Postgres isn't slow, your storage is

Your PostgreSQL performance issues might be fixed by changing your storage configuration, not tuning the database itself.

A recent talk at POSETTE 2026 highlighted that PostgreSQL performance issues at scale, such as slow ingestion, inconsistent read latency, lagging autovacuum, lengthy checkpoints, and logical replication lag, can often be attributed to storage performance rather than the database itself. The presenter, Sai Srirampur, conducted a benchmark comparing PostgreSQL on baseline gp3 EBS and local NVMe storage. The benchmark utilized an m6id.4xlarge instance with 16 vCPUs and 64 GiB RAM, running a 482 GiB, 3.3-billion-row workload with 64 clients and 16 threads performing random UPDATEs. Results showed local NVMe achieved a median of 16,030 TPS compared to 1,734 TPS on EBS, a 9.24x improvement, with median transaction latency dropping from 36.9 ms to 4.0 ms. NVMe also outperformed EBS significantly in VACUUM and logical replication tasks. The analysis indicated that EBS latency was largely due to storage I/O, causing PostgreSQL processes to wait, while NVMe allowed for faster I/O and less CPU waiting time. The talk also addressed operational concerns regarding local NVMe's ephemeral nature, proposing solutions for production environments.

The benchmark reveals a dramatic performance gap between local NVMe and EBS for PostgreSQL, suggesting that many perceived database bottlenecks are actually storage-bound. By using local NVMe, users can achieve an order of magnitude improvement in transactions per second and significantly reduce latency, allowing databases to utilize CPU resources more effectively instead of waiting on I/O. This implies a shift in how organizations should approach database performance optimization, prioritizing storage infrastructure for workloads that push the limits of traditional cloud storage.

The proposed production architecture for NVMe-backed PostgreSQL addresses the inherent ephemerality of local storage by focusing on cluster-level replication and external backup solutions. This strategy enables businesses to leverage the speed of NVMe while maintaining enterprise-grade availability and disaster recovery capabilities, potentially making faster, more responsive databases accessible without sacrificing data durability. Organizations should evaluate their current storage solutions and consider the operational changes required to adopt local NVMe for critical PostgreSQL deployments.