HeatWave—In-Memory Query Accelerator with Built-in ML

100X faster data recovery

When data is loaded from MySQL into HeatWave, a copy of the in-memory representation is made to the scale-out data management layer built on the OCI object store. Changes made to data in MySQL are transparently propagated to this data layer. When an operation requires reloading of data to HeatWave, such as during error recovery, data can be accessed from the HeatWave data layer, in parallel, by multiple HeatWave nodes. This results in a significant improvement in performance. For example, for a 10 TB HeatWave cluster, the time it takes to recover and reload data reduces from 7.5 hours to 4 minutes—an improvement of over 100X.

No changes required to applications with built-in analytics

HeatWave is designed as a MySQL pluggable storage engine, which completely shields all the low-level implementation details from customers. As a result, applications and tools seamlessly access HeatWave through MySQL, using standard connectors. HeatWave supports the same ANSI SQL standard and ACID properties as MySQL, and supports diverse data types. This enables existing applications to take advantage of HeatWave without any changes.

Hybrid cloud—OLTP on-premises, OLAP in the cloud

On-premises customers who can’t move their MySQL deployments to a cloud because of compliance or regulatory requirements can still leverage HeatWave by using the hybrid deployment model. In such a hybrid deployment, customers can use MySQL replication to replicate on-premises MySQL data to HeatWave without the need for ETL.

In-database machine learning with AutoML

With in-database machine learning in MySQL HeatWave, available at no extra cost, users don’t need to move data to a separate machine learning service such as Amazon SageMaker—accelerating their ML initiatives, increasing security, and reducing costs. They can apply machine learning training, inference, and explanation to data stored both inside MySQL and in the object store with HeatWave Lakehouse. HeatWave AutoML automates the machine learning lifecycle, including algorithm selection, intelligent data sampling for model training, feature selection, and hyperparameter tuning—saving customers significant time and effort.

Developers and data analysts can build machine learning models using familiar SQL commands; they don’t have to learn new tools and languages. Additionally, HeatWave AutoML is integrated with popular notebooks such as Jupyter and Apache Zeppelin. HeatWave AutoML delivers predictions with an explanation of the results, helping organizations with regulatory compliance, fairness, repeatability, causality, and trust.

Generative AI with MySQL HeatWave vector store

Currently in private preview, the vector store will enable customers to leverage the power of large language models (LLMs) with their proprietary data to get answers that are more accurate than using models trained on only public data. With generative AI and vector store capabilities, customers can interact with MySQL HeatWave in natural language and efficiently search documents in various file formats in HeatWave Lakehouse.

The vector store ingests documents in a variety of formats, including PDF, and stores them as embeddings generated via an encoder model. For a given user query, the vector store identifies the most similar documents by performing a similarity search against the stored embeddings and the embedded query. These documents are used to augment the prompt given to the LLM so that it provides a more contextual answer.

• Read the blog post

Consistent high performance and reduced costs with no downtime

Real-time elasticity enables customers to increase or decrease the size of their HeatWave cluster by any number of nodes without incurring any downtime or read-only time. The resizing operation takes only a few minutes, during which time HeatWave remains online, available for all operations. Once resized, data is downloaded from object storage, automatically repartitioned among all available cluster nodes, and becomes immediately available for queries. As a result, customers benefit from consistently high performance, even at peak times, and reduce costs by downsizing their HeatWave cluster when appropriate—without incurring any downtime or read-only time. Customers aren’t constrained to overprovisioned instances forced by rigid sizing models offered by other cloud database providers. With efficient data reloading from object storage, customers can also pause and resume their HeatWave cluster to reduce costs.

Fast analytics across databases and object storage

MySQL HeatWave Lakehouse enables users to query half a petabyte of data in object storage—in a variety of file formats, such as CSV, Parquet, Avro, and export files from other databases. The query processing is done entirely in the HeatWave engine, enabling customers to take advantage of HeatWave for non-MySQL workloads in addition to MySQL-compatible workloads. Customers can query data in various formats in object storage, transactional data in MySQL databases, or a combination of both using standard SQL commands. Querying the data in object storage is as fast as querying the databases. With HeatWave AutoML, customers can use data in object storage, the database, or both to automatically build, train, deploy, and explain ML models—without moving the data to a separate ML cloud service. The HeatWave cluster scales to 512 nodes to process half a petabyte of data, and the data isn’t copied to the MySQL database.