Jiangxi Bank Audit Platform Replacement and Upgrade Project

1 Project Overview

• Background

In recent years, the pace of interest rate liberalization has accelerated and internet finance has grown rapidly, exerting significant pressure on traditional banks. As big data technology matures, it offers higher cost efficiency, more powerful computing and storage capabilities, and easier scalability for data processing—unlocking the full value of complete datasets. A robust big data platform has therefore become essential for banks to adapt to the internet finance era. Many banks are building or have already built such platforms, harnessing advanced IT to drive business growth and generating substantial value through innovative applications.

Jiangxi Bank’s legacy audit system relied on a tightly coupled deployment architecture with DB2 as its database, which posed several challenges:

Performance, storage bottlenecks, and high costs

The traditional centralized DB2 database faced inherent performance and capacity ceilings, depended heavily on high-end hardware such as minicomputers and large-scale storage to sustain performance and stability, and incurred high procurement and maintenance expenses.

Lack of elastic scalability

As data volumes and model complexity grew, overall system throughput hit hardware and technology bottlenecks. In particular, when running multiple models concurrently, the system struggled to handle high-concurrency queries, impeding the efficiency of business audit data analysis.

Need for independently controllable innovation

The core underlying hardware and software technologies in use posed “chokehold” risks, potentially constraining future IT development and digital transformation. An urgent goal was to replace foreign-controlled components—operating systems, chips, databases, and middleware—with China-developed alternatives, ensuring reliability, availability, controllability, and security of the information systems.

2 Solution

Jiangxi Bank’s audit platform was designed as a big data architecture that stores and computes structured data, providing a summarized audit data mart to serve front-end query services. The legacy DB2 system was replaced with the GBase 8a MPP Cluster, a massively parallel distributed database cluster, upgrading the architecture from a traditional centralized model to a distributed cluster. The results exceeded expectations.

The audit platform leverages a highly integrated solution combining GBase 8a MPP Cluster and Hadoop. Data is first processed by Hadoop and then loaded into the main warehouse hosted by GBase 8a MPP Cluster, which provides multi-tenancy features such as model design, report development, version release, scheduling management, and permission control. GBase 8a MPP Cluster employs columnar storage, partitioned storage, smart indexing, and efficient data compression to minimize disk I/O. Its high-availability fault-tolerant design, high-performance data loading, and high-concurrency processing capabilities meet the demands of complex computational analysis, high-concurrency queries, and ad-hoc queries. The online expansion mechanism also allows the bank to deploy additional data analysis and marketing modules as business needs evolve in the future.

Jiangxi Bank Data Platform Architecture Diagram

3 Benefits

The previous audit data mart was built on the DB2 DBF product. After migrating to GBase 8a MPP Cluster, the new data mart delivers significantly higher throughput and approximately 10x performance improvement, empowering the bank to support high-concurrency and large-data-volume scenarios well into the future.

High concurrency: In stress tests with single transactions, the system ran normally under various concurrent user loads and different analytical models.

Responsiveness: Except for data analysis model execution, all system functions responded within 2 seconds. Audit analysts can run data analysis models and obtain results on billion-level datasets in seconds.

Analytical power: The upgrade achieved roughly 10x performance gains over the legacy system, with complex statistical analyses returning results in seconds.

High throughput: The platform sustains over 50 concurrent logged-in users.

Stability: Since go-live, resource utilization has remained stable with no abnormal fluctuations, and both transaction response times and system processing capacity have been consistently reliable.