OceanBase’s TPC‑C Benchmark: Technical Deep Dive into Test Preparation, System Architecture, and Performance Results

Background Ant Financial’s self‑developed database OceanBase recently topped the TPC‑C benchmark, attracting wide industry attention. To explain the technical details, OceanBase core engineers released a five‑part series; this is the second article focusing on the test execution.

Preparation Achieving a TPC‑C record required over a year of preparation, including a six‑month audit certification. The team optimized both performance and stability, and the 60.88 million tpmC figure represents only a small part of the overall test metrics.

Test System The TPC‑C test involves a Remote Terminal Emulator (RTE) and a System Under Test (SUT) that can be split into a Web Application Server (WAS) and a DB server. OceanBase used OpenResty for the WAS and built a fully compliant, scalable RTE implementation. The test ran on 64 cloud servers (64 CPU × 128 GB) with 960 RTE clients simulating 47,942,400 user terminals.

Test Planning Planning covered hardware selection, parameter tuning (e.g., number of warehouses per server), and resource allocation. Because OceanBase leveraged cloud resources, hardware provisioning could be adjusted quickly to meet performance and cost goals.

Performance Testing The benchmark follows three phases: ramp‑up, steady‑state, and measurement interval. OceanBase performed a 10‑minute warm‑up, a 25‑minute steady‑state run with a checkpoint, and an 8‑hour measurement phase, keeping tpmC variation under 0.5 % and completing the required checkpoints.

ACID Tests OceanBase passed the A (atomicity), C (consistency), I (isolation), and D (durability) tests. Notably, it executed both local and distributed transaction modes for the I test, and demonstrated rapid recovery from node failures during the D test, restoring 60 million tpmC within two minutes without relying on RAID.

Author The article is authored by Cao Hui, a technical expert in OceanBase’s storage engine team.