Guangxi Mobile Big Data Platform IOE Replacement and Upgrade Project – Hybrid Architecture Powers System Upgrade

Project Background

With the continuous growth of users and services at Guangxi Mobile, data volumes in the business analysis system surged. Against this backdrop, to meet escalating data and business demands, Guangxi Mobile decided to cloudify its NGBASS system. By adopting x86 cloud-native technologies such as Hadoop and MPP, it aimed to migrate its business analysis system away from IOE architecture, enabling fast processing of massive data and scaling capabilities for future growth.

Requirement Analysis

Guangxi Mobile needed to leverage mature cloud technologies to move all data processing previously handled by minicomputers in the legacy business analysis system onto x86 platforms. As its business continued expanding, all future 4G statistical analysis would also be carried out on the cloudified platform, which required online capacity expansion when necessary.

Solution

With over 1 TB of new data generated daily, Guangxi Mobile deployed an 87-node x86 server cluster to build a GBase 8a MPP Cluster. The cluster comprises 11 GCluster/GCWare nodes and 76 GNode nodes, currently hosting 610 TB of data.

As the foundational ETL layer, a Hadoop platform ingests data from BOSS, customer service, websites, and other interfaces, where cleansing and transformation take place. The Hadoop platform also performs lightweight summarization, offloading ODS-layer aggregation from the data warehouse.

The GBase 8a MPP Cluster serves as both the core data warehouse—handling modeling and aggregation—and the data mart. External users can access the cluster to perform self-service queries and analytics on data within their access permissions.

Value Delivered

• High Performance: Data loading, summarization, and querying are 10–20x faster than traditional databases.

• Seamless Hadoop Integration: Quickly interfaces with Hadoop platforms for efficient data exchange.

• Cost Efficiency: The GBase 8a MPP Cluster delivers equivalent computing power at just one-tenth the cost of a minicomputer-based solution.