GBase Enables Seamless Cross-Data Center Migration of Petabyte-Scale Data Warehouse for Shandong Mobile's Big Data Platform
Recently, Shandong Mobile completed the relocation and expansion of its petabyte-scale data warehouse across data centers. The project implementation achieved zero code changes to business applications. By simply modifying the JDBC connection string and restarting services, the system cutover was completed. This truly achieved a completely non-disruptive cross-data center relocation of the core business analysis data warehouse. The expanded system has now been running smoothly for more than 30 days.
Project Background
Since July 2012, Shandong Mobile's business analysis system has been adopting the GBase 8a MPP database to fully replace DB2. After several phases of construction, a big data open operation platform was gradually established by 2020, leveraging MPP database service capabilities to support both internal and external data services on the big data platform.
With a dramatic increase in data volume, the storage utilization of the 72-node cluster for the cloud-based business analysis data warehouse exceeded 90% starting in 2021. Sustained high storage levels could trigger frequent hardware failures and degrade overall database performance. This required the O&M team to constantly monitor the cluster, leading to increased operational costs on one hand and impacting new business development on the other. Therefore, cluster expansion became an urgent problem to solve.
Shandong Mobile's No. 5 Data Center, one of the earliest data centers built by the provincial company, was already operating at full capacity with no remaining space for business expansion. The only option was to relocate and expand the big data platform across data centers. The objective conditions, including the petabyte-scale data volume of the business analysis warehouse, 150,000 tables, over 3,000 daily intra-database jobs, and daily processing times exceeding 16 hours, all posed significant challenges to the migration effort.
Solution
After meticulous planning, GBase developed a two-step relocation and expansion plan:
1. Data Replication Based on GBase 8a Virtual Clusters for Data Center Relocation
A cross-data center dual-active cluster architecture was established using GBase 8a MPP V95. A virtual cluster, VC2, was created in the No. 6 Data Center, configured identically to VC1 in the No. 5 Data Center in terms of cluster size, shard count, distribution, and hash map. After creating VC2, a full data synchronization from the cluster was initiated through a mirroring relationship. Simultaneously, the upper-layer applications were configured to access the Coordinator nodes of VC2 by modifying the JDBC connection. At this stage, the datanode nodes of VC1 continued to provide data services as the production data cluster.
Once data synchronization was complete, the downstream data cluster access for VC2's Coordinators was switched to VC2's datanodes, finalizing the cross-data center migration of the core data warehouse.
This solution ensured zero code changes to business applications. System cutover was completed by merely modifying the JDBC connection string and restarting services, truly achieving a completely non-disruptive cross-data center relocation of the core business analysis data warehouse.
2. Node Expansion of the Migrated Cluster to Increase Storage Capacity and Computing Power of the Big Data Platform
Ultimately, through a 7-day data redistribution process, the expansion from 72 nodes to 158 nodes was completed for the petabyte-scale data. The expanded system has now been running smoothly for over 30 days.
Value Delivered
By leveraging the virtual cluster feature of GBase 8a MPP V95, a mirror relationship is established between two sub-clusters of the same size and data distribution policy. This enables real-time data replication, allowing for the construction of a real-time, dual-active disaster recovery solution for big data business platforms.
The successful migration of Shandong Mobile's main data warehouse validates the feasibility of building dual-active or disaster recovery architectures using virtual clusters. This project has accumulated valuable experience for other provincial companies planning to build similar architectures. It also demonstrates the robust product capabilities of GBase 8a MPP and the independent R&D strength of General Data Technology Co., Ltd., contributing to the comprehensive digital transformation of China Mobile.