Two sizes fit most: PostgreSQL and ClickHouse
10 days ago
- #databases
- #PostgreSQL
- #OLAP
- Relational databases, particularly SQL databases, have remained dominant since their introduction in 1974, despite various challengers.
- PostgreSQL has emerged as an improvement over its predecessors and competitors, maintaining the dominance of relational databases.
- Specialized databases like Clickhouse have gained traction in OLAP (Online Analytical Processing) due to their performance benefits for analytics.
- General-purpose relational databases are often misused for inappropriate data storage, leading to inefficiencies.
- Michael Stonebraker and Uğur Çetintemel argued against the 'one size fits all' approach in database architecture, highlighting inefficiencies in non-OLTP workloads.
- Column stores have proven 10-100x more efficient than traditional row stores for data warehousing.
- OLAP databases like Clickhouse and Vertica have become significant, offering real-time analytics with SQL interfaces.
- Stonebraker later argued that even OLTP databases could be improved by eliminating overheads like buffer management and locking, proposing single-threaded, in-memory architectures.
- H-Store and VoltDB were developed as prototypes and commercial products based on these principles, but tradeoffs like high memory costs and single-thread limitations hindered widespread adoption.
- PostgreSQL has become the dominant relational database, especially in open-source environments, due to its versatility and feature set.
- NoSQL databases addressed some shortcomings but are often unnecessary as modern relational databases like PostgreSQL handle JSON, key-value, and specialized use-cases effectively.