Managed vs self-hosted databases: compare the software, service and full cost

Separate the database engine from the running service

A database engine is the software that stores and queries your application’s data. A managed service adds a particular operating arrangement around an engine. Start a shortlist with two columns: the exact software and edition, and the service that will run it. Comparing a free software download with an entire hosted deployment hides the work and infrastructure in the second offer. Keep application compatibility, licensing, hosting and support as distinct buying questions.

Write a workload brief before asking for prices

Describe the records, relationships and transactions that the application needs. Include a representative import, the queries that users wait for, concurrent activity and an expected growth point. Use an approved sample rather than copying a production customer database into a trial. Ask the application owner which extensions, drivers and versions it requires. A broad claim about scale cannot establish whether an engine supports your particular transaction or query correctly.

Name an owner for each operating task

For each proposal, assign backups, restoration, patch planning, monitoring, database permissions and application changes to a named provider or internal role. A managed offer can change the division of work without removing every customer responsibility. Amazon RDS, for example, uses a shared responsibility model and gives customers controls over resource and database access. Ask for the actual service scope in writing. Include the time and skills your team still needs in the cost comparison.

Turn recovery requirements into an acceptance check

State how much recent work the business could lose and how long it could wait for the application to work again. An example brief might allow one hour of lost changes and two hours to restore service; those numbers are a hypothetical requirement, not a result for any database here. In an isolated test environment, restore an approved sample and reconnect a copy of the application. Record the recovered data, elapsed time and any manual steps. A successful backup job alone does not show that the application has been recovered.

Check which kind of backup and restore is included

Different backup methods support different recovery paths. PostgreSQL documents SQL dumps, file-system backups and continuous archiving. Amazon RDS offers automated backups and manual snapshots; its documented point-in-time restore creates a new instance. Do not assume every engine, edition or hosting package implements the same arrangement. Ask about retention, recoverable points, separate storage charges and the steps needed to redirect an application after restoration. Keep the chosen version and deployment with the answer.

Build a cost sheet for a normal month and a growth month

Record software or edition fees, compute, primary storage, backup retention, additional replicas, network transfer, monitoring, support and staff time. Mark each line as included, separately quoted or unknown. Then model a larger data set and a busier period using the same assumptions in each proposal. Keep temporary credits and trial allowances outside the standard recurring estimate. Use the software cost calculator for the fixed parts of your own quote; consumption charges need their own workload estimate.

Check portability before choosing the deployment

Export a small representative sample and prove it can be read outside the original environment. Include schema, indexes, roles, attachments or external objects where relevant to your application. Ask which extensions or service-specific features need replacement in an exit. Record the change window, transfer charges and who owns the migration. A familiar database name does not by itself guarantee that every managed configuration can move unchanged to another provider.

Keep the decision and its evidence together

Retain the workload brief, exact edition, service region, dated quote, responsibility list and recovery observations. Record unresolved requirements as unknowns. Start with PostgreSQL vs MySQL, MySQL vs MariaDB or SQL Server vs Oracle Database to frame the engine decision. Use your own comparison scorecard for the retained offers. These proposed checks do not establish measured performance or recovery guarantees for the products.

Frequently asked questions