Oracle’s MySQL HeatWave is recognized as the only cloud database service capable of delivering transactions, real-time analytics across data warehouses and data lakes, and machine learning in one MySQL database. All this comes without the complexity, risks, or costs of ETL duplication. Now MySQL HeatWave—dubbed a potential “killer app” for Oracle in a Forbes article—takes another leap forward with a new vector store, generative AI capabilities, MySQL HeatWave Lakehouse availability on Amazon Web Services (AWS), and more.
Support for generative AI, along with the ability to augment large language model (LLM) training with a business’s own data, was the marquee announcement for MySQL HeatWave at Oracle CloudWorld. “Today’s enhancements to MySQL HeatWave are another significant step on our journey to address pressing customer data, analytics, and AI issues,” said Edward Screven, Oracle’s chief corporate architect. “Now vector store and generative AI bring the power of LLMs to customers, providing them with an intuitive way to interact with data in their enterprise and get the accurate answers that they need for their business.”
“Now vector store and generative AI bring the power of LLMs to customers, providing them with an intuitive way to interact with data in their enterprise and get the accurate answers that they need for their business.”
The vector store ingests enterprise data from a data lakehouse, then stores it as embeddings generated via an encoder for use as additional context for LLM queries. User prompts to LLMs undergo a similarity search against the vector store. The result is used as an input query that combines the original prompt and the similarity search, allowing the LLM to deliver a more contextual, relevant answer derived from enterprise data.
“The vector store output is just an input to the LLM,” said Nipun Agarwal, an Oracle senior vice president for MySQL HeatWave development. “LLMs are not trained on this proprietary data, so you don’t run the risk of any information leakage.”
Currently in private preview, the combination of vector store and generative AI represent a powerful tool to help organizations leverage proprietary data for queries and searches, ultimately leading to more accurate responses to natural language queries.
Launched in July 2023 on Oracle Cloud Infrastructure (OCI), MySQL HeatWave Lakehouse accelerates analytics on object stores’ data across various formats without importing it to MySQL, making HeatWave accessible for even non-MySQL workloads. Now AWS customers can enjoy these same capabilities because MySQL HeatWave Lakehouse runs natively on the platform.
“We use AWS infrastructure to run MySQL HeatWave in AWS, including HeatWave Lakehouse,” said Screven. “That means that if you use HeatWave Lakehouse in AWS, there is no data egress fee. You can analyze files that are stored in S3 buckets.”
Hundreds of terabytes of data in CSV, Apache Parquet, Avro, and other formats—including exports from other databases—can be natively queried, replacing five AWS services (Aurora, Redshift, Glue, SageMaker, and Athena) with just one to reduce both cost and complexity. MySQL HeatWave is also available to Microsoft Azure users as part of Oracle Database Service for Microsoft Azure.
“MySQL HeatWave can now directly query data stored in files in object store. That means you do not need to ingest it in MySQL,” said Screven. “That also means that MySQL HeatWave Lakehouse becomes a service that you can use for all your data: IoT devices, log files, exports from other databases.”
MySQL Autopilot Indexing is another new feature that Screven announced at CloudWorld. Currently in limited availability, Autopilot Indexing speeds up and improves indexing by using machine learning to make recommendations to create and drop indexes. Autopilot Indexing also generates performance impact assessments alongside recommendations to help inform tuning decisions. Other recently announced features for MySQL HeatWave include.
Clay Magouyrk and Edward Screven discuss Oracle’s latest advances in cloud technology.