OceanBase Technical Overview: Distributed Transaction Processing and TPC‑C Benchmark

On November 19, 2019, Ant Financial announced the release of OceanBase 2.2 and the SOFAStack dual‑mode micro‑service platform, and shared a full transcript of a talk by senior researcher and OceanBase founder Yang Zhenkun titled “OceanBase: An Enterprise‑Grade Relational Database for the Future.”

The speaker explains that the goal is to transform the traditional centralized OLTP system into a distributed one, emphasizing that the real value lies not only in OLTP performance but also in supporting OLAP workloads.

He describes the historical evolution of relational databases from the 1980s ATM era, the emergence of separate OLTP and OLAP systems, and the challenges introduced by the explosive growth of internet traffic, which makes a single‑node solution neither scalable nor economical.

Key challenges of the split architecture include the need for ETL bridges between transaction and warehouse systems, global uniqueness of identifiers in sharded databases, and the non‑real‑time nature of data warehouses.

OceanBase addresses these issues by designing a horizontally scalable, highly available system that runs on ordinary hardware. It uses a Paxos group for each logical node, with three to five physical nodes forming a virtual node, and applies a majority‑vote protocol (2 out of 3 or 3 out of 5) to commit transactions, ensuring durability even if one or more nodes fail.

The system also adds a second replica to the primary, requiring at least two of three replicas to acknowledge a transaction, thereby achieving high availability without sacrificing performance.

The talk then details the TPC‑C benchmark, its origins, the five transaction types it models, and the specific workload characteristics (e.g., order creation, payment, distribution across warehouses). It explains how OceanBase performed the benchmark at 60 million tpmC over an eight‑hour stable run, meeting strict requirements for continuous operation, minimal performance variance, and full ACID compliance.

Results were publicly disclosed for a 60‑day review period, after which the benchmark becomes “Active” and the system can be purchased by any customer in the region.

Historical comparisons show OceanBase’s performance surpassing earlier results from DB2 and Oracle, highlighting the impact of using virtual machines and a large cluster (210 nodes total) to achieve the scale.

Beyond raw OLTP throughput, the speaker emphasizes that OceanBase aims to provide integrated OLAP capabilities, eliminating the need for separate data warehouses and enabling real‑time analytics on transactional data.

In conclusion, the presentation asserts that OceanBase is the first distributed database capable of full transaction processing while also supporting intelligent analytical scenarios, fulfilling a long‑standing demand for a unified OLTP/OLAP solution.