Zhejiang Mobile Cloud-Based Historical Database Transformation —— Building a Cloud-Based Historical Database at Low Cost

l  Low Cost: GBase 8a MPP Cluster runs on low-cost x86 PC servers, significantly reducing hardware investment costs.

l  Elastic Scalability: The cluster efficiently handles petabyte-scale data, meeting storage demands for all historical business data. It shifts from traditional vertical scaling to horizontal scaling based on data volume, enabling dynamic expansion without downtime and ensuring uninterrupted service.

l  High Availability: Backup strategies based on security groups ensure that node failures do not impact service continuity.

l  High Performance: Delivers superior query performance on historical data — monthly data queries in under 3 seconds and statistical processing over 10x faster than the previous system.

This GBase platform solution consists of 28 worker nodes and 2 loading nodes. The cluster is interconnected via a 10 Gigabit Ethernet network. After the transformation, the historical data scale expanded from 100TB to 476TB. It incorporates data from Internet service models and the DW/ST layers generated by the primary data warehouse and big data analytics platform. The system retains detailed billing data for 6+1 months, detailed network usage data for 3+1 months, daily summary data for 3+1 months, and monthly summary data for 12+1 months. It also provides the computation power to support data mining, analysis, and long-term trend analysis on historical data.

Build a cost-effective cloud-based historical data repository that consolidates billing and business analytics historical databases, reconstructs the basic data model for detailed bills, serves as the historical data storage for the primary data warehouse and big data analytics platform, and supports historical data mining, trend analysis, and forecasting.

Zhejiang Mobile's existing business analysis history database lacked network-related data, preventing related applications from being launched. It was urgent to incorporate network data such as Internet logs, DPI, and location signaling. The detailed records in the billing history database and the business analysis history database were redundant and needed to be consolidated, with the business analysis history database absorbing the billing history database for unification. The retention periods of historical data lacked unified management, with varying lengths, making it impossible to meet the business needs of long-term data analysis and mining as well as historical data retrieval and analysis. Building the history database using traditional database architectures would be prohibitively expensive.