Preemptible queues with advance reservations: Strategic behavior and revenue management
Preemptible queues with advance reservations: Strategic behavior and revenue management
复制标题
DOI：
10.1016/j.ejor.2020.12.044
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Jonathan Chamberlain;Eran Simhon;D. Starobinski
中科院分区：
文献类型：
--
作者：
Jonathan Chamberlain;Eran Simhon;D. Starobinski
考虑支持提前预订的 M/G/1 排队系统。在该系统中，战略客户必须决定是否提前预订服务器（从而获得更高的优先级）或放弃预订。提前预订服务器需要付费。提供商可以通过实施几种基于优先级的抢占策略之一来进一步影响客户的预订决策：(i) 任何客户都受到更高优先级客户 (PR) 服务抢占的政策；(ii) 不发生服务抢占的政策 (NP)； (iii) 混合政策，其中只有没有优先预订的客户才会受到服务抢占 (HPR) 的影响。在这项工作中，我们描述了每项政策下客户的战略行为、均衡结果以及提供商的收入最大化。在所有情况下，我们证明（i）唯一可能的纳什均衡类型是基于客户优先级的阈值均衡； (ii) 系统负载影响纳什均衡的结构和数量。我们还证明 HPR 是唯一能够满足 (i) 所有客户都进行预订的均衡的策略； (ii) 第二个服务时刻影响平衡。最后，我们证明，对于任何系统负载和任何服务分配，HPR 策略产生最高的最大收入，其次是 PR 策略和 NP 策略。我们进一步表明，HPR 和 PR 策略的性能相对差异在低系统负载和低服务方差下最大。
Consider an M/G/1 queuing system that supports advance reservations. In this system, strategic customers must decide whether to reserve a server in advance (thereby gaining higher priority) or forgo reservations. Reserving a server in advance bears a cost. The provider can further impact the customers’ reservation decisions via implementation of one of several priority-based preemption policies:(i) one in which any customer is subject to service preemption by a higher priority customer (PR);(ii) one in which service preemption does not occur (NP); and (iii) a hybrid policy in which only customers without a priority reservation are subject to service preemption (HPR). In this work, we characterize the strategic behavior of customers, equilibrium outcomes, and provider’s revenue maximization under each of these policies. In all the cases, we prove that (i) the only possible type of Nash equilibria is a threshold one based on the customers’ priorities; and (ii) the system load impacts both the structure and number of Nash equilibria. We also prove that HPR is the only policy in which (i) an equilibrium where all customers make reservations may exist; and (ii) the second moment of service impacts the equilibria. Finally, we prove that for any system load and any service distribution, the HPR policy yields the highest maximum revenue, followed in turn by the PR policy and the NP policy. We further show that the relative difference in the performance of the HPR and PR policies is greatest at low system load and under low service variance.