A Short Summary of the Last Decades of Data Management • Hannes Mühleisen • GOTO 2024

Relational databases and SQL have proven remarkably resilient and continue to dominate data management after 50+ years

Key-value stores, document databases, graph databases and other “NoSQL” alternatives are increasingly being absorbed back into relational systems

The fundamental split in database architectures is between transactional (OLTP) and analytical (OLAP) workloads, which have different optimization requirements

Distributed “big data” systems often introduce more complexity and costs than running on modern powerful single nodes - scaling up can be better than scaling out

Tables as a data organization concept predate written text by ~1000 years and remain a fundamental way to structure information

MongoDB, Cassandra and other NoSQL databases have gradually re-added SQL, schemas and ACID properties as developers struggled without them

DuckDB represents a new generation of analytical databases optimized for modern hardware and in-process execution

Vector databases and AI embeddings are likely to be absorbed into relational systems rather than remaining separate specialized databases

Application developers should generally not have to deal directly with storage, schemas and consistency - that complexity belongs in the database

SQL and relational databases continue to evolve and add capabilities while maintaining their core strengths of declarative queries and ACID guarantees