Database selection is a foundational decision when building an application. Choosing the right database enables you to focus effort on building customer value while the database scales with demand. Choosing a suboptimal database can lead to performance, scale, and functionality issues and a database migration in the future. AWS offers several fully managed database services. That breadth serves the diversity of the millions of customers who operate databases on AWS, but it raises a practical question: Where do I start? Making a database decision in the face of deadlines, organizational inertia, and decision paralysis can further complicate the choice. Or, increasingly, your agent is making your database selection on your behalf. Based on our collective experience working with database customers, this post provides a direct, opinionated answer focused on the job you are trying to perform: migrating an existing workload, refactoring a legacy application, or building something new. We have also encoded this guidance in AWS database selection skill to help guide your agents.
Migrating an existing workload
When migrating to AWS, choose the managed equivalent of the database you already run. This is the fastest path to production: your application code, schema, and team expertise transfer unchanged, and you gain automated patching, backups, and scaling out of the box.
| Source Database | AWS Managed Equivalent |
| PostgreSQL | Aurora PostgreSQL serverless |
| MySQL | Aurora MySQL serverless |
| MariaDB | Amazon RDS for MariaDB |
| Oracle | Amazon RDS for Oracle |
| Exadata | Oracle AI Database@AWS |
| SQL Server | Amazon RDS for SQL Server |
| Db2 | Amazon RDS for Db2 |
| MongoDB | Amazon DocumentDB (with MongoDB compatibility) serverless |
| Cassandra | Amazon Keyspaces (for Apache Cassandra) |
| Redis / Valkey | Amazon ElastiCache for Valkey serverless |
| Neo4j | Amazon Neptune serverless |
| InfluxDB / time-series | Amazon Timestream for InfluxDB |
For migrations, the principle is straightforward. If an AWS-managed equivalent exists, use it, and start with serverless when possible.
Refactoring a legacy application
Refactoring is different from migration. Migration moves the workload as-is. Refactoring modifies the application to improve performance, scale, or price, or to take advantage of cloud-native capabilities, without changing the behavior of the application. That often means changing the database architecture.
If you are refactoring from Oracle, Db2, SQL Server, or another legacy relational database, to eliminate commercial license costs, start with Aurora PostgreSQL serverless. PostgreSQL has the broadest feature compatibility with commercial relational databases and the largest open source community.
If you already know the access patterns and scaling limitations of existing applications, modernizing a monolithic database can result in adopting multiple purpose-built databases to better optimize for performance, cost, and functionality. Customers frequently use Aurora, DynamoDB, and ElastiCache together for a single application.
If you are refactoring an application to improve performance and scale, Aurora PostgreSQL is a strong starting point. Aurora PostgreSQL delivers the performance, scalability, and availability that enterprise workloads require. If you require horizontal scale-out in a single Region or across Regions, Aurora DSQL and DynamoDB are your options: Aurora DSQL for relational workloads that require horizontal scale, and DynamoDB for consistent single-digit-millisecond performance at any scale. DynamoDB delivers the same performance whether your application has tens of thousands or hundreds of millions of simultaneous users. Both Aurora DSQL and DynamoDB provide multi-Region strong consistency and up to 99.999% multi-Region availability Service Level Agreements (SLAs). If the workload has a specialized data model (graph, time-series, document), choose the corresponding purpose-built service from the table above.
Building a new application
If you are building a new application and you are already familiar with a database like PostgreSQL, MySQL, Valkey, or DynamoDB, and you are confident and productive with that database, use what you know. If you do not have a compelling need to choose a different database, choose what will help you deliver value to your customers faster.
If you are building a new application and do not yet know your access patterns, data model, or scale requirements, start with PostgreSQL. As a relational database, PostgreSQL provides flexibility to evolve your database schema as you iterate on your application requirements. PostgreSQL is the most widely used database among developers (57.9% of respondents) and the most admired database (62.3% of respondents) per the 2026 Stack Overflow Developer Survey, for the fourth consecutive year. If you require a PostgreSQL database, choose Aurora PostgreSQL serverless, which scales compute and storage automatically, reducing operational complexity and cost, especially when developing new applications. Aurora PostgreSQL serverless supports relational, JSON, and hybrid search (vector similarity combined with full-text search) in a single engine. If you require a relational database and are building for scale and operational simplicity, choose Aurora DSQL. Aurora DSQL scales to meet any workload demand with zero infrastructure management and zero downtime maintenance. For most applications, Aurora PostgreSQL serverless or Aurora DSQL will be strong options. As your application grows, consider a more specialized service only if the workload demands it. For instance, if your agentic application requires contextual knowledge through connected data, such as knowledge graphs for grounding agent reasoning and retrieval-augmented generation (RAG), choose Amazon Neptune.
If you already know you are building for massive scale, hundreds of thousands to millions of requests per second, single-digit-millisecond latency, and no tolerance for capacity planning or infrastructure management, start with Amazon DynamoDB. DynamoDB is a strong choice for high throughput workloads where you can define your primary access patterns upfront. The Amazon.com shopping cart is a well-known example.
Caching is complementary to any database architecture and pervasive across all data stores and workload types. To achieve microsecond latency, reduce read costs, and improve throughput when using AWS databases, start with Amazon ElastiCache for Valkey serverless. If you are building an agentic application, using ElastiCache for Valkey serverless can reduce token costs and improve performance.
Beyond these defaults (Aurora PostgreSQL serverless, Aurora DSQL, DynamoDB, and ElastiCache for Valkey), match the database to the workloads in the table below.
| Database type | Examples | Recommendation |
| Relational | Relational databases store data with predefined schemas and relationships between them. These databases are designed to support ACID transactions and maintain referential integrity and strong data consistency. | Aurora |
| Distributed NoSQL | Distributed NoSQL databases are optimized for common access patterns, typically to store and retrieve large volumes of data. These databases deliver quick response times, even in extreme volumes of concurrent requests. | DynamoDB |
| Distributed SQL | Distributed SQL databases are optimized for relational workloads that require strong consistency across multiple nodes, typically to process transactions and complex queries at scale. These databases deliver ACID guarantees and familiar SQL interfaces, even across geographically distributed deployments. | Aurora DSQL |
| In-memory, caching | In-memory databases are used for applications that require real-time access to data. By storing data directly in memory, these databases deliver microsecond latency to applications for session management, real-time leaderboards, feature stores, and semantic caching. | ElastiCache for Valkey |
| Document | A document database is designed to store semi-structured data with flexible schema as JSON-like documents. These databases help developers build and update applications quickly. | Amazon DocumentDB (with MongoDB compatibility) |
| Graph | Graph databases are for applications that need to navigate and query of relationships between highly connected graph datasets with millisecond latency at large scale. | Amazon Neptune |
| Time-series data (Internet of Things (IoT), DevOps metrics) | Time-series databases efficiently collect, synthesize, and derive insights from data that changes over time and with queries spanning time intervals. | Amazon Timestream for InfluxDB |
The bottom line
Databases are foundational. Do not overthink database selection. The simplest answer is often the right answer. Choose the database that satisfies your access patterns, scale requirements, and operational constraints. Do not add complexity you do not yet require. For most new applications, PostgreSQL is the right starting point. If you are looking for a PostgreSQL database, choose Aurora PostgreSQL serverless. If you require a relational database and are building for scale and operational simplicity, choose Aurora DSQL. For scale-first applications, choose DynamoDB. For caching of any data store, choose ElastiCache for Valkey. For migrations, match the managed equivalent. For refactoring from Oracle, SQL Server, Db2, or another legacy database, default to Aurora PostgreSQL serverless and specialize only when the workload demands it. Consider adding in-memory caching to reduce costs and improve database performance. If you are unsure, we built an AWS database selection skill that guides your agent to work backward from your application requirements to recommend the best database for your workload. If you want to talk through options with a human, we are happy to help. Our AWS Solutions Architects work with customers daily to build secure, reliable, performant, and scalable database architectures.


