DorisDB Powers a Unified High‑Performance OLAP Platform for Beike Real Estate

Beike, a technology‑driven housing service provider, aims to digitize and intelligent‑ify its residential services for hundreds of millions of families, relying on a robust data platform to support various scenarios including second‑hand homes, new homes, rentals, and renovations.

The company's data platform team maintains six to seven different OLAP engines (Impala, Presto, Kylin, Druid, ClickHouse, Hive) to meet diverse analytical needs such as KPI reporting, ad‑hoc exploration, visual dashboards, user behavior analysis, risk control, and DMP functions. This multi‑engine landscape caused high operational overhead and steep learning curves for users.

Since 2021, Beike introduced DorisDB as the primary analysis engine, unifying the metric and reporting platforms under a single component. DorisDB’s MPP architecture, columnar storage, ANSI‑SQL support, multi‑table join capability, materialized view automation, and high‑throughput ingestion dramatically improved query efficiency and reduced maintenance complexity.

Performance Comparison

Against ClickHouse on the SSB benchmark (≈6 billion rows), DorisDB outperformed ClickHouse in 9 of 13 queries when limited to ≤8 threads, and in 7 of 13 queries without thread limits.

In multi‑table join tests, DorisDB was 5‑10× faster than Apache Doris.

High‑concurrency tests showed DorisDB handling 1 500‑2 000 QPS with ~50 ms average latency, compared to Druid’s 600‑700 QPS and ~100 ms latency.

Additional metrics (shown in accompanying charts) further confirm DorisDB’s superiority in throughput, latency, and resource utilization.

Key Benefits in Production

Supports tens of thousands of metric queries per day with sub‑3‑second TP99 latency.

Enables thousands of active visual reports with tens of thousands of daily calls.

Provides real‑time dashboard updates via automatically refreshed materialized views.

Offers flexible data modeling: both wide‑table and multi‑table join modes are efficiently handled.

Reduces operational pressure: a 35‑node cluster (80 cores, 192 GB RAM, 3 TB SSD per node) runs 35 BE and 3 FE instances, covering metric, reporting, and core business scenarios.

Use cases include high‑QPS metric queries during agent performance assessments, real‑time large‑screen dashboards, A/B testing, transaction monitoring, risk control, and live‑streaming platforms, all of which have migrated from ClickHouse or Apache Doris to DorisDB with noticeable latency reductions (often >7×).

Looking forward, Beike plans to extend DorisDB across more business domains, gradually replacing Druid, ClickHouse, and Kylin to build a unified, ultra‑fast OLAP analytics platform.