At POSETTE 2026, Sai Srirampur presented findings that storage performance, not PostgreSQL itself, often causes slowdowns in large-scale deployments. The benchmark compared a 3.3-billion-row PostgreSQL workload running on baseline gp3 EBS storage versus local NVMe, revealing significant performance differences. The talk highlighted common symptoms such as slow ingestion, unpredictable read latency, and replication lag that arise as data scales into hundreds of gigabytes or terabytes, according to clickhouse.com.
Sai’s benchmark focused on real-world concurrency and throughput scenarios, identifying five common issues affecting PostgreSQL at scale. These include ingestion pipelines suffering from heavy UPDATEs and UPSERTs that slow from seconds to minutes, inconsistent read latency impacting query performance, autovacuum processes falling behind, checkpoint resource contention, and logical replication lag. The analysis showed that local NVMe storage can mitigate many of these problems compared to gp3 EBS, enabling smoother operation under heavy workloads.
This insight is significant as PostgreSQL remains a popular open-source database for large-scale applications, but users often attribute performance bottlenecks to the database engine rather than underlying storage. By isolating storage as a key factor, organizations can optimize deployments with faster local NVMe drives to improve ingestion speed, reduce latency spikes, and maintain replication health. Sai proposed production patterns for running PostgreSQL on local NVMe to address these challenges effectively.
The full talk and benchmark details are available on clickhouse.com, providing technical guidance for teams managing PostgreSQL workloads at scale. Sai’s presentation offers a data-driven approach to diagnosing and resolving performance issues linked to storage infrastructure.