HTAP benchmark in financial scenariosOACSCD

With the development of HTAP(Hybrid Transactional/Analytical Processing),the use of HTAP databases has become increasingly common. HTAP databases integrate OLTP(Online Transaction Processing) and OLAP(Online Analytical Processing) into a single system, replacing traditional ETL methods and thereby reducing the cost of system usage and maintenance. However, financial scenarios impose higher demands on data and service solutions, with significantly more complex data structures and business processes.This complexity poses a series of challenges for the application of HTAP databases in financial scenarios. To address these challenges, this paper proposes a benchmark tailored for financial scenarios. This benchmark evaluates the key characteristics of HTAP systems in financial contexts: throughput frontier and freshness. It simulates data distributions specific to financial scenarios and generates data consistent with financial use cases. Based on financial data patterns, it defines a variety of analytical queries, including those for marketing,risk control, single-table aggregation, and multi-table joins. It also simulates common banking transactions such as account operations, deposits, and withdrawals, and supports risk detection rollback to reflect the real-world behavior of banking transactions. The testing program allows the generation of transactional and analytical clients in varying proportions based on parameters, enabling concurrent database testing. This paper presents adaptations of the benchmark to a community edition of a specific database for testing purposes. Considering the growing adoption of cloud databases, where resource allocation has become more flexible, this paper also investigates the changes in database performance and workload isolation under different resource allocation scenarios for transactional and analytical clients. The study finds that appropriate resource allocation can significantly enhance workload isolation.Additionally, the benchmark allows database users to assign intent weights to transactional and analytical workloads, enabling a unified performance evaluation metric for the overall performance of the database. A comparison of two different resource allocation strategies shows that appropriate resource allocation improved the comprehensive performance score of the database from 55.0 to 60.0.