Shortest-Job-First Scheduling in Many-Server Queues with Impatient Customers and Noisy Service-Time Estimates
研究了在多服务器系统中，即使服务时间预测有噪声，最短作业优先策略也能有效优化性能，并提出了一个更简单的两优先级规则来近似其效果。
Queue scheduling, in which limited resources must be allocated to incoming customers, has numerous applications in service operations management. With increasing data availability and advances in predictive models, personalized scheduling—which leverages individual information about underlying stochastic processes beyond just probability distributions—has gained significant attention. A new study reveals that, even with noisy service-time predictions, the (predicted) shortest-job-first (SJF) policy can effectively optimize performance in many-server systems with inpatient customers. The study also characterizes the impact of prediction errors on the policy’s effectiveness. Additionally, the study shows that a two-class priority rule, in which customers with shorter predicted service times (below a carefully designed threshold) are prioritized, can asymptotically match the performance of SJF, offering a simpler policy for implementation in practice.