OceanBase: Crossing the Death Valley of Relational Databases – A Technical Journey

On October 15, 2018, Ant Financial senior researcher Yang Zhenkun delivered a talk titled “OceanBase: Crossing the Death Valley of Relational Databases” at Beijing Jiaotong University, describing how OceanBase grew from an innovative prototype to a product that overcomes the three stringent demands of modern databases: high‑concurrency transaction processing, zero data errors, and uninterrupted service.

The piece explains why databases face a high market barrier, how most enterprises struggle to build a generic system, and why Ant Financial chose to develop its own solution instead of relying on open‑source or commercial offerings.

Key technical innovations are highlighted: a shift to a distributed architecture, the use of a majority‑vote consensus protocol, in‑memory incremental updates merged to disk nightly, and a cost‑effective PC‑server‑based design that achieves five‑nine reliability without expensive hardware.

Case studies such as the Taobao Favorites service illustrate how OceanBase reduced I/O from hundreds of random reads per request to a single read by materializing a wide table (materialized view) and merging updates nightly, dramatically cutting server count and operational cost.

Major milestones are traced: early 2010‑2012 demand‑driven development, the 0.3‑0.4 versions that nearly led to team dissolution, the 0.5 version that survived a 10% traffic cut during Double‑11 2013, and the 1.0 version that became the only commercial database supporting distributed transactions.

The article also describes OceanBase’s “three‑site five‑center” deployment, enabling city‑level fault tolerance and automatic, loss‑free recovery within seconds, as demonstrated at the Cloud Expo ATEC forum.

Finally, the narrative notes OceanBase’s expansion beyond Alibaba, serving external banks and enterprises, and promotes the upcoming OceanBase TechTalk event in Beijing, where product features of OceanBase 2.0 will be showcased.