Noah Wealth’s CDH-to-Cloud Migration: Boosting OLAP with Hologres

Customer Overview

Noah Holdings Ltd., operating under the "Noah Wealth" brand, is a Chinese independent wealth management firm listed in Hong Kong and the United States. Its data intelligence department builds a big‑data framework to support BI analysis, dashboards, user profiling, and self‑service analytics.

Business Challenges

The rapid digital transformation led to explosive data growth and a fragmented data service landscape using MySQL, Impala, Greenplum, Elasticsearch, and others. Key pain points included:

Insufficient analytical performance for complex, high‑concurrency queries.

Delayed real‑time analytics unable to meet timely decision‑making.

Multiple query engines with different DSLs causing learning overhead and data silos.

Difficulty for analysts to obtain data from disparate systems, leading to manual Excel extraction.

High operational complexity and cost of maintaining a self‑built CDH stack.

Limited scalability and slow query performance of Impala.

Technical Options Comparison

Hologres : Full SQL support, high‑concurrency asynchronous processing, rich dashboard for query logs and slow SQL, multiple storage modes (row, column, hybrid), fast community response.

StarRocks : MySQL‑compatible, supports wide tables and multi‑table joins, limited high‑concurrency support.

ClickHouse : Strong single‑node performance, but limited high‑concurrency (recommended QPS ≈ 100) and higher operational overhead.

Solution

After evaluating the four dimensions, Noah migrated its self‑built CDH environment to Alibaba Cloud’s unified big‑data platform, adopting Hologres as the core OLAP engine. The new architecture integrates DataWorks, MaxCompute for offline processing, Flink for real‑time Kafka streams, and Hologres for both real‑time and batch analytics.

The migration was performed in parallel with the legacy platform, allowing new workloads to run on the cloud while gradually refactoring existing tables into ODS, CDM (DIM, DWD, DWS), and ADS layers. This reduced compute resources by half, clarified DAG dependencies, and eliminated the need for ad‑hoc script modifications.

Business Value

Lowered IDC and operational costs by moving to a cloud‑native, elastic environment.

Enabled sub‑second query responses (average 300 ms) for millions‑row tables, improving analyst productivity.

Supported real‑time, multi‑dimensional advertising analytics via Flink + Hologres, eliminating data loss in previous pipelines.

Accelerated user‑profile and segmentation analysis using Hologres’ Roaring Bitmap functions, achieving sub‑second query times on massive user attribute datasets.

Improved API performance for data services, reducing average response time from 800 ms to 300 ms.