How OceanBase Set a New TPC‑H Record and the Future of HTAP Databases

On May 20, 2021, the Transaction Processing Performance Council (TPC) announced that Ant Group's self‑developed distributed relational database OceanBase set a new world record in the TPC‑H analytical benchmark with a performance score of 15.26 million QphH.

OceanBase also became the only Chinese‑developed database to achieve world‑first rankings in both transaction processing (TPC‑C) and data analysis (TPC‑H) benchmarks.

We invited Chen Mengmeng, a core member responsible for the OceanBase test, to share the technical considerations behind this achievement.

HTAP: The Original Purpose and Future of Databases

HTAP (Hybrid Transaction and Analytical Processing) combines online transaction processing (TP) and online analytical processing (AP) in a single database system, eliminating costly ETL processes and enabling faster analysis of the latest data—a key competitive advantage for enterprises.

The concept was introduced by Gartner in 2014, emphasizing the benefits of unified TP and AP workloads.

Historically, TP and AP have evolved through phases of integration, separation, and re‑integration, with early milestones such as Codd's relational model (1970), System R (1974), Jim Gray's transaction theory (late 1970s), and the emergence of data warehouses (1988) and OLAP (1993).

As data volumes grew, the limitations of single‑node databases became apparent, prompting the development of distributed systems capable of handling both TP and AP workloads.

OceanBase Embeds HTAP into Its DNA

OceanBase began internal development within Alibaba in 2010 and evolved into a full‑featured distributed database by 2014. Early on, traditional databases excelled in TP benchmarks, while OceanBase aimed to challenge both TP and AP performance.

Key technical milestones include a cost‑based optimizer (2014), integrated distributed execution (2016), vectorized execution engine and resource isolation for TP/AP (2019‑2020), and the introduction of OceanBase File System (OFS) for virtually unlimited storage.

OceanBase's architecture emphasizes horizontal scalability (e.g., 1,554 nodes in TPC‑C) and vertical scalability, enabling thousands of CPU cores across dozens of machines to serve queries efficiently.

Advanced query optimization, cache‑aware vectorized processing, and just‑in‑time compilation have been crucial for achieving high AP performance, though gaps remain compared to leading competitors.

Three Major Challenges Behind the TPC‑H Record

1. Capacity and scalability: HTAP systems must handle massive data volumes, requiring clusters that can scale to hundreds of nodes.

2. Query optimization and execution: AP workloads involve complex, multi‑table joins, aggregations, and sub‑queries, demanding sophisticated optimizers and high‑throughput execution engines.

3. Resource isolation: Mixing TP and AP requests can cause interference; OceanBase addresses this with virtualized resources and resource groups to isolate workloads.

Where Does HTAP End? Future Directions for OceanBase

OceanBase aims to simplify system complexity while maintaining strong TP performance as its foundation. Future development will continue to enhance HTAP capabilities without compromising transactional strength, targeting broader enterprise adoption and addressing remaining challenges in workload isolation and performance.