Distributed Database Performance Optimization
Led infrastructure work to reduce tail latencies and improve query planning for a distributed database system.
Context & Problem
A distributed DB startup had customers reporting unpredictable query latencies, especially in multi-region deployments. This was becoming a blocker for enterprise sales.
Role & Responsibilities
Staff engineer, responsible for query planner and replication layer improvements.
Architecture & Key Decisions
Rewrote query planning logic to optimize for geo-locality, implemented replica preference hints, added cost-based routing for follower reads.
Impact & Outcomes
Reduced p99 latency by 65%, cut customer-reported slowdowns by 80%, enabled enterprise sales to close 3 major deals.
What I'd Do Differently
Understanding the actual bottleneck beats generic optimization. Spend time with customers and trace real queries. Replica selection and follower reads unlock a lot of performance without major rewrites.