MariaDB 11.8 delivers up to 2.5x higher OLTP throughput on Dell PowerEdge R7715 servers powered by AMD EPYC™ processors
MariaDB Enterprise Server 11.8 Improves Transactional Performance
The database engine that powers global enterprise systems has enhanced its OLTP performance. MariaDB Enterprise Server 11.8 delivers up to 2.5x greater transactional throughput than the long-term release version of 10.6 when tested with the industry-standard HammerDB TPROC-C benchmark on Dell PowerEdge R7715 servers powered by AMD EPYC™ processors. Hardware was provisioned and hosted by StorageReview, the world-renowned authority on enterprise-class infrastructure.
In less than six months, the MariaDB engineering team advanced the server’s performance profile for transactional workloads. This performance gain wasn’t about removing old limits; it was about meeting the challenge of modern hardware head-on. With CPUs such as the 96-core AMD EPYC 9655P, database scalability needed to grow just as fast – and open source collaboration made that possible.
The key optimization built on the earlier work described in How Intel Helps MariaDB Become Even Faster. MariaDB engineers worked with HammerDB performance engineers to identify an opportunity to further refine InnoDB’s internal log handling to keep pace with today’s multi-core servers.
The initial fix for MDEV-33515 (log_sys.lsn_lock causes excessive context switching) introduced spinning on a mutex protecting the log sequence number (LSN), delivering a measurable uplift on high-core systems. However, with this feature, DBA expertise was required to set the innodb_log_spin_wait_delay parameter correctly for maximum performance.
In MariaDB 11.8, MDEV-21923 (LSN allocation is a bottleneck) went further by removing that mutex entirely, redesigning LSN generation to scale linearly with thread counts. This fully eliminated one of the final serialization points in the redo-logging subsystem. The result: around 2x the throughput measured with the earlier “spinning” implementation, and up to 2.5x higher performance compared with the 10.6 release – all with no DBA intervention required for maximum transactional performance.
This is open source at its best: rapid, collaborative innovation delivering measurable progress in 6 months from design to delivery.
Performance Evolution at a Glance
| MariaDB Version | Relative Performance |
| 10.6 (Baseline) | 1.0x |
| 11.8 Enterprise | 2.5x |
MariaDB Server 11.8 delivers a 2.5x increase in transactional throughput compared with the 10.6 release when tested with HammerDB TPROC-C on an AMD EPYC™ 9655P processor.
Figure 1. MariaDB Performance 11.8 vs. 10.6 Baseline
Engineered on Proven Enterprise Platforms
All testing was performed on production-grade infrastructure representative of modern data-center deployments:
| Component | Specification |
| Server | Dell PowerEdge R7715 |
| Processor | AMD EPYC 9655P (96 cores / 192 threads) |
| Memory | 768 GB DDR5 RAM |
| Storage | Dual Broadcom PERC13 NVMe RAID |
| Operating System | Ubuntu 24.04 LTS |
| Workload | HammerDB v5.0 TPROC-C |
Continuous Performance Engineering
Of course, no single change delivers performance in isolation. Improving redo log concurrency exposed other areas for optimization, most notably MDEV-19749, addressing MDL scalability regressions introduced by backup locks. This, along with a series of related efficiency improvements, collectively contributed to the throughput gains observed in MariaDB 11.8.
The table below lists a selection of the performance-related development tasks, known as MDEV tickets, from the MariaDB Jira issue tracker. Each represents a focused engineering change – ranging from concurrency improvements to memory and buffer management optimizations – that together enhanced the efficiency and scalability of the server. One of the great advantages of MariaDB open source development is the transparency it provides: anyone can review these tickets, follow the discussions, and see exactly how incremental changes between releases translate into measurable performance gains.
Ticket | Summary |
| MDEV-21923 | LSN allocation is a bottleneck |
| MDEV-19749 | MDL scalability regression after backup locks |
| MDEV-16168 | Performance regression on sysbench write benchmarks from 10.2 to 10.3 |
| MDEV-37244 | buf_read_page() should return a buffer-fixed block |
| MDEV-37152 | buf_page_get_gen → buf_pool->stat.n_page_gets++ high CPU utilisation |
| MDEV-37482 | AHI only contention on btr_sea::partition::latch |
| MDEV-36024 | Performance regression with encrypted InnoDB log |
| MDEV-24 | Segmented key cache for Aria |
| MDEV-34680 | Asynchronous and buffered logging for Audit Plugin |
| MDEV-36190 | Intel TSX-NI can cause a significant regression when TAA mitigation is needed |
Together, these improvements deliver enterprise-class scalability in MariaDB 11.8, cementing MariaDB’s role as a trusted platform for large-scale, business-critical systems.
HammerDB: Industry Standard Database Benchmarking
Many MySQL and MariaDB DBAs are familiar with sysbench, which excels at microbenchmarking – isolating and measuring specific low-level operations such as read, write or transaction latency. However, microbenchmarks alone don’t reveal how a database performs under the complex, sustained workloads seen in real-world environments.
HammerDB complements sysbench by providing macrobenchmarking at scale – pushing databases to their limits in terms of performance, scalability and stability. Its TPROC-C workload – derived from the industry-standard TPC-C and hosted and overseen by the TPC emulates genuine online transaction processing (OLTP) activity that scales to thousands of concurrent connections.
These results are complemented by benchmarks from Mark Callaghan of Small Datum, who has shown that MariaDB 11.8 maintains exceptional performance and stability under intensive sysbench workloads, reinforcing the broader picture of sustained scalability and robustness across both micro and macro-level testing.
What It Means for Enterprises
MariaDB 11.8 converts low-level engineering improvements directly into measurable performance gains. The server now scales efficiently across the full parallelism of modern AMD EPYC processors and high-bandwidth NVMe storage, sustaining millions of transactions per minute under HammerDB workloads.
These results require no tuning or configuration changes – throughput increases of up to 2.5x are achieved out of the box through optimized redo log concurrency, metadata locking and buffer pool management. The efficiency gains translate into higher utilization per core and reduced hardware footprint, lowering overall power and infrastructure cost.
Tested on production-grade systems rather than synthetic models, MariaDB 11.8 demonstrates that open source engineering can match and often exceed proprietary alternatives in both scalability and stability for enterprise-class workloads.
Conclusion
MariaDB Enterprise Server 11.8 achieved a new standard for open source OLTP performance. On Dell PowerEdge R7715 servers powered by AMD EPYC processors, MariaDB 11.8 delivers up to 2.5x greater throughput than 10.6 – demonstrating how focused engineering and open benchmarking translate directly into measurable performance gains. MariaDB continues to work closely with external benchmarking tools such as HammerDB to uncover new optimization opportunities and further improve scalability and throughput. The collaboration exemplifies how open engineering and transparent testing drive sustained innovation and progress in enterprise-class database performance.
