PICC Property and Casualty Company Limited Data Platform Project

Background

The original PICC data platform was built on a Teradata appliance, resulting in tight hardware-software coupling and high operational and maintenance costs. Currently, storage capacity and processing performance have reached a bottleneck, making vertical scaling impossible without service interruption.

This project aims to rebuild the system using an MPP database product, leveraging a massively parallel processing architecture on open x86 hardware to build a big data platform with distributed computing and linear horizontal scalability. Independent platforms will be established for different application scenarios to achieve workload isolation and avoid resource contention, thereby addressing the storage and computational performance bottlenecks of PICC Property and Casualty Company Limited (PICC P&C).

Requirements

In the future, PICC P&C will maintain a data service platform where the Teradata appliance and MPP database coexist.

1. Migrate certain functionalities running on Teradata—such as the business analysis platform, auto insurance analysis, and claims analysis—to the MPP database, reducing the load on the Teradata appliance.

2. Build a big data resource center using the MPP database to facilitate data integration and sharing, and enhance data asset operation and management capabilities.

3. Migrate data analysis systems currently running on Oracle and Informix to the MPP database, significantly reducing costs while improving system efficiency, and providing better data services for PICC P&C’s business analysis and development.

To address PICC's specific requirements and future roadmap, GBase 8a cluster was deployed to build a big data resource center. Leveraging GBase 8a's columnar storage, intelligent indexing, linear scalability, and distributed parallel computing, the solution eliminates the performance bottlenecks and scaling limitations previously experienced with Teradata appliances and Oracle. It enables unified integration, management, and scheduling of data analytics, laying a solid foundation for PICC's new strategic goals.

The GBase 8a cluster database forms the core of the big data resource center, promoting the convergence and sharing of data resources and strengthening data asset operations and management capabilities. The overall system architecture is shown below:

PICC Data Platform Architecture

Migrating the existing analytics systems from Oracle and Teradata to the MPP database dramatically reduces costs while boosting operational efficiency, empowering PICC with superior data services for business analysis and growth.

A big data lake platform based on GBase 8a MPP Cluster ingests upstream business data and gradually offloads main warehouse workloads. Starting with data marts, it progressively reduces the load on the primary Teradata warehouse, decouples core operations, and transitions from parallel operation to phased migration and full replacement.

Implementation

PICC has deployed over a hundred nodes across multiple production clusters, managing several petabytes of data. In the ACRM system, clusters A and B handle primary business functions, enabling rapid data publishing for daily incremental data and monthly reports. After migrating from Oracle, overall performance has improved by 2 to 10 times. The data layers SGA, ODS, F, DW, and DM total hundreds of terabytes, with over a thousand programs successfully migrated.

Benefits and Value

High Performance: With intelligent indexing and a fully parallel architecture, the system delivers lightning-fast query analysis, fully supporting high-performance analytical workloads. Migrating the ACRM system from Oracle on IBM P780 servers to a GBase 8a cluster on x86 servers (with comparable hardware specs) resulted in a 2-10x performance boost.

Scalability: An open architecture allows on-demand horizontal scaling. The Shared-Nothing design supports online expansion without service interruption.

High Reliability: All components feature high availability with no single point of failure. The federated architecture ensures full HA design for management and data nodes, and supports active-active clusters to maximize system availability.

Easy Management: One-click node replacement simplifies hardware fault handling. The DBLink tool between clusters enables cross-cluster data access, while centralized management and monitoring tools provide a graphical interface for operating multiple clusters.

Expert Services: GBase’s original team provided PICC with data warehouse migration solutions (from Teradata) and Oracle migration expertise along with dedicated support.