OceanBase: Achieving Extreme Database Performance on Commodity Hardware

OceanBase is a fully self‑developed, financial‑grade distributed relational database created by Ant Group. It scales horizontally by adding machines and uses a multi‑replica replication model that runs on ordinary PC servers without requiring high‑end hardware.

The storage engine resembles an LSM‑Tree: all writes first go to an in‑memory Memtable and are later flushed to disk in the background, avoiding random writes and making the system friendly to both SSDs and HDDs.

Performance goals focus on extracting the maximum possible throughput from commodity hardware. Latency and throughput are the two key metrics; the aim is to keep latency within acceptable bounds while pushing throughput as high as possible.

Key CPU optimizations include reducing the number of instructions executed per SQL statement and improving the cycles‑per‑instruction (CPI) ratio. Commit operations are made asynchronous so that transaction‑commit threads do not block while waiting for log persistence.

System scalability is enhanced by minimizing unnecessary work and reducing contention on shared resources such as memory allocators and counters, allowing the database to efficiently utilize servers with dozens or even hundreds of CPU cores.

During the 2018 Double‑11 event, OceanBase 2.0 powered the Alipay core transaction path, handling a record‑breaking 213.5 billion CNY sales volume. Compared with version 1.4, version 2.0 achieved a 63 % performance increase in the order‑placement scenario and a 58 % increase in the payment scenario.

Future work will continue to optimize the full stack, adapt to new workloads, and target emerging hardware to maintain the highest possible performance‑to‑cost ratio.