How OceanBase’s Shared‑Nothing Architecture Powers Scalable, High‑Availability Databases

OceanBase uses a shared‑nothing distributed cluster architecture in which every node hosts its own SQL engine, storage engine, and transaction engine. The system runs on ordinary PC servers, offering high scalability, high availability, high performance, low cost, and strong compatibility with mainstream databases.

Development began in 2010, and version 0.5 featured a two‑layer design: a stateless SQL service layer above a storage cluster composed of two types of servers.

The initial architecture, based on UpdateServer, provided good read scalability and a stateless, horizontally scalable SQL layer, but suffered from a single‑write‑node bottleneck and latency challenges caused by separating storage and SQL layers.

To address these issues, OceanBase introduced a new architecture (versions 1.0‑3.0) where all nodes are peers capable of handling SQL, transactions, and data storage simultaneously. This design delivers distributed vertical scalability and horizontal high availability, with scalability improving as more machines are added.

Before version 4.0, the architecture already demonstrated excellent scalability. Version 3.0 passed the TPC‑C benchmark, achieving 7.07 billion tpmC with linear scaling as nodes increased, and demonstrated 20 million transactions per second across a 1,557‑machine cluster in an eight‑hour stress test.

Version 4.0 introduced a dynamic log‑stream concept, decoupling transaction logs from storage shards. This allows multiple storage shards to share a single log stream and high‑availability service, reducing overhead for smaller‑scale deployments and better supporting medium‑sized enterprises.

OceanBase runs on generic server hardware without special requirements, using a shared‑nothing design. Each server runs a single‑process program called observer , which stores data and transaction redo logs locally.

Cluster deployment requires configuring zones (availability zones), logical groups of nodes that may reside in the same data center or span multiple data centers.

Data is replicated across zones for fault tolerance; each zone holds one replica, while multiple zones hold additional replicas synchronized via a consensus protocol.

The platform includes built‑in multi‑tenant capabilities, isolating CPU, memory, and I/O for each tenant, effectively giving each tenant an independent database.

Through this architecture, OceanBase achieves high availability, high compatibility, and horizontal scalability, supporting a wide range of business scenarios.