ODS/EDW System Transformation Support Project

Shanghai Telecom's ODS/EDW application systems faced the following challenges:

• Limited database storage capacity;

• ETL performance bottlenecks, with new applications causing resource contention and degrading overall system performance;

To address these issues, a distributed data processing platform was built to complement the ODS/EDW data warehouse. Leveraging a massively parallel processing architecture, the GBase 8a MPP Cluster delivers high throughput and powerful join query performance. It breaks through the performance bottlenecks of loading massive business data, overcomes the drawbacks of traditional relational databases—such as high costs, long query times, and delayed analytical results and business reports—and adopts a PC server model, significantly reducing new storage acquisition costs.

This project uses the GBase 8a MPP Cluster database. All high-level aggregation and analysis of business data, processing and uploading of corporate wide tables, local wide table layer processing, and local summary layer processing have been migrated to the MPP platform. Programs related to the sales control system and panoramic sales views were also migrated.

Accelerated Operations: Enterprise-wide data operations mandated by the group have entered a faster phase, with all product instances, sales offerings, and other data uploaded before 10:00 AM daily.

Ad-hoc Queries: Self-service report queries now respond in seconds, resolving slow performance for large-scale data applications and noticeably improving user experience and satisfaction.

Performance Boost: Harnessing the high efficiency and powerful join capabilities of a massively parallel processing (MPP) architecture, it breaks through performance bottlenecks in loading big data workloads and overcomes the high cost, long processing times, and delayed analytics and reporting typical of traditional relational databases.

High Scalability: The previous vertical scale-up model has shifted to a horizontal scale-out approach based on data volume, supporting online linear scaling where performance improves linearly as nodes are added.