySQL HeatWave will enable database deployments to scale up with new database nodes when needed without interrupting a running workload.

"The infrastructure enables the elasticity, but if we don't do anything at our layer, the queries will get interrupted when the cluster makes the transition and add a node," Agarwal said. "So that's the work we have done."

Oracle also increased the amount of data that can be processed per database node, boosting processing capability from approximately 400 gigabytes (GB) of data per node to 820 GB with a memory optimization approach known as a Bloom Filter.

The vendor also added support for the open source LZ4 compression algorithm, which further optimizes data in a MySQL HeatWave node.

"What we have done is [that] we have doubled the amount of data that can be processed per node," Agarwal said.

Dig Deeper on Database management

Business Analytics
Content Management