MariaDB Enterprise Platform 2027 Is Generally Available; Its 56% Throughput Claim Rests on a MariaDB-Sponsored Test MariaDB Calls Non-Durable
MariaDB plc says Enterprise Platform 2027 is generally available. Its 56% peak-throughput claim over version 11.8 rests on a MariaDB-sponsored HammerDB test that MariaDB's own blog calls in-memory and non-durable.
Overview
MariaDB plc announced MariaDB Enterprise Platform 2027 on November 4, 2026, and said in its press release that the platform “is generally available today.” According to the same release, the platform includes MariaDB Enterprise Server 12.3, MariaDB Exa, MariaDB GridGain, MariaDB MaxScale, MariaDB Enterprise Manager and the MariaDB Enterprise Kubernetes Operator. The headline performance claim is 56% higher peak throughput over the last major release. That figure comes from a benchmark that, per MariaDB’s own footnote, was “Sponsored by MariaDB.”
The Machine Herald previously reported that MariaDB announced a definitive agreement to acquire GridGain Systems in March.
What MariaDB Says Is New
The press release describes the new AI features this way: “Agent Memory gives agents a durable, in-memory store shared across sessions, so long-running agents pick up where they left off. Semantic Cache returns stored answers to questions already asked in different words, cutting response times and token costs.” It adds that “AI plugins bring MariaDB expertise into coding agents such as Claude Code and Codex.”
Other items, all as described by MariaDB:
- Vector search. The press release says native vector search “now delivers up to 30% more queries per second than the last major release.” MariaDB’s launch blog post names the comparison as version 11.8 “at the same accuracy.”
- Multi-region. The blog post says MariaDB Raft Cluster is now generally available and that “Writes commit as soon as a majority of nodes confirm, so one slow or distant node no longer sets the pace.”
- Deployment and security. The blog post says MariaDB Exa is now generally available on premises, that Bring Your Own Cloud runs the MariaDB Cloud data plane in a customer’s AWS, Microsoft Azure or Google Cloud account, and that FIPS 140-3 is enabled through the MariaDB Enterprise Kubernetes Operator.
Independent trade outlet Blocks & Files covered the launch the same day. Its article lists the same components and repeats the 56 percent figure with the qualifier “in third-party testing.” It does not mention the sponsorship or the test conditions.
How the 56% Figure Was Produced
The press release footnote attributes the number to a McKnight Consulting Group report titled “High-Concurrency TPC-C Benchmarking,” dated October 2026. The footnote describes the workload as “HammerDB TPROC-C (derived from TPC-C, not an official TPC-C result),” calls it an “in-memory engine scalability test,” and compares MariaDB Enterprise Server 12.3.3-0 with 11.8.3-1 on Google Cloud. MariaDB’s report page describes the report as “An independent performance test, sponsored by MariaDB.” The full report sits behind a form, and this article is based on MariaDB’s own pages about it, not on the report itself.
A separate MariaDB engineering blog post by Adam Luciano and Steve Shaw gives more detail:
- Version 11.8.3-1 reached 1,702,758 NOPM (HammerDB’s New Orders Per Minute metric) at 88 virtual users, while 12.3.3 reached 2,653,370 NOPM at 152 virtual users. The post puts the ratio at “~1.56x.”
- The post states: “The configuration was deliberately in-memory and non-durable.”
- It also says 12.3.3 “ran with innodb_index_shrink = OFF, a non-default setting.”
- The hardware was a Google Cloud c4d-highmem-192-lssd instance with 96 cores and 192 vCPUs, running a 768-warehouse dataset.
MariaDB’s pages also give a comparison with MySQL, and they word it differently. The launch blog post says the platform “runs 25% faster than MySQL 9.7.2 at peak performance.” The report page says MariaDB 12.3 handled about 25% more orders per minute than MySQL 9.7.2 “between 64 and 120 users, and 21% more at 152 users,” and adds: “At each database’s best result, MariaDB’s lead was 21%.”
The engineering blog post also headlines “About 4x More OLTP Throughput than 10.6.” It states that “McKnight did not test 10.6.” That comparison lines up the McKnight 12.3.3 results against an earlier, separate run on “a similar 96-core system,” and the post says to “treat the comparisons in this section as indicative rather than like-for-like.”
Availability Wording Differs
The sources do not agree on whether MariaDB AI Retrieval is already in preview. The press release says: “MariaDB AI Retrieval follows in public preview on MariaDB Cloud later this month.” The launch blog post says: “MariaDB AI Retrieval is available in public preview on MariaDB Cloud.” Blocks & Files uses the same present-tense wording as the blog.
What We Don’t Know
- The full McKnight report is behind a form, so its configuration details and results beyond those MariaDB publishes on its own pages were not checked here.
- None of the sources cited here reports a repeat of the test by a party other than the MariaDB-sponsored run. The report page says the setup is published “so anyone can repeat the test and check the claims.”
- The sources do not give throughput results for configurations with durable commits enabled, so how the 56% gap carries over to production settings is not stated.
- The sources do not say whether 11.8.3-1 ran with a setting equivalent to the non-default innodb_index_shrink = OFF used for 12.3.3.