Data Warehouse IOE Replacement and Upgrade Project with Domestically Developed Technology

Guangxi Mobile leveraged Hadoop, MPP, and other x86-based cloud technologies to eliminate IOE (IBM, Oracle, EMC) dependency in its business analysis system. The overall solution deployed nearly one hundred x86 servers as computing nodes to build a GBase 8a MPP Cluster, which now spans close to one hundred nodes and stores several hundred terabytes of data.

Hadoop is used as the ETL platform at the underlying layer.

The data warehouse built on GBase 8a MPP Cluster not only handles all data modeling and aggregation tasks, but also functions as a data mart. External users can access the cluster to perform self-service query and analysis on authorized data.

Guangxi Mobile’s requirement is to use mature cloud technologies to migrate all data processing from its legacy business analysis system running on minicomputers to x86 platforms. As the business grows, future 4G statistical analysis will be fully handled on the cloud platform, which must support online scaling on demand.

China Mobile Guangxi manages a massive data volume, with over 1 TB of new data added daily. The overall solution deploys an 87-node GBase 8a MPP Cluster using x86 servers as compute nodes, comprising an 11-node GCluster/GCWare cluster and a 76-node GNode cluster. The current total data volume is 610 TB.

The underlying Hadoop platform serves as the ETL system, ingesting interface data from BOSS, customer service, and website systems. Data cleansing and transformation are performed within the Hadoop platform, which also handles lightweight aggregation of the ODS layer to reduce the pressure on the data warehouse.

The data warehouse built on the GBase 8a MPP Cluster not only performs all data modeling and aggregation for the entire warehouse but also functions as data marts. External users can access the cluster to conduct self-service analysis on authorized data.

• High Performance: Data ingestion, aggregation, and query speeds are 10–20x faster than traditional databases.
• Seamless integration with Hadoop platforms for rapid data exchange, enabling hybrid architecture convergence.
• Low Investment, High Efficiency: New system costs only 1/10 of a minicomputer with equivalent computing power.