started · updated
PostgreSQL performance and management in large-scale environments
Technical reports highlight different approaches to managing large-scale PostgreSQL databases. Lakebase Postgres utilizes a tiered architecture to mitigate the latency issues typically associated with object storage. By using a local compute cache for 99.46% of page accesses and a pageserver with local NVMe for 0.54%, the system ensures that object storage is only accessed in approximately 1 out of every 28 million requests.
In a separate implementation, Zerodha manages massive PostgreSQL instances on bare EC2 instances to handle historical transactional data. Their database architecture stores nearly 20 TB of data across four sharded nodes, managing hundreds of billions of rows. To handle peak traffic, Zerodha utilizes a specialized setup where one PostgreSQL instance acts as a hot caching layer for other large databases, supporting up to 50,000 queries per second during peak periods.
Entities
AWS · EC2 · Lakebase · PostgreSQL · Zerodha