How a Chinese Startup Cracked the 12‑Year TPC‑DS Benchmark: Inside the Database Performance Breakthrough

What is TPC‑DS?

The Transaction Processing Performance Council (TPC) defines standardized benchmarks for data‑management systems. TPC‑DS, introduced in 2006, measures analytical query performance using 99 distinct queries executed against a multi‑step workload.

Benchmark workflow (six steps)

Data Load Test (LD)

Power Test (PT) – run 99 queries in a single stream

Throughput Test 1 (TT1) – run 99 queries across multiple streams

Maintenance Test 1 (DM1) – refresh the database

Throughput Test 2 (TT2) – re‑run 99 queries across multiple streams

Maintenance Test 2 (DM2) – refresh the database again

The queries are generated from templates by the dsqgen tool using a seed value, so parameter values are unknown before execution. Query order varies across streams. Typical data volumes start at 10 TB.

Test environment

The benchmark was executed on a Cisco UCS integrated infrastructure combined with Transwarp Data Hub v5.1. The software stack included StarRing’s distributed graph database StellarDB , flash‑storage database ArgoDB , and the TDH 6.0 big‑data platform.

Official results are published at: http://www.tpc.org/tpcds/results/tpcds_advanced_sort.asp

Why TPC‑DS is difficult

Successful completion requires full SQL compliance, robust transaction support, and scalable execution under massive data‑exchange loads. Traditional RDBMS vendors (e.g., IBM, Oracle, Teradata) have struggled with node‑level bottlenecks at the required scale, often resorting to specialized hardware. Hadoop‑based platforms (e.g., Cloudera, Hortonworks, Databricks) are limited by architectural constraints that prevent execution of the complete benchmark suite.

StarRing testing timeline

StarRing began the first phase (Data Load Test) in late 2014. The remaining five steps were completed over the next three years. Results were submitted to TPC in July of the prior year and underwent a five‑month audit before public release.

Implications

Passing TPC‑DS demonstrates that StarRing’s platform can achieve performance comparable to, or exceeding, leading foreign vendors in large‑scale analytical workloads, providing a new reference point for evaluating high‑performance data‑analytics systems.