Unlocking enterprise performance at scale: Autonomous PostgreSQL performance tuning with DBtune surpasses manual tuning for Intel
87% higher throughput, then another 17% on an already heavily optimized system, with 43% less disk I/O and 23% less WAL generation.
Exec summary
Intel evaluated DBtune's AI-driven platform on a large-scale PostgreSQL environment on Intel® Xeon® processors to determine whether autonomous optimization could uncover meaningful performance gains beyond extensive manual parameter tuning. DBtune increased transaction throughput by 87% during the initial optimization phase and then delivered a further 17% improvement on an already heavily optimized system, while reducing disk I/O activity by 43% and WAL generation by 23%.
The key result was not a single parameter change. DBtune autonomously identified the dominant bottlenecks, validated their root causes through experimentation and translated those findings into measurable performance and efficiency gains—without changes to the application, schema or hardware.
The challenge: Maximizing database performance and stability on modern hardware
Modern enterprise servers provide unprecedented CPU and memory capacity, but extracting maximum performance from PostgreSQL remains a complex task. Intel worked with DBtune to evaluate how autonomous optimization could identify scalability bottlenecks and unlock additional platform performance.
Test environment
Testing was conducted using a dedicated two-node configuration connected through an SSH tunnel across separate subnets.
- Database server: Intel Xeon 6972P, 384 vCPUs, 1.5 TB RAM, 1.7 TB NVMe SSD
- Workload client: Intel Xeon Platinum 8480CTDX, 224 vCPUs, 251 GB RAM, 1.7 TB NVMe SSD
- PostgreSQL 18.4, HammerDB 6.0, TPC-C workload, 1,536 warehouses, 384 virtual users
Frictionless deployment and autonomous optimization
The DBtune agent was installed quickly and optimization loops executed autonomously, continuously evaluating PostgreSQL configuration changes without manual intervention.
What autonomous optimization revealed
The analysis consistently pointed to checkpoint behavior, WAL pressure, I/O utilization, and LWLock contention as the dominant bottlenecks. The objective was to identify root causes and validate the most impactful change through experimentation.
Phase 1: Nearly doubling throughput through autonomous optimization
Autonomous optimization identified checkpoint-related contention and delivered an 87% increase in transaction throughput without application, schema, or hardware changes.
Phase 2: Finding additional headroom in an already optimized environment
Even on a heavily manually optimized system, DBtune’s AI engine delivered an additional 17% throughput improvement while reducing overhead.
More than speed: Improving efficiency
The optimized configuration reduced disk I/O activity by 43%, reduced WAL generation by 23%, and maintained zero deadlocks and zero temporary file spills during testing.
Why this matters for Intel platforms
As CPU core counts, memory capacities, and storage performance continue to increase, software configuration becomes an increasingly important factor in realizing platform potential.
For this evaluation, the objective was not simply to improve a default PostgreSQL deployment. The more challenging goal was to determine whether meaningful performance gains remained after extensive manual tuning. A key target was identifying approximately 20% additional throughput on an already configured and optimized PostgreSQL environment, a level of gain that is typically difficult to achieve through conventional optimization techniques.
The evaluation demonstrated that autonomous tuning can help PostgreSQL scale more effectively on large Intel Xeon platforms by identifying bottlenecks that may otherwise remain hidden. Rather than relying solely on manual performance investigations, engineers can use intelligent optimization systems to continuously evaluate configurations and discover workload-specific opportunities for improvement.
Notably, the autonomous optimization process delivered a 17% throughput improvement on a system that had already undergone significant manual optimization, coming close to the original performance objective while simultaneously reducing disk I/O activity and WAL generation. This suggests that autonomous optimization can uncover performance headroom even when traditional tuning efforts have already addressed the most obvious opportunities.
“Autonomous PostgreSQL optimization can identify hidden bottlenecks, validate root causes, and uncover additional headroom even in environments that have already been extensively optimized” — Srinath Duvuru, Cloud Software Engineer at Intel.
The most important outcome was not that a single parameter changed. It was that the system automatically determined why performance was limited, validated the root cause through experimentation, and delivered measurable gains while reducing system overhead.
Conclusion
The evaluation demonstrated that autonomous optimization can improve performance and efficiency by automatically identifying and addressing dominant scalability bottlenecks. The most valuable outcome was not a specific parameter change, but the ability to diagnose and validate root causes automatically.