Shanghai Mobile User Profile Database Project — Efficiently Processing Massive Data

With nearly 20 million subscribers, Shanghai Mobile aimed to maintain its strong growth trajectory by integrating data from disparate business systems and consolidating customer profile and behavioral data. The goal was to build a database of customer consumption characteristics, deliver personalized marketing messages based on customer profiles, and embed marketing mechanisms for channel-specific push strategies. This enables timely product recommendations, significantly boosting marketing capabilities, identifying target users and latent demand, and achieving precision marketing.

The project requires building a large-capacity customer feature database that stores over one year of call detail records, more than two years of billing data, and all customer profiles permanently—totaling hundreds of terabytes of raw data storage. It must support near real-time data loading, handle up to 200 concurrent queries with second-level response times, and enable ad-hoc queries, multidimensional analysis, and interactive reporting, delivering visual big data management and fast, intuitive data presentation.

This solution leverages GBase 8a MPP Cluster to build a feature library. It ingests source data from existing systems—such as business analysis platforms and CRM—via file receiving servers and consolidates them into the analytics platform. Built on MPP architecture, the distributed cloud computing data analytics platform delivers exceptional scalability with online dynamic expansion, meeting the massive data storage and processing demands of the feature library. Parallel loading and parallel computing enable near-real-time data ingestion and support high-concurrency requirements. The cluster’s BI tools offer multidimensional analysis, ad-hoc queries, and interactive dashboards.

l  High compression ratio: Provides comprehensive compressed-state storage management for massive data, fully retaining user information and consumption behavior data, laying a solid data foundation for operators to perform more complex customer feature analysis.

l  High performance: 100 concurrent ad-hoc queries achieve sub‑second response. Data loading and model computation speeds are 5–10 times faster than the original system. The system architecture is highly scalable, with performance scaling linearly as nodes increase, guaranteeing high performance for the customer profiling database. This enables comprehensive improvements in product recommendation precision and overall marketing capabilities.