No abstract

Abstract. We investigate the general multi-armed bandit problem with multiple servers. We determine a condition on the reward processes su½cient to guarantee the optimality of the strategy that operates at each instant of time the projects with the highest Gittins indices. We call this strategy the Gittins index rule for multi-armed bandits with multiple plays, or brie¯y the Gittins index rule. We show by examples that: (i) the aforementioned su½cient condition is not necessary for the optimality of the Gittins index rule; and (ii) when the su½cient condition is relaxed the Gittins index rule is not necessarily optimal. Finally, we present an application of the general results to the multiserver scheduling of parallel queues without arrivals.

No abstract

We consider two-stage tandem queuing systems with dedicated servers in each station and flexible servers that can serve in both stations. We assume exponential service times, linear holding costs, and operating costs incurred by the servers at rates proportional to their speeds. Under conditions that ensure the optimality of nonidling policies, we show that the optimal allocation of flexible servers is determined by a transition-monotone policy. Moreover, we present conditions under which the optimal policy can be explicitly determined. Andradottir, Ayhan,, Gel, Hopp, and Van Oyen [10,11], Hopp, Tekin, and Van Oyen [12], Iravani, Van Oyen, and Sims [15], and Ahn and Righter [3]. These models include collaborative and noncollaborative work disciplines, open systems with external arrivals that are independent of the system state, closed systems (e.g., CONWIP systems where a departure triggers a new arrival), and various forms of cross-training (e.g., full, zone, or hierarchical cross-training).There is also much research on the optimal use of flexible servers in tandem systems with holding costs. Because the mathematical models involved are quite complex, most results refer to two-stage systems and exponential service times. Rosberg, Varaiya, and Walrand [19] considered a system with Poisson arrivals, a server with a constant service rate in the downstream station, and a server with a controllable service rate in the upstream station. They showed that the optimal service rate is nondecreasing in the length of the first queue and nonincreasing in the length of the second queue. Weber and Stidham [23] considered a system of n stations with arrivals at each station, where, in addition to convex holding costs, servers incur operating costs that are convex functions of their service rates. They showed that the optimal policy is transition-monotone; that is, when a job leaves a queue, the optimal service rate in that queue is not increased and the optimal service rates in the other queues are not decreased. Duenyas, Gupta, and Olsen [7] and Iravani, Posner, and Buzacott [14] characterized optimal policies for models of n and two-stage systems, respectively, with one flexible server and setup costs. Pandelis and Teneketzis [18] studied two-stage clearing systems with multiple flexible servers where jobs join the second queue with probability p after completing service in the upstream station. They derived conditions under which the policy that gives priority to jobs in the upstream station is optimal for general service times. For a two-stage clearing system with two flexible servers, Ahn, Duenyas, and Zhang [2] provided necessary and sufficient conditions under which an exhaustive policy for the upstream or the downstream station is optimal. Similar results were obtained by Ahn, Duenyas, and Lewis [1] for the model with arrivals. The results of Ahn et al. [2] have been extended in two more directions. First, Schiefermayr and Weichbold [20] obtained the optimal policy for all values of holding costs and service times. Se...

We investigate the impact of switching penalties on the nature of optimal scheduling policies for systems of parallel queues without arrivals. We study two types of switching penalties incurred when switching between queues: lump sum costs and time delays. Under the assumption that the service periods of jobs in a given queue possess the same distribution, we derive an index rule that defines an optimal policy. For switching penalties that depend on the particular nodes involved in a switch, we show that although an index rule is not optimal in general, there is an exhaustive service policy that is optimal.