Road to Percona Live 2022 - Yahav Biran - PostgreSQL Talk
summary
Shows reproducible LWLock:buffer_content contention on large deployments and isolates the pathology to time‑series/transactional access patterns; runs a benchmark and real-world workload against Amazon Aurora and vanilla PostgreSQL on a 96 vCPUs, 768 GiB, 30 TB dataset to demonstrate where contention spikes. Proposes concrete schema and application changes and operational tips that reduce simultaneous buffer lock waiters and avoid hotspot pages, and reports which patterns actually lower contention versus those that merely sound plausible.
AI-written summary · about
related talks
48mUnderstanding and Reducing Application Contention
56mPerformance Improvements in PostgreSQL9 2
46mOptimizing High-Contention Writes in PostgreSQL
31mDemanding the impossible rigorous database benchm
56mLightning Talks
29mPerformance Archaeology - 20 years of improvements
11h 21mPercona Community Live Conference - Day 3 - Open Source Databases Conference
14mBuilding a better PostgreSQL together