Inferring Queue Performance from Transactional Data

在排队等候或排队等候时出现的拥塞（如 电信、零售、ATM等）是这方面的重点 研究提案。 该提案建立在一个新的统计数据基础上， 推理方法-“队列推理引擎（QIE）”-计算 几乎所有相关队列的统计无偏估计 仅从交易数据中衡量业绩， 需要队列观测或参数估计。 项目 以四种方式建立在早期工作的基础上（1）推导出各种界限 以及加速计算和便于分析的近似值 非常拥挤的系统;（2）创建新的队列推理 算法在更复杂的设置;（3）推导出等效 QIE结果的统计置信限， 可以事先知道所需的样本量;（4）尝试 在QIE中创建服务器分配或调度算法 框架，这将是数据密集型的，基本上没有参数 并且将准确地反映瞬态操作。

Congestion as occurs in waiting lines or queues (as found in telecommunications, retailing, ATM's, etc) is the focus of this research proposal. The proposal builds on a new statistical inference method -- "Queue Inference Engine (QIE)" -- that computes statistically unbiased estimates of virtually all relevant queue performance measures from the transactional data alone, with no queue observation or parameter estimation required. The project builds on earlier work in four ways (1) to derive various bounds and approximations that speed computation and facilitate analysis of very congested systems; (2) to create new queue inference algorithms in more complex settings; (3) to derive the equivalent of statistical confidence limits on the QIE results, so that one can know a priori required sample sizes; and (4) to attempt to create a server allocation or scheduling algorithm within the QIE framework, that would be data intensive, essentially parameter free and would reflect accurately transient operations.