Business Analysis System Upgrade Project

As China Mobile's business continues to expand, its business analytics systems and thematic big data platforms are required to deliver increasingly comprehensive, in-depth, and efficient data for decision-making. In this context, carriers keep advancing data convergence across business (B), operations (O), and management (M) domains, forcing traditional business analysis systems and big data platforms to handle larger data volumes and workloads.

To address the mounting pressure on Shandong Mobile's business analytics data and various thematic analysis scenarios, a cloud transformation project based on x86 PC servers was deployed. By blending a Hadoop-based ETL platform with an MPP data warehouse platform, it successfully supports the storage of massive historical data and meets diverse big data thematic analysis requirements on Shandong Mobile's big data platform.

(1)    Massive and rapidly growing data scale

l  Active users reach 72 million; the system must ingest traffic analytics data, with an average daily volume exceeding 2.1 TB.

l  Total data scale is expanding rapidly and is approaching the petabyte level.

l  As a forward-looking requirement, compute processing and storage capacity must scale continuously to meet future growth.

(2)    Accelerating existing system responsiveness

l  The system experiences high concurrency and heavy load, requiring urgently improved responsiveness under high-concurrency conditions.

l  Enhance join computation performance across wide tables, including user tags and related attributes.

l  Shorten the end-to-end data production cycle from source extraction to report-layer generation.

(3)    High data availability requirements

l  For all business analytics workloads—including daily, monthly, analytical, reporting, and data mart processing—any delay caused by system failure must not exceed one business day.

l  On the X86 PC server-based architecture, the system must deliver strong fault tolerance and self-healing recovery capabilities.

The overall system architecture leverages a hybrid design combining a Hadoop ETL platform with an MPP-based core analytics database to process and store cross-domain data from BSS, OSS, and MSS domains.

The MPP cluster environment for this project is divided into three functional segments: 26 nodes for the legacy analytical database, 72 nodes for the new analytical database, and 40 nodes for the big data platform. Initial data structuring, cleansing, and light aggregation take place on the Hadoop ETL platform. The processed data is then loaded into the MPP distributed database using the GBase 8a MPP Cluster data distribution tool. As the data foundation for the business analysis system, GBase 8a MPP Cluster manages data processing from the data layer to the shared layer, along with transformations between these layers. Finally, highly aggregated results within the core analytical database are imported into the legacy Oracle database, which serves the data to existing reports and graphical applications..

Enables Deep, Granular Business Analytics: High-efficiency data analytics enable customers to tackle highly complex scenarios demanding real-time insights, performing multidimensional in-depth analysis on diverse data to unlock accurate value.

Supports Massive Data with a Hybrid Architecture: Distributed computing and storage combined with a Hadoop + MPP hybrid structure effectively handles massive datasets.

Low Investment, High Efficiency: The system runs on cost-effective x86 PC servers, delivering performance comparable to the original system while reducing overall costs to one-tenth.