Benchmarking PostgreSQL Connection Pooling with PgBouncer (7432)

Technical Overview

Engineering breakdown of Benchmarking PostgreSQL Connection Pooling with PgBouncer (7432). Bare-metal hardware performance requires isolated kernel parameters such as net.core.somaxconn, net.ipv4.tcp_tw_reuse, and tuned vm.swappiness to eliminate OS‑level bottlenecks.

Test Methodology

We used pgbench with 500 concurrent clients, each running a mixed read/write workload for 30 minutes. PgBouncer was configured in transaction pooling mode with a pool size of 200, max_client_conn=1000, and default_pool_size=200. The PostgreSQL instance ran on a dedicated 32‑core, 256 GiB RAM node with NVMe storage, PostgreSQL 15, and max_connections=500.

Key Findings

• Transaction pooling reduced average query latency by 38 % compared to session pooling.
• Connection acquisition time dropped from 2.4 ms to 0.6 ms under peak load.
• CPU utilization on the PgBouncer host stayed below 12 % even at 10 k TPS.

Full raw data and statistical analysis are available in the attached telemetry chart.