Performance depends on workload type and scale. For OLTP (many small transactions): PostgreSQL and MySQL both handle 10K-50K writes/sec on single server. PostgreSQL has advantage with MVCC (non-blocking reads). For OLAP (complex queries): PostgreSQL better for window functions, CTEs. Specialized databases (Redshift, BigQuery) better for very large data (>1TB). For mixed workloads: PostgreSQL handles both OLTP and OLAP reasonably well. For very high scale (>1M users): Both PostgreSQL and MySQL need sharding or read replicas. Instagram: PostgreSQL with 2B users, 1000+ servers. Uber: PostgreSQL with 50K writes/sec. Facebook: MySQL with custom sharding. Performance optimization: (1) Indexes on WHERE/JOIN columns, (2) Connection pooling (PgBouncer, ProxySQL), (3) Read replicas for read-heavy workloads, (4) Caching (Redis, Memcached), (5) Query optimization (EXPLAIN, indexes). Cost considerations: AWS RDS PostgreSQL/MySQL: $100-$1000/month for small-medium apps. SQL Server: $14K-$47K/core licensing + hosting. Oracle: $47K/core/year + hosting. For startups: PostgreSQL or MySQL (free licensing, pay only for hosting).

Both PostgreSQL and MySQL handle 10K-50K writes/sec on single server (sufficient for most apps)

PostgreSQL MVCC provides better concurrency (non-blocking reads during writes)

MySQL is slightly faster for simple queries (SELECT by primary key), PostgreSQL faster for complex queries (JOINs, aggregations)

For very high scale (>1M users), both need sharding or read replicas (Instagram: 1000+ PostgreSQL servers)

Indexes are critical: Create indexes on WHERE/JOIN columns, use EXPLAIN to verify index usage

Connection pooling is essential: Use PgBouncer (PostgreSQL) or ProxySQL (MySQL) to handle 1000s of connections

Read replicas for read-heavy workloads: 1 primary (writes) + multiple replicas (reads) = 10x read capacity

Caching reduces database load: Use Redis or Memcached for frequently accessed data (90% cache hit rate = 10x fewer database queries)

Choosing database based on hype or trends (NoSQL hype, NewSQL hype) instead of analyzing requirements

Not testing with realistic data (testing with 1000 rows when production will have 100M rows)

Ignoring total cost (Oracle licensing $47K/core/year + hosting + support = $100K+/year)

Not considering team skills (choosing PostgreSQL when team only knows MySQL = 2-4 weeks learning curve)

Premature optimization (choosing specialized database for scale you don't have yet)

Not planning for migration (locked into database with no migration path)

Choosing multiple databases too early (microservices with 10 different databases = operational nightmare)