HeatWave on OCI vs. Aurora

Oracle HeatWave is a high performance, highly scalable, in-memory cloud service that lets organizations build and run applications for transaction processing, analytics across data warehouses and data lakes, generative AI, and machine learning from a single service. HeatWave customers can get real-time analytics on their transactional data in MySQL Database without duplicating it via extract, transform, and load (ETL) to a separate analytics database. That reduces security risks and cost while query performance improves via HeatWave’s in-memory accelerator.

For artificial intelligence–powered applications, users can take advantage of an automated, in-database vector store to perform retrieval-augmented generation (RAG), which lets AI answer queries using a company’s own data. Developers and analysts without AI expertise can use in-database large language models (LLMs) for AI development. And businesses can deploy HeatWave on Oracle Cloud Infrastructure (OCI), Amazon Web Services, and Microsoft Azure. HeatWave MySQL executes complex analytics queries on a TPC-H decision-support benchmark faster than Amazon Aurora.

HeatWave also supports the same tools for business intelligence and data visualization as MySQL Database, including Oracle Analytics Cloud, Tableau, and Looker. Here are the top reasons that customers choose HeatWave on OCI over Aurora.

  • Integrated, automated, and secure generative AI: Heatwave GenAI features include LLMs available in the database, an automated vector store, and the ability to hold natural language conversations informed by a customer’s own data.
  • Better performance with cost savings: HeatWave ran a 4 TB TPC-H analytics workload 1,400 times faster than Aurora and was shown to provide 2,200 times better price-performance as of a March 2022 benchmark test.
  • A single database for transactional and analytics workloads: HeatWave eliminates the risk, cost, and complexity of using two separate databases by combining transaction processing and analysis in the same database.
  • Machine learning-powered automation: HeatWave Autopilot uses ML techniques to automate functions, including database tuning for performance and scalability.
  • Native, in-database machine learning: HeatWave AutoML provides the ability to build, train, and explain ML models within HeatWave, eliminating the need to move data to a separate cloud service.

Integrated, automated, and secure generative AI

Businesses want to build generative AI applications informed by their own data; HeatWave GenAI lets them do so quickly, easily, and securely. Developers can benefit from the automatic creation of vector embeddings needed for RAG to provide company data that augments built-in LLMs. They can use a combination of SQL statements and vector queries, without moving data from the database. To create the same applications with Aurora requires using different cloud services, so developers must move or integrate data across these services.


Capability and evidence

HeatWave

Aurora

Can customers build generative AI applications using in-database LLMs?

  • HeatWave customers can put LLMs that ship inside the database to work in their apps, without selecting and integrating external ones. They can also choose external LLMs if desired. HeatWave helps customers cut infrastructure costs by eliminating the need for external LLMs and for provisioning GPUs.
  • AWS doesn’t provide the ability to run LLMs inside its analytics or transactional database engines: Customers must use an external service called Bedrock for selecting and integrating AI models. Using multiple cloud services can increase complexity and cost.
yes
no

Can customers automate AI vector creation and processing?

  • HeatWave automates the generation of vectors, including parsing, extracting metadata, splitting documents into “chunks,” and choosing embedding models for data and queries. Users don’t need AI expertise; they can complete all steps in the database, simplifying processes and reducing costs.
  • AWS requires manual work to create a vector store. It’s only possible in Aurora PostgreSQL (and in AWS Relational Database Service for PostgreSQL). Users require computing resources to connect to Aurora, create vectors manually using the LangChain framework, select embedding models with Amazon Bedrock, and use pgvector and LangChain to retrieve vectors. These complex, manual operations can take days, versus minutes with HeatWave.
yes
no

Can the system provide optimized resources to accelerate vector processing?

  • Vector processing is parallelized across up to 512 HeatWave cluster nodes and executed at near memory bandwidth, helping deliver fast results.
  • AWS vector processing isn’t unified in a single database instance. It requires multiple systems, making it difficult to optimize performance.
yes
no

Can customers use a chat interface for natural language conversations?

  • HeatWave Chat lets users hold natural language conversations with the system, with answers informed by their company’s own data. Developers don’t need to build back-end chat code, accelerating development.
  • AWS doesn’t provide a chat UI out of the box; users need to use a separate service to build chatbots.
yes
no

HeatWave: Faster than Aurora

Optimized for performance and scalability, HeatWave accelerates MySQL queries. It’s faster than other cloud database services, including Aurora, at lower cost.

Query Performance: 4TB TPC-H

See the performance details and learn about the benchmark.


Capability and evidence

HeatWave

Aurora

Can customers run analytics and complex queries without indexing?

  • HeatWave doesn’t require customers to index data to accelerate queries. Data in the MySQL InnoDB storage engine is transparently transformed and propagated to HeatWave, available for query acceleration as soon as it’s updated in MySQL. This reduces manual tuning by DBAs.
  • Aurora requires DBAs to manually create and maintain indexes for better query performance. As users run more complex queries, more indexes are needed. Maintaining them can be tedious and time-consuming: For a 4 TB TPC-H data set, it can take five days to create the indexes for query processing in Aurora versus just four hours to load MySQL data into HeatWave.
yes
no

Can customers run ad hoc queries efficiently?

  • HeatWave’s algorithms for distributed, in-memory, analytics processing include scans, joins, group-by, aggregation, and top-K (a SQL query that yields answers ranked in order of probability). HeatWave runs ad hoc queries as fast as planned ones, eliminating the need for manual tuning by DBAs.
  • Without knowing the queries in advance, DBAs can’t create appropriate indexes in Aurora to improve query performance.
yes
no

Can the database scale as data volume increases while still maintaining performance?

  • HeatWave is designed for massive scalability and performance. It has a highly partitioned architecture, which enables high parallelism for scaling out. HeatWave MySQL’s performance improvements increase with data size. It’s 151 times faster than Aurora for 256 GB of data, nearly 1,000 times faster for 1 TB of data, and 1,400 times faster for 4 TB of data.
  • Aurora scales out by adding database replicas for groups of users who need to access the database concurrently. But that doesn’t help with query performance.
yes
no

HeatWave vs. Aurora: Cost and performance

HeatWave MySQL provides 2,200 times better price-performance than Aurora, as demonstrated by the 4 TB data set TPC-H benchmark.

Cost Comparison: 4 TB TPC-H

See the performance details and cost comparisons


Broctagon logo

After Broctagon migrated its CRM system from Aurora to HeatWave on OCI, the Singapore-based fintech increased performance by 30% and reduced costs by 35%.



A single database for transactions and analytics

HeatWave eliminates the risk, cost, and complexity of using separate databases for transactions and analytics. HeatWave MySQL, with its integrated in-memory query accelerator, lets database administrators and developers run OLTP and OLAP workloads directly from the MySQL Database. For a 100 GB mixed workload with OLTP and OLAP queries, HeatWave MySQL is 18 times faster than Aurora. It provides 110 times more throughput than Aurora for OLAP queries while maintaining the same performance.


Capability and evidence

HeatWave

Aurora

Can customers run combined OLTP and OLAP workloads efficiently using a single database service?

  • HeatWave lets customers run OLTP and OLAP in one MySQL Database—without the need to extract, transform, and load (ETL) data to a separate database for analytics processing. No changes to existing applications are needed either.
  • Aurora customers that want to analyze their data need to use a separate OLAP database service, called Amazon Redshift. Using two databases can increase complexity and cost and introduces consistency concerns. Security and compliance risks also increase as data moves. While Aurora’s and Redshift’s zero-ETL integration simplifies the process, data still gets replicated between separate OLTP and OLAP database services.
yes
no

Can customers run real-time analytics workloads on their MySQL Database?

  • New MySQL transactions are immediately available to HeatWave for analytics queries.
  • Aurora users needing fast data analysis must move data to a separate analytics database to avoid lengthy delays from long-running queries. By the time data is available in this separate database, it’s stale. Even with zero-ETL integration between Aurora and Redshift, data replication latency can be problematic for applications that need real-time analytics.
yes
no

Can customers eliminate the cost, complexity, and risk of ETL by using a single database service?

  • HeatWave provides a platform for both analytics and transactional workloads, eliminating the need for the complex, time-consuming, expensive ETL and integration that would be required to use a separate analytics database.
  • AWS has no unified service for OLTP and OLAP, so using Aurora for OLTP and Redshift for analytics requires customers to replicate data across databases, increasing complexity, risk, and costs. AWS itself says that customers need to pay for the Aurora and Redshift resources used to create and process change data that’s part of zero-ETL integrations. That may include fees for I/O, storage, and data transfer.
yes
no
Fancom logo

For Japanese ad network FANCOMI, migrating from Aurora to HeatWave increased performance tenfold for real-time analytics while reducing costs. The company no longer needed to move data to an analytical database nor modify its application to get the performance improvement.



Real-time analytics without modifications

ISVs and enterprises need fast, accurate predictions to improve business results. Customers running MySQL applications can immediately take advantage of HeatWave for real-time analytics, without making any changes to their applications. HeatWave can run existing MySQL applications without the need for recoding, and databases and applications running on HeatWave benefit from the additional features and performance.


Capability and evidence

HeatWave

Aurora

Can customers process transactions and get real-time analytics in the same database without changing existing applications?

  • HeatWave runs existing MySQL applications with no changes, so they can benefit from real-time analytics, performance improvements, and other features.
  • Aurora isn’t designed to provide real-time analytics; to efficiently analyze data, customers must use a separate analytics database, such as Amazon Redshift.
yes
no

HeatWave Autopilot: ML-powered automation

  • HeatWave Autopilot automates many of the most important and challenging aspects of achieving high query performance at scale. It can automatically provision optimal cluster sizes for given data sets—whether the data resides in MySQL or in an object store—and handle database tuning, including performance, scalability, and query plans. Autopilot is available at no additional charge for HeatWave customers.

Capability and evidence

HeatWave

Aurora

Does the database service include built-in machine learning automation for operations?

  • HeatWave Autopilot uses ML to automate database operations, including provisioning, data loading, query processing, and error handling.
  • Aurora lacks the equivalent built-in ML-powered automation, requiring DBAs to manually provision, maintain, and tune the database.
yes
no

Does the database service include built-in machine learning automation for auto provisioning?

  • HeatWave Autopilot auto provisioning estimates the HeatWave cluster size based on a workload’s data, including the attributes of its tables and required compression needed to load it into memory. Customers no longer need to manually estimate optimal cluster sizes.
  • Aurora lacks the equivalent capability for auto provisioning. Developers and DBAs need to guess or manually test the optimal computing instance type for their workload, increasing cost and development time.
yes
no

Does the database service include built-in automation for query plan improvement?

  • HeatWave Autopilot auto query plan improvement learns from executing queries and improves future execution. As customers run more queries, performance improves.
  • Aurora doesn’t learn from and improve query plans based on those that were previously executed, requiring DBAs to analyze and manually update table statistics for the Aurora query optimizer.
yes
no

Does the database service include built-in automation for auto scheduling?

  • HeatWave Autopilot auto scheduling determines which queries in the queue are short running and prioritizes them over long-running queries to reduce users’ waiting times.
  • Aurora lacks the equivalent workload management capability. Queries are executed based on which ones come in first, which can degrade performance for simple queries.
yes
no

HeatWave seamlessly integrates machine learning

With HeatWave AutoML, customers can build, train, and explain machine learning models within HeatWave.


Capability and evidence

HeatWave

Aurora

Does the database service provide in-database machine learning?

  • HeatWave AutoML lets developers and data analysts build, train, and explain machine learning models within HeatWave.
  • Aurora lacks the equivalent in-database machine learning capability. As a result, customers must move their data to a separate machine learning service, such as Amazon SageMaker, to build and train ML models. Security and compliance risks increase as data moves between systems.
yes
no

Are all machine learning models explainable, so users can understand and explain how they deliver their output?

  • HeatWave AutoML makes models’ workings and the way they make predictions explainable, helping with regulatory compliance, fairness, repeatability, and trust.
  • Not all models in Aurora ML are explainable, which can reduce trust, increase the risk of bias, and introduce regulatory complexity.
yes
no

Is the ML lifecycle automated?

  • HeatWave AutoML fully automates processes including algorithm selection, data sampling, feature selection, and hyperparameter tuning, saving time and effort. HeatWave AutoML also lets developers and data analysts build ML models using familiar SQL commands.
  • Aurora ML doesn’t automate many elements of the ML process and requires data science expertise to influence the performance, accuracy, and cost of training.
yes
no

Elevating beyond Aurora: The HeatWave edge

HeatWave lets businesses speed up queries of their key data and analyze data from transactions as they happen—without cumbersome ETL processes. It can run on Oracle’s cloud, AWS, and Azure. Businesses can take advantage of features such as LLMs that ship inside the database and automatic generation of the vectors needed to perform retrieval-augmented generation, which combines pretrained AI models with companies’ own proprietary data for more accurate, relevant results. Aurora can’t tune itself for OLTP workloads nor improve the execution of queries based on what it’s learned from past ones, contributing to lower absolute performance and lower price-performance than HeatWave.


HeatWave vs. Aurora FAQs

What makes HeatWave faster than Aurora?

HeatWave stores data in memory to accelerate queries and can scale to thousands of processing cores. The software synchronizes data between object storage and the MySQL Database or both, and an in-memory cluster, caching it in memory and removing reads from disk as a performance bottleneck.

How does HeatWave reduce costs compared to Aurora?

HeatWave eliminates the need for a separate analytics database and complex ETL operations between the two databases. It also eliminates the need for separate ML and generative AI services.

How do the machine learning capabilities of HeatWave compare with Aurora’s?

HeatWave AutoML includes ML models inside the database that business analysts can use to build systems for recommending content or products or running what-if scenarios for decisions without SQL commands or coding. Aurora users need to move data to a separate ML service, such as Amazon SageMaker, to build and train models.


Try HeatWave for free.

Try HeatWave for free

Learn more about HeatWave

Access free HeatWave migration resources

Request a free HeatWave workshop