Game-Theoretic Analysis Of Queueing Systems Based On Full Rationality And Bounded Rationality

In random service systems,the uncertainty regarding the arrival and service time of customers is prevalent and inherent.Because of limited service resources,the imbal-ance between the supply and demand is rather common,causing queuing phenomena,such as queueing for tickets,seeking for services to doctors,spectrum access,etc.In queueing systems where humans play dominative roles,the subjects of the system,including customers,service providers,managers,and so on,are usually rational.In-corporating the strategic behavior,the operator of the system moves first to select the revenue-optimal or socially optimal price to charge.Then customers make join-or-balk decisions,aiming to maximize their own utilities and thus causing the non-cooperative game among customers,as well as the Stackelberg game between the operator of the system and customers.According to the different levels of customers' rationality,we first assume cus-tomers are fully rational and a retrial queueing system with complementary services is analyzed for cloud computing systems,where the optimal pricing strategies of the service providers are provided.Then considering the trade-off between the power con-sumption and customers' delayed services,we study a retrial queueing system with reserved idle time and setup time,and the optimal pricing strategy and reserved idle time are given.Sequently,a retrial queueing model with unreliable servers is employed to characterize the process secondary users(SUs)access to the spectrum in cognitive radio networks.An admission price is proposed to make SUs' decentralized joining probability coincide with the centralized one.Lastly,for boundedly rational customers,we turn to queueing models with customers' reneging,and the manager's decisions on how to set the admission fee and whether to reveal the information of the system or not are investigated.When customers are fully rational,customers can perfectly anticipate the values of system parameters and the utility of joining the system.Under the assumption of full rationality,firstly,in.cloud computing systems,since the processing of the request generated by the user first requires the scheduler to allocate a computing node for it and then the corresponding computing node provides services for this request,we study a retrial queueing system with two servers providing complementary services.Under the ex-ante and ex-post pricing schemes,we investigate the servers' pricing strategies according to their different ownership.It is found that the optimal price set by the monopolist can also maximize the social welfare.In addition,when the two service providers are private and adopt the ex-post pricing scheme,there is no equilibrium.In other words,the revenue-optimal pricing strategy cannot converge to a stable one in this situation.Secondly,since the idle devices in information and communication technology systems still consume 50%of the energy they consume in their peak processing a job,a reserved idle time is introduced for idle devices,where a longer reserved idle time consumes a large amount of energy,while more idle time reduces customers' sojourn time in the system.In this thesis,we discuss the optimal reserved idle time and optimal pricing strategies for the server and social planner.The results indicate that when the setup cost per time unit is small,turning off the idle devices can improve the server's revenue effectively.Lastly,in cognitive radio networks,when assuming the band is operated by a single primary user(PU)and shared by multiple homogeneous SUs,we investigate the decen-tralized and centralized behavior of SUs,to maximize their own utility and the social welfare respectively.The results show that decentralized joining probability is not less than the centralized one,and an admission price is proposed to make them coincide.Se-quently,when the band is licensed to multiple legal PUs and opportunistically accessed by heterogeneous SUs,the revenue-optimal and socially optimal prices are presented,and the conclusion that the revenue-optimal price is not less than the socially optimal one is drawn.Note that the above analysis is based on the assumption that customers are fully rational.However,in the daily-life,it is not easy for customers to learn the values of system parameters and then compute the expected utility of joining.So we relax this assumption and consider the situation where customers' cognitive ability is limited,i.e.,customers are boundedly rational.Firstly,when the system parameters are known,we capture customers' behavior in an M/M/1 retrial queue through the logit model and the revenue-optimal and socially optimal pricing strategies are studied.We find in the unobservable setting,the revenue-optimal price may be less than the socially optimal price and competition can reduce the price.Then when a stream of customers does not know the information of true service speed,we develop an observable M/M/1 queue-ing system in which the uninformed customers join the queue according to the intrinsic balking thresholds formed by their own experience and others' feedback.However,once these customers are in the line,within-queue observations give them the infor-mation about the true service speed,which may help them to decide to renege or to stay in the queue.In the presence of such reneging behavior,we investigate whether a revenue maximizing or consumer-welfare maximizing server is in its best interest to reveal the true service speed information to customers upfront.It is found revealing the service speed information brings a win-win situation when customers' learning rate is moderate.