PICC Life Insurance Data Warehouse Project

Project Overview

Background

In 2018, China Life procured the GBase 8a MPP Cluster database and deployed a cluster of dozens of nodes to support production systems, including a branch BI data delivery system and a data warehouse upgrade project. This deployment partially alleviated the company's reliance on Oracle and data appliance environments for data-centric projects, meeting computing efficiency requirements while reducing the cost of scaling data appliances.

Since its deployment in 2018, the system has consistently supported the branch BI data delivery system with stable operations. In January 2020, the data warehouse upgrade and renovation project officially commenced, with system construction completed and trial operations initiated in 2021. The trial-phase data warehouse system primarily operates on the GBase database, responsible for data storage across the Operational Data Area, Subject Data Area, Common Data Area, and Application Data Area, as well as data processing logic between these layers. It also handles certain direct data queries and API-based data queries for applications. Based on the current implementation phase, which considers source data, multi-layer data, logic, and supporting applications, it is recommended to scale out the database on the existing foundation to ensure stable and efficient data warehouse operations.

Objectives

The data warehouse upgrade project aims to expand the existing GBase 8a cluster to nearly one hundred nodes, serving as the foundational support for future data systems and numerous data analysis applications, while ensuring stable and efficient operations. Project Requirements

  • Complete node expansion of the GBase 8a cluster with linear resource scaling and proportional performance improvement;

  • Perform online node expansion without impacting business system operations;

  • Monitor the database during expansion, support adjusting the degree of parallelism, and enable control over data redistribution priorities to minimize the impact of data redistribution on business execution during scale-out.

General Data Technology Co., Ltd. (GBase) Solution Success Story: Adjusting the degree of parallelism for data redistribution and supporting priority settings to reduce the impact on business operations during scale-out.

Solutions

To better support PICC Life's business growth and accelerate its digital operations, PICC Life procured a distributed database to complete data architecture upgrade projects. The system is built on GBase 8a MPP Cluster. GBase 8a is purpose-built for analytic applications, using columnar storage as the fundamental data layout and processing model. It integrates advanced technologies such as columnar compression, parallel processing, and rapid intelligent indexing, delivering exceptional performance for query, statistics, analytics, and bulk loading. 

In the first phase, dozens of nodes were deployed. The data warehouse primarily stores data across operational data areas, subject data areas, common data areas, and application data areas, while handling data processing logic between these zones. It also supports direct query access from certain applications and API-based queries. As the business expanded, nearly a hundred additional nodes were added in this phase to support the head office’s reporting system, executive dashboard system, and big data platform.

Application Results

Implementation Status 

Cluster Scale: Currently nearly a hundred data nodes and several coordinator nodes, with coordinator and data nodes running on shared instances (General Data Technology Co., Ltd. GBase solution success case); 

Data Volume: Total data volume is approximately over a hundred TB; 

Daily Incremental Data: Several hundred GB per day; 

Concurrency: Overnight batch processing peaks at tens of concurrent jobs; daytime query concurrency is currently low, peaking at around a few concurrent queries; 

Performance: Overnight batch processing is generally completed within 5 hours. Since files arrive in batches and are loaded immediately upon arrival, this timeframe includes file waiting time, so a direct comparison with previous batch performance is not yet feasible; complex daytime report queries are typically completed within 2 minutes. 

Results and Value 

The GBase 8a cluster has scaled to nearly a hundred nodes, improving the data warehouse's computing power and application outcomes: 

  • Unified data service platform: expanding from dozens to nearly a hundred nodes increases storage capacity, enabling a unified data service to support multi-scenario analytics; 

  • Performance improvement: after expansion, despite a multi-fold increase in data warehouse workloads, branch BI batch processing performance improved by over 70%, with all batch tasks completing within 5 hours; 

  • Cost-effectiveness: achieves a favorable return on investment, offering linear and flexible scalability; 

  • A domestic database solution that meets China's requirements for indigenous, self-controlled information technology systems.