An industrial study of applying input space partitioning to test financial calculation engines

Abstract

This paper presents results from an industrial study that applied input space partitioning and semi-automated requirements modeling to large-scale industrial software, specifically financial calculation engines. Calculation engines are used in financial service applications such as banking, mortgage, insurance, and trading to compute complex, multi-conditional formulas to make high risk financial decisions. They form the heart of financial applications, and can cause severe economic harm if incorrect. Controllability and observability of these calculation engines are low, so robust and sophisticated test methods are needed to ensure the results are valid. However, the industry norm is to use pure human-based, requirements-driven test design, usually with very little automation. The Federal Home Loan Mortgage Corporation (FHLMC), commonly known as Freddie Mac, concerned that these test design techniques may lead to ineffective and inefficient testing, partnered with a university to use high quality, sophisticated test design on several ongoing projects. The goal was to determine if such test design can be cost-effective on this type of critical software. In this study, input space partitioning, along with automation, were applied with the help of several special-purpose tools to validate the effectiveness of input space partitioning. Results showed that these techniques were far more effective (finding more software faults) and more efficient (requiring fewer tests and less labor), and the managers reported that the testing cycle was reduced from five human days to 0.5. This study convinced upper management to begin infusing this approach into other software development projects.