MySQL, MariaDB or Percona Server: the 2026 comparison and the benchmark dispute
Editorial note: The information in this article was compiled to the best of our knowledge at the time of publication. Technical details, prices, versions, licensing terms, and external content may change. Please verify the information provided independently, particularly before making business-critical or security-related decisions. This article does not replace individual professional, legal, or tax advice.
Do you run a database holding business-critical data? WZ-IT provides managed MySQL hosting, managed MariaDB hosting and managed Percona hosting, including backups, tested restores and monitoring. Discuss your setup
Anyone choosing a database from the MySQL family in 2026 will find contradictory performance comparisons. That is not sloppy research; it is because the vendors are currently disputing the measurements in public.
In January 2026 Percona published a comparison across the MySQL family. In April, MariaDB.org formally objected. Both pieces are documented and public. They reach different conclusions, and the difference lies in the configuration.
This article assesses the three systems against verifiable criteria, reports the dispute without settling it, and names the questions a selection actually has to answer.
Last reviewed: 24 August 2026.
Table of contents
- Why this comparison is difficult right now
- The three systems at a glance
- The 2026 benchmark dispute
- What can be drawn from it
- Origin and licence
- Compatibility: the costliest misconception
- Which system for which situation
- What to check before deciding
- How we approach this at WZ-IT
- Further guides
Why this comparison is difficult right now
Database performance comparisons are rarely neutral, because they almost always come from one of the parties involved. In 2026 that is unusually well documented: two detailed, publicly available accounts exist that contradict each other and are both reasoned.
For readers that is more valuable than another comparison with fresh numbers. Knowing both positions lets you judge how much weight published figures on this topic can carry at all.
The three systems at a glance
Figures from the public repositories, as of 24 August 2026.
| System | Repository | GitHub stars | Licence | Origin |
|---|---|---|---|---|
| MySQL | mysql/mysql-server | 12,398 | GPL-2.0 with FOSS exception | Oracle |
| MariaDB | MariaDB/server | 8,130 | GPL-2.0 | MariaDB Foundation, fork from 2009 |
| Percona Server | percona/percona-server | 1,270 | follows MySQL | Percona, variant of MySQL |
The star counts say little about adoption. MySQL and MariaDB are mostly installed from distribution packages rather than GitHub, and Percona Server is a tool for operators, not developers. All three repositories were updated in the days before 24 August 2026.
A caution on version numbers: for MySQL and Percona Server, GitHub releases are not a reliable source. Percona shows a 2017 state there while shipping releases continuously through its own channels. Check versions on the vendor sites.
The 2026 benchmark dispute
Percona, 23 January 2026. The performance review compares MySQL Community 8.4 and 9.5, Percona Server 8.4 and 9.5, and MariaDB 11.8.5. Testing used TPC-C-like workloads at a 50/50 read-write mix plus sysbench.
The result: Percona Server and MySQL sit close together. MariaDB performs well up to 32 concurrent connections but cannot serve traffic at higher concurrency in the TPC-C test, while doing well on isolated reads. The conclusion is that MySQL and Percona 9.5 are the best-performing version of the family currently available.
MariaDB.org, 3 April 2026. The response does not dispute the measurements but the test setup. Three points:
- Version choice. The MariaDB version tested was publicly known to have performance issues while a newer one was available.
- Unequal configuration. MySQL, Percona Server and MariaDB 11 ran with 4 GB InnoDB log files, MariaDB 12 with half that.
- Incorrectly set base parameters. Deprecated variables were not adjusted.
The disclosure is explicitly acknowledged: putting several releases on the same graphs and publishing the test harness is welcomed. The conclusion, however, should be read much more narrowly than the article suggests.
A note on reading these: Percona's January piece names MariaDB 11.8.5, while MariaDB's response argues about 12.1.2 versus 12.3.1. The response therefore addresses a later state of the measurement series. Anyone following the discussion should read both sources in the original.
What can be drawn from it
Not which system is faster. But three things that are more useful for a decision:
Configuration beats product choice. The dispute is entirely about settings, not architecture. A badly sized InnoDB log file costs more performance than the difference between the systems. Anyone with a performance problem should measure their own configuration rather than swap products.
Behaviour under load differs from behaviour in isolation. In both accounts the systems sit close together at low concurrency and separate only with many simultaneous connections. A comparison measuring only a few connections says little about production.
Published figures on this topic warrant caution, including those quoted here. A statement becomes reliable only with your own data and your own query patterns.
Origin and licence
MySQL has belonged to Oracle since the Sun acquisition in 2010. The Community Edition is under GPL-2.0 with a FOSS exception, alongside commercial licences with different terms. For self-operation the Community Edition is freely usable.
MariaDB emerged in 2009 as a fork led by original MySQL developers, with the stated aim of keeping development independent of Oracle. The server is under GPL-2.0. The current version in the repository is mariadb-12.3.2 from 28 May 2026.
Percona Server is not a fork but a variant: it tracks MySQL releases and adds tooling for operations and diagnostics. Using Percona keeps you in the MySQL branch.
Worth noting against other fields: none of the three has taken the path to a restricted licence seen at MongoDB, Elasticsearch or Redis. In that respect the MySQL family has stayed quiet.
Compatibility: the costliest misconception
The most common false assumption is that MariaDB is a drop-in replacement for MySQL. That was largely true in the early years after the fork and is not true today.
Diverged areas include:
- JSON handling. MySQL and MariaDB took different routes.
- System tables. Anything reading them directly needs checking.
- Replication. Setup and identifiers differ.
- Timestamps and character sets. Edge-case behaviour deviates.
Practical consequence: a switch is a project with testing, not a swap during a maintenance window. For simple applications it often passes unnoticed; for grown ones it rarely does.
And with standard applications the vendor decides anyway: if the support matrix names MySQL 8.0, running MariaDB shifts the risk onto the operator, even where it works at first.
Which system for which situation
| Starting point | Obvious choice | Reason |
|---|---|---|
| Standard application with a vendor support matrix | the supported system | deviating shifts the risk onto you |
| Existing MySQL installation, operations should improve | Percona Server | stays in the MySQL branch, adds operational tooling |
| New in-house development, independence from Oracle wanted | MariaDB | own development, GPL-2.0, clear provenance |
| Existing MariaDB installation without complaints | keep MariaDB | switching without cause only creates risk |
| Application does not exist yet | reopen the question | then PostgreSQL belongs in the selection |
What to check before deciding
What does the application require? For standard software, the support matrix; for in-house development, the functions actually used.
What does the load look like? Not the data volume but the number of concurrent connections and the read-write ratio. That is exactly where the systems separate, as both sides of the dispute show.
Who operates it? Backups with tested restores, monitoring of replication and locking, an upgrade path across major versions. That work applies to all three and is the larger item than the product choice.
How do you get out again? A logical dump in a vendor-neutral format should run regularly and be verified. It is also the best insurance against a wrong decision.
How we approach this at WZ-IT
We operate all three for clients and do not select on performance claims but on the application above them. In most cases the choice is already made before we arrive, because a standard application dictates it or an existing estate is being migrated.
Where we do decide, the question is almost never "which is faster" but "who keeps it running and how do we get back in an emergency". A backup whose restore has never been rehearsed is equally worthless on all three.
We do not solve performance problems by switching systems. First we measure where the time actually goes. In most cases it is queries, indexes or memory settings, not the product name.
Further guides
- PostgreSQL expertise - when the application is not yet fixed
- MySQL, MariaDB and Percona - the systems individually
- Managed operations - operations, backup and restore under contract
- Managed open source - application and database from one team
Unclear which system fits your application? We review the application, the load and the operating model and say what actually drives the decision. Book a call
Sources
Choose a database or have it operated
We assess which system fits your application and can take on operations, backups and tested restores.
Frequently Asked Questions
Answers to important questions about this topic
No longer consistently. MariaDB began in 2009 as a fork of MySQL and was largely interchangeable for years. Since MySQL 8.0 and MariaDB 10.x, data types, JSON functions, system tables and replication have diverged. A switch is possible, but it is not a binary swap.
A variant of MySQL that keeps its behaviour and interfaces and adds tooling for operations, diagnostics and backup. Unlike MariaDB it is not a fork on its own path but tracks the MySQL releases.
On 23 January 2026 Percona published a performance review of the MySQL family in which MariaDB fell off sharply under high concurrency. MariaDB.org responded publicly on 3 April 2026, criticising the configuration, including differently sized InnoDB log files and incorrectly set base parameters.
That cannot be settled from the outside, and this article does not attempt to. What is documented is that Percona published its test harness and MariaDB explicitly welcomes that openness while considering the configuration unrealistic. The dispute is therefore not a sound basis for a purchasing decision on its own.
MariaDB Server is under GPL-2.0. MySQL Community Edition is under GPL-2.0 with a FOSS exception, alongside commercial licences from Oracle. Percona Server follows MySQL licensing and is freely available. None of the three has moved to a restricted source-available model.
Not answerable in general. Published measurements contradict each other, and each side attributes the differences to the other's configuration. Anyone needing a reliable answer for their own application measures with their own data and query patterns.
Often yes for simple applications, no for complex ones. Check JSON functions, timestamp and character set behaviour, stored procedures, replication setup and anything touching system tables. A test against a copy of production does not replace that review but is the fastest route to the open questions.
The one the vendor supports. Many standard applications explicitly name MySQL or MariaDB in specific versions. Deviating shifts the risk onto the operator, even where it works technically.
If the application does not exist yet, that question comes first. This comparison assumes the MySQL family is already set, because a standard application requires it or an existing estate is being migrated.
Less than expected. Backups with tested restores, monitoring of replication and locking, an upgrade path and a plan for major version changes are needed with all three. The differences lie in the bundled tooling and in behaviour under load.
Written by
Timo Wevelsiep
Co-Founder & CEO
Co-Founder of WZ-IT. Specialized in cloud infrastructure, open-source platforms and managed services for SMEs and enterprise clients worldwide.
LinkedInLet's Talk About Your Idea
Whether a specific IT challenge or just an idea - we look forward to the exchange. In a brief conversation, we'll evaluate together if and how your project fits with WZ-IT.
Arrange a callback
Companies worldwide trust WZ-IT
