一句话：提出 FM4MC，通过切片为部分特征模型并按复杂度选择性 SAT 求解，大幅加速微服务链特征模型验证。
基于 AI 的应用常以微服务链形式部署，其配置问题可用特征模型（Feature Models, FMs）刻画，但模型规模迅速增长，导致现有验证方法不可行。论文提出 FM4MC，一种保持正确性的高效验证方法：首先将大型特征模型切片为部分特征模型（Partial Feature Models, PFMs），然后根据估计的复杂度选择性地调用 SAT 求解器。该组合在保留精确结果的同时显著减少求解器调用次数。作者在按真实重尾规模分布采样的合成特征模型上评估，结果显示 FM4MC 在典型场景中最高加速 23 倍，部分场景超过 100 倍，极端情况下可达 1000 倍，使任务关键时间约束下的微服务 AI 应用配置与验证变得可行。
AI-based applications deployed as microservice chains pose challenging configuration problems, as their Feature Models (FMs) quickly grow to sizes where state-of-the-art validation approaches become infeasible. We propose FM4MC, a correctness-preserving method that validates such models efficiently by (i) slicing them into Partial Feature Models (PFMs) and (ii) applying SAT solving selectively based on estimated complexity. This combination drastically reduces the number of solver calls while retaining exact results. Our evaluation on synthetic FMs, sampled to reflect realistic heavy-tail size distributions, shows that FM4MC validates large models up to 23 × faster in typical scenarios, reaches speedups of more than 100 ×, and scales in extreme cases even to 1, 000 × faster than state-of-the-art techniques. These results demonstrate that FM4MC makes configuration and validation feasible for microservice-based AI applications under mission-critical time constraints, where existing approaches become impractical.
@inproceedings{paperbot11505,
title = {FM4MC: Improving Feature Models for Microservice Chains—Towards More Efficient Configuration and Validation},
author = {Uwe Gropengießer and Paul Wolfart and Julian Liphardt and Max Mühlhäuser},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3744916.3764544}
}