How Michael Stonebraker Is Turning Databases Into Operating Systems

From Markov Chains to Databases

Michael Stonebraker, the 2014 Turing Award laureate and founder of Postgres, began his research on relational databases at Berkeley after realizing his PhD work on Markov chains lacked practical value. Together with Eugene Wong, he was inspired by Edgar Codd’s relational model proposal and set out to build Ingres, despite having no prior experience in system software.

Ingres competed with IBM’s System R and later Oracle’s early relational systems, but faced platform challenges—Unix at the time could not run Ingres for large deployments such as Arizona State University’s student records. To commercialize Ingres, Stonebraker founded Relational Technology, which later became Ingres Corporation and was eventually absorbed by larger software firms.

Former Ingres team members went on to create Sybase, whose technology was later licensed to Microsoft for early SQL Server versions.

A New Era

In 1986 Stonebraker and Larry Rowe published a 28‑page paper describing Postgres, emphasizing extensibility for complex objects and user‑definable data types, operators, and access methods. Feedback from Ingres customers highlighted the need for flexible time handling—Postgres later added support for JSON documents, bridging relational and NoSQL worlds.

Stonebraker has been critical of the NoSQL movement, arguing that modern NoSQL systems are converging toward SQL‑like interfaces to meet consistency requirements and improve developer experience.

Closed‑Source Databases Don’t Fit Future Trends

Stonebraker’s venture Illustra was sold to Informix, and he later served as CTO of Informix. He has repeatedly noted that open‑source databases outperform proprietary ones in flexibility and cost, and that the industry’s shift toward cloud‑compatible PostgreSQL services—such as Google Cloud SQL, Azure PostgreSQL, AWS Aurora, CockroachDB, and YugabyteDB—validates this view.

Disrupting Databases Again

Stonebraker’s latest concept, DBOS (DataBase Operating System), envisions an operating system built on a database rather than the reverse. Inspired by discussions with Matei Zaharia of Databricks, DBOS would store all OS state in database tables, enabling seamless upgrades, machine‑learning‑driven decisions, and robust security without downtime. The prototype draws on multi‑node, multi‑core, transactional, high‑availability technology from VoltDB.

Even at age 80, Stonebraker remains active, stating he cannot imagine a life of three days of golf a week and will continue working as long as his mind is engaged.