Adopting StarRocks for Real‑Time Analytics in ZhongAn’s JiZhi Platform: A Performance Comparison with ClickHouse

In recent years ZhongAn Insurance has built the JiZhi platform to provide zero‑code, drag‑and‑drop analytics for dozens of business lines, achieving over 50% faster data analysis and a 40% reduction in labor cost.

The platform originally used ClickHouse as its unified OLAP engine, but growing data volume and concurrent dashboard usage exposed several limitations: poor multi‑query performance, slow multi‑table joins, lack of transactional DDL/DML, heavy reliance on ZooKeeper, and no automatic resharding.

To address these issues, the team evaluated StarRocks, a next‑generation MPP OLAP engine. StarRocks offers high‑concurrency query support (over 10 k QPS), distributed join strategies, MySQL‑compatible transactional DDL/DML, and a simple FE/BE architecture without external dependencies.

Benchmark tests using the SSB dataset on four 8‑core machines showed that StarRocks matches ClickHouse on single‑threaded queries, outperforms it by 1.8× on single‑table multi‑concurrency, and up to 8× on multi‑table multi‑concurrency scenarios. For real‑time workloads, StarRocks’ Primary‑Key model delivered 3‑10× faster queries than ClickHouse’s Replacing engine and more stable performance.

Based on these results, JiZhi integrated StarRocks for its real‑time data warehouse. Users can now create data models that automatically generate Flink SQL sink statements and obtain StarRocks connection details, enabling rapid construction of real‑time dashboards with lower latency (≈3 s vs >10 s) and support for near‑hundred‑million‑row datasets.

Beyond real‑time analytics, ZhongAn plans to extend StarRocks to offline scenarios, lightweight data‑warehouse use cases, and user‑behavior analysis, consolidating its OLAP capabilities into a single engine.