11th International Conference on Software Engineering and Applications (SOEA 2027)
January 23 ~ 24, 2027, Zurich, Switzerland
Accepted Papers
Green Architectural Reconstruction: A Petstore Case Study
ABSTRACT
We present an automated architecture reconstruction (AR) framework that uses OpenAPI v3.0 as a language-agnostic intermediate representation to generate Java 21, Python, C#, and Rust service implementations from a legacy Java EE 7 Petstore API. We evaluated the legacy service and all four reconstructed stacks under controlled laboratory conditions (Lab) at loads of up to 1,000 concurrent virtual users. We also compared the two Java stacks in a laboratory scenario labelled Production, which combined network delay, jitter, and packet loss with CPU and memory contention on the server. For each configuration, we conducted 30 trials, measured CPU package energy using Intel RAPL, and calculated mean package power from the energy and elapsed time of each trial. At 1,000 virtual users under Lab conditions, mean package power was 37.8% lower for reconstructed Java than for legacy Java. Across the four reconstructed stacks, mean package power ranged from 16.47 W for Rust to 27.93 W for C#, a spread of 11.46 W. At the same load, the difference in mean package energy between legacy and reconstructed Java narrowed from 8,437.00 J in Lab to 368.07J in Production. For this workload, reconstructed Java used less package energy in both scenarios, with a smaller advantage under the injected disturbances.
KEYWORDS
Architecture reconstruction, Software modernisation, Energy measurement, OpenAPI, Green software engineering