At POSET TE 2026, Sai Srirampur delivered "Postgres Isn’t Slow. Your Storage Is," highlighting how storage performance underpins Postgre SQL scalability. The talk outlines five common symptoms—slow ingestion, erratic read latency, autovacuum lag, prolonged checkpoints, and logical replication delays—often misattributed to the database engine itself. Sai benchmarked a 3.3‑billion‑row workload on baseline gp3 EBS and local NVMe, using eight identical Postgre SQL clusters (four per storage type) on m6id.4xlarge instances. Results showed network‑attached storage operations in milliseconds versus tens of microseconds on local NVMe, demonstrating that local NVMe sharply reduces I/O cost and mitigates latency spikes. The presentation concludes with production patterns for running Postgre SQL on local NVMe to achieve consistent performance at scale.
Key takeaways: identify storage as the hidden bottleneck, isolate I/O variables in testing, and adopt local NVMe for high‑throughput Postgre SQL workloads.
Key_entities
Source: Hacker News · Summarized by HeadlinesBriefing