TPC-H
The TPC-H benchmark, developed by the Transaction Processing Performance Council (TPC), is a leading industry-standard benchmark for decision support systems. It is widely recognized by industry and academia as a critical measure of database performance and is a key factor in database selection.
After rigorous large-scale analytics and ACID transaction testing, AnalyticDB for PostgreSQL set a new world record on the TPC-H 30 TB benchmark. It now leads the world in both performance and price-performance, marking the first time a product from China has topped this prestigious list.
On the TPC-H 30,000 SF benchmark, Alibaba Cloud AnalyticDB for PostgreSQL 6.0 achieved a performance of 5,057,263 QphH at a price-performance ratio of 1.46 CNY/QphH. This result is nearly four times the performance of the second-place entry at the same scale, the Cisco UCS C480 M5 running Microsoft SQL Server 2019, which scored 1,278,277 QphH.
The TPC-H benchmark consists of a suite of business-oriented ad-hoc queries and concurrent data modifications. It tests a database's combined transactional and analytical capabilities.
The primary challenges of this benchmark include:
-
A 30 TB dataset: The benchmark uses a 30 TB dataset, with the largest table containing 180 billion rows. This massive scale presents a significant challenge to data import, storage, and computation performance.
-
Complex association analysis: The queries involve operations like multi-table joins, correlated queries, multi-dimensional data filtering, and high-precision calculations. This challenges not only the optimizer (which must handle query decorrelation, optimal join ordering, and table distribution strategies) but also the compute engine's implementation (including memory management, code generation, and execution scheduling).
-
High-volume real-time writes: The benchmark involves write and delete operations on hundreds of millions of records. This tests transaction processing performance and poses a major challenge to the column-oriented storage engine, which is essential for analytical processing (AP) systems.
-
High-throughput concurrent reads and writes: With a 30 TB dataset, the test requires running at least 10 concurrent query streams while executing a refresh stream for multiple rounds of large-scale data insertions and deletions. This challenges the system's concurrency and ACID capabilities.
-
Distributed transaction validation: The benchmark thoroughly tests the database's distributed transaction capabilities, covering all ACID (Atomicity, Consistency, Isolation, and Durability) properties. Tests include shutting down coordinator nodes and compute nodes to evaluate the system's transactional integrity and high availability under failure scenarios, such as a power outage.
Related documents
TPC-H testing requires multiple machine restarts and places strict requirements on overall system capabilities and operational control. For the detailed test procedure, see TPC-H.