一句话：提出并工业部署 MLOLET：用时间序列模型预测负载测试的吞吐与响应时间并确定提前停止点。
现代大规模系统需应对大量并发请求，负载测试对保障其性能与稳定性至关重要；但传统负载测试执行时间长、成本高，难以适应当代软件短发布周期。本文报告在爱立信部署 MLOLET 这一机器学习优化负载测试框架的工业经验。MLOLET 针对负载测试的关键挑战，从负载测试采集的关键性能指标（KPI）训练时间序列模型，实现测试提前停止点的判定，并预测生产环境中的吞吐量与响应时间趋势。该能力可及早发现系统异常行为、给出较准确的性能预测，帮助负载测试工程师就资源分配做出有依据的决策，从而提升测试效率与系统可靠性。文中介绍了 MLOLET 的设计、在工业环境中的应用以及实施反馈，并说明其对负载测试流程与运营性能的影响。
Load testing is essential for ensuring the performance and stability of modern large-scale systems, which must handle vast numbers of concurrent requests. Traditional load tests, often requiring extensive execution times, are costly and impractical within the short release cycles typical of contemporary software development. In this paper, we present our experience deploying MLOLET, a machine learning optimized load testing framework, at Ericsson. MLOLET addresses key challenges in load testing by determining early stop points for tests and forecasting throughput and response time trends in production environments. By training a time-series model on key performance indicators (KPIs) collected from load tests, MLOLET enables early detection of abnormal system behavior and provides accurate performance forecasting. This capability allows load test engineers to make informed decisions on resource allocation, enhancing both testing efficiency and system reliability. We document the design of MLOLET, its application in industrial settings, and the feedback received from its implementation, highlighting its impact on improving load testing processes and operational performance.