Asymptotically optimal importance sampling for Jackson networks with a tree topology

This note describes an importance sampling (IS) algorithm to estimate buffer overflows of stable Jackson networks with a tree topology. Three new measures of service capacity and traffic in Jackson networks are introduced and the algorithm is defined in their terms. These measures are effective service rate, effective utilization and effective service-to-arrival ratio of a node. They depend on the nonempty/empty states of the queues of the network. For a node with a nonempty queue, the effective service rate equals the node’s nominal service rate. For a node i with an empty queue, it is either a weighted sum of the effective service rates of the nodes receiving traffic directly from node i, or the nominal service rate, whichever smaller. The effective utilization is the ratio of arrival rate to the effective service rate and the effective service-to-arrival ratio is its reciprocal. The rare overflow event of interest is the following: given that initially the network is empty, the system experiences a buffer overflow before returning to the empty state. Two types of buffer structures are considered: (1) a single system-wide buffer shared by all nodes, and (2) each node has its own fixed size buffer. The constructed IS algorithm is asymptotically optimal, i.e., the variance of the associated estimator decays exponentially in the buffer size at the maximum possible rate. This is proved using methods from (Dupuis et al. in Ann. Appl. Probab. 17(4):1306–1346, 2007), which are based on a limit Hamilton–Jacobi–Bellman equation and its boundary conditions and their smooth subsolutions. Numerical examples involving networks with as many as eight nodes are provided.     

Admission control of a service station when the arrivals are internal

All studies in the admission control of a service station make decisions at arrival epochs. When arrivals are internal and are rejected from a queue, the rejected jobs have to be routed to other stations in the system. However the system will not know whether a job will be admitted to a queue or not until its arrival epoch to that queue. Thus, the system has to react dynamically and agilely to the decisions made at a specific queue and may try several queues before finding a queue that admits the job. This paper remedies these difficulties by changing the decision epochs of the admission control from arrival epochs to departure epochs with the actions of switching (keeping) the arrival stream on or off. Thus upstream stations will have information on the admission status of their downstream stations all the time. It is proved that the optimal policy for this revised admission control system is of control limit type for an M/G/1 queue. Comparisons of the optimal values and optimal policies for the admission controls made at arrival epochs and at departure epochs are included in the paper.     

Asymptotic Analysis of Two Coupled Queues with Vastly Different Arrival Rates and Finite Customer Capacities

We consider two coupled queues, with each having a finite capacity of customers. When both queues are nonempty they evolve independently, but when one becomes empty the service rate in the other changes. Such a model corresponds to a generalized processor sharing (GPS) discipline. We study the joint distribution p(m, n) of finding (m, n) customers in the (first, second) queue, in the steady state. We study the problem in an asymptotic limit of “heavy traffic,” where also the arrival rate to the second queue is assumed to be small relative to that of the first. The capacity of the first queue is scaled to be large, while that of the second queue is held constant. We consider several different scalings, and in each case obtain limiting differential and/or difference equation for p(m, n), and these we explicitly solve. We show that our asymptotic approximations are quite accurate numerically. This work supplements previous investigations into this GPS model, which assumed infinite capacities/buffers. The present model corresponds to a random walk in a lattice rectangle, where p(m, n) satisfies a different boundary condition on each edge.     

Strong approximations for a Kumar-Seidman network under a priority service discipline

In this paper,we derive the strong approximations for a four-class two station multi-class queuing network（Kumar-Seidman network） under a priority service discipline.Within a group,jobs are served in the order of arrival,that is,a first-in-first-out disciple,and among groups,jobs are served under a pre-emptiveresume priority disciple.We show that if the input data（i.e.,the arrival and service processe） satisfy an approximation（such as the functional law-of-iterated logarithm approximation or the strong approximation）,the output data（the departure processes） and the performance measures（such as the queue length,the work load and the sojourn time process） satisfy a similar approximation.     

Estimating the throughput of an exponential CONWIP assembly system

We consider a production system consisting of several fabrication lines feeding an assembly station where both fabrication and assembly lines consist of multiple machine exponential workstations and the CONWIP (CONstant Work-In-Process) mechanism is used to regulate work releases. We model this system as an assembly-like queue and develop approximations for the throughput and average number of jobs in queue. These approximations use an estimate of the time that jobs from each line spend waiting for jobs from other lines before being assembled. We use our approximations to gain insight into the related problems of capacity allocation, bottleneck placement and WIP setting.     

On optimal polling policies

In a single-server polling system, the server visits the queues according to a routing policy and while at a queue, serves some or all of the customers there according to a service policy. A polling (or scheduling) policy is a sequence of decisions on whether to serve a customer, idle the server, or switch the server to another queue. The goal of this paper is to find polling policies that stochastically minimize the unfinished work and the number of customers in the system at all times. This optimization problem is decomposed into three subproblems: determine the optimal action (i.e., serve, switch, idle) when the server is at a nonempty queue; determine the optimal action (i.e., switch, idle) when the server empties a queue; determine the optimal routing (i.e., choice of the queue) when the server decides to switch. Under fairly general assumptions, we show for the first subproblem that optimal policies are greedy and exhaustive, i.e., the server should neither idle nor switch when it is at a nonempty queue. For the second subproblem, we prove that in symmetric polling systems patient policies are optimal, i.e., the server should stay idling at the last visited queue whenever the system is empty. When the system is slotted, we further prove that non-idling and impatient policies are optimal. For the third subproblem, we establish that in symmetric polling systems optimal policies belong to the class of Stochastically Largest Queue (SLQ) policies. An SLQ policy is one that never routes the server to a queue known to have a queue length that is stochastically smaller than that of another queue. This result implies, in particular, that the policy that routes the server to the queue with the largest queue length is optimal when all queue lengths are known and that the cyclic routing policy is optimal in the case that the only information available is the previous decisions.This work was supported in part by NSF under Contract ASC-8802764.     

On the Introduction of an Agile, Temporary Workforce into a Tandem Queueing System

We consider a two-station tandem queueing system where customers arrive according to a Poisson process and must receive service at both stations before leaving the system. Neither queue is equipped with dedicated servers. Instead, we consider three scenarios for the fluctuations of workforce level. In the first, a decision-maker can increase and decrease the capacity as is deemed appropriate; the unrestricted case. In the other two cases, workers arrive randomly and can be rejected or allocated to either station. In one case the number of workers can then be reduced (the controlled capacity reduction case). In the other they leave randomly (the uncontrolled capacity reduction case). All servers are capable of working collaboratively on a single job and can work at either station as long as they remain in the system. We show in each scenario that all workers should be allocated to one queue or the other (never split between queues) and that they should serve exhaustively at one of the queues depending on the direction of an inequality. This extends previous studies on flexible systems to the case where the capacity varies over time. We then show in the unrestricted case that the optimal number of workers to have in the system is non-decreasing in the number of customers in either queue. AMS subject classification: 90B22, 90B36     

Work-Conserving Tandem Queues

Consider a tandem queue of two single-server stations with only one server for both stations, who may allocate a fraction α of the service capacity to station 1 and 1−α to station 2 when both are busy. A recent paper treats this model under classical Poisson, exponential assumptions.Using work conservation and FIFO, we show that on every sample path (no stochastic assumptions), the waiting time in system of every customer increases with α. For Poisson arrivals and an arbitrary joint distribution of service times of the same customer at each station, we find the average waiting time at each station for α = 0 and α = 1. We extend these results to k ≥ 3 stations, sample paths that allow for server breakdown and repair, and to a tandem arrangement of single-server tandem queues.This revised version was published online in June 2005 with corrected coverdate     

A tandem queueing model with coupled processors

We consider a tandem queue with coupled processors and analyze the two-dimensional Markov process representing the numbers of jobs in the two stations. A functional equation for the generating function of the stationary distribution of this two-dimensional process is derived and solved through the theory of Riemann-Hilbert boundary value problems.     

Analysis of a Fork/Join Synchronization Station with Inputs from Coxian Servers in a Closed Queuing Network

Fork/join stations are commonly used to model the synchronization constraints in queuing models of computer networks, fabrication/assembly systems and material control strategies for manufacturing systems. This paper presents an exact analysis of a fork/join station in a closed queuing network with inputs from servers with two-phase Coxian service distributions, which models a wide range of variability in the input processes. The underlying queue length and departure processes are analyzed to determine performance measures such as throughput, distributions of the queue length and inter-departure times from the fork/join station. The results show that, for certain parameter settings, variability in the arrival processes has a significant impact on system performance. The model is also used to study the sensitivity of performance measures such as throughput, mean queue lengths, and variability of inter-departure times for a wide range of input parameters and network populations.     

Admission control strategies for tandem Markovian loss systems

Motivated by communication networks, we study an admission control problem for a Markovian loss system comprised of two finite capacity service stations in tandem. Customers arrive to station 1 according to a Poisson process, and a gatekeeper, who has complete knowledge of the number of customers at both stations, decides whether to accept or reject each arriving customer. If a customer is rejected, a rejection cost is incurred. If an admitted customer finds that station 2 is full at the time of his service completion at station 1, he leaves the system and a loss cost is incurred. The goal is to find easy-to-implement policies that minimize long-run average cost per unit time. We formulate two intuitive, extremal policies and provide analytical results on their performances. We also present necessary and/or sufficient conditions under which each of these policies is optimal. Next, we show that for some states of the system it is always optimal to admit new arrivals. We also fully characterize the optimal policy when the capacity of each station is two and discuss some characteristics of optimal policies in general. Finally, we design heuristic admission control policies using these insights. Numerical experiments indicate that these heuristic policies yield near-optimal long-run average cost performance.     

On the stability of a partially accessible multi‐station queue with state‐dependent routing

We consider a multi‐station queue with a multi‐class input process when any station is available for the service of only some (not all) customer classes. Upon arrival, any customer may choose one of its accessible stations according to some state‐dependent policy. We obtain simple stability criteria for this model in two particular cases when service rates are either station‐ or class‐independent. Then, we study a two‐station queue under general assumptions on service rates. Our proofs are based on the fluid approximation approach. This revised version was published online in June 2006 with corrections to the Cover Date.     

Rare event asymptotics for a random walk in the quarter plane

This paper presents a novel technique for deriving asymptotic expressions for the occurrence of rare events for a random walk in the quarter plane. In particular, we study a tandem queue with Poisson arrivals, exponential service times and coupled processors. The service rate for one queue is only a fraction of the global service rate when the other queue is non-empty; when one queue is empty, the other queue has full service rate. The bivariate generating function of the queue lengths gives rise to a functional equation. In order to derive asymptotic expressions for large queue lengths, we combine the kernel method for functional equations with boundary value problems and singularity analysis.     

A Tandem Queue with Blocking and Markovian Arrival Process

Queueing networks with blocking have proved useful in modelling of data communications and production lines. We study such a network consisting of a sequence of two service stations with an infinite queue allowed before the first station and no intermediate queue allowed between them. This restriction results in the blocking of the first station whenever a unit having completed its service in that station cannot enter into the second one due to the presence of another unit there. The input of units to the network is the MAP (Markovian Arrival Process). At the first station, service requirements are of phase type whereas service times at the second station are arbitrarily distributed. The focus is on the embedded process at departures. The essential tool in our analysis is the general theory on Markov renewal processes of M/G/1-type.     

A simple heuristic for load balancing in parallel processing networks with highly variable service time distributions

Suppose that customers arrive at a service center (call center, web server, etc.) with two stations in accordance with independent Poisson processes. Service times at either station follow the same general distribution, are independent of each other and are independent of the arrival process. The system is charged station-dependent holding costs at each station per customer per unit time. At any point in time, a decision-maker may decide to move, at a cost, some number of jobs in one queue to the other. The goals of this paper are twofold. First, we are interested in providing insights into this decision-making scenario. We do so, in the important case that the service time distribution is highly variable or simply has a heavy tail. Secondly, we propose that the savvy use of Markov decision processes can lead to easily implementable heuristics when features of the service time distribution can be captured by introducing multiple customer classes. To this end, we consider a two-station proxy for the original system, where the service times are assumed to be exponential, but of one of two classes with different rates. We prove structural results for this proxy and show that these results lead to heuristics that perform well.     

On a Two-Queue Priority System with Impatience and its Application to a Call Center*

We consider an s-server priority system with a protected and an unprotected queue. The arrival rates at the queues and the service rate may depend on the number n of customers being in service or in the protected queue, but the service rate is assumed to be constant for n > s. As soon as any server is idle, a customer from the protected queue will be served according to the FCFS discipline. However, the customers in the protected queue are impatient. If the offered waiting time exceeds a random maximal waiting time I, then the customer leaves the protected queue after time I. If I is less than a given deterministic time, then he leaves the system, else he will be transferred by the system to the unprotected queue. The service of a customer from the unprotected queue will be started if the protected queue is empty and more than a given number of servers become idle. The model is a generalization of the many-server queue with impatient customers. The global balance conditions seem to have no explicit solution. However, the balance conditions for the density of the stationary state process for the subsystem of customers being in service or in the protected queue can be solved. This yields the stability conditions and the probabilities that precisely n customers are in service or in the protected queue. For obtaining performance measures for the unprotected queue, a system approximation based on fitting impatience intensities is constructed. The results are applied to the performance analysis of a call center with an integrated voice-mail-server.     

Optimal control of a queue under a quality-of-service constraint with bounded and unbounded rates

We consider a simple Markovian queue with Poisson arrivals and exponential service times for jobs. The controller chooses state-dependent service rates from an action space. The queue has a finite buffer, and when full, new jobs get rejected. The controller’s objective is to choose optimal service rates that meet a quality-of-service constraint. We solve this problem analytically and compute it numerically under two cases: When the action space is unbounded and when it is bounded.     

Tandem Queues with Dynamic Priorities

In a line production system, if the sequencing at each work station is done according to the times that the jobs are due out of the system then the sequencing is according to what is called the dynamic priority rule. The priorities are dynamic because the longer a job waits in a queue for service, the less the likelihood that a later arrival will precede it. In this paper interest is focused on the equilibrium probability distribution of the time that a job spends in such a system (called the flow time). Reported here are results of simulation studies which suggest a technique for locating these distributions graphically from theoretically derived flow time distributions for a similar system, but in which queue discipline is governed by a first-come, first-served rule.     

A heavy-traffic analysis of a closed queueing system with a GI/∞ service center

This paper studies the heavy-traffic behavior of a closed system consisting of two service stations. The first station is an infinite server and the second is a single server whose service rate depends on the size of the queue at the station. We consider the regime when both the number of customers, n, and the service rate at the single-server station go to infinity while the service rate at the infinite-server station is held fixed. We show that, as n→∞, the process of the number of customers at the infinite-server station normalized by n converges in probability to a deterministic function satisfying a Volterra integral equation. The deviations of the normalized queue from its deterministic limit multiplied by √n converge in distribution to the solution of a stochastic Volterra equation. The proof uses a new approach to studying infinite-server queues in heavy traffic whose main novelty is to express the number of customers at the infinite server as a time-space integral with respect to a time-changed sequential empirical process. This gives a new insight into the structure of the limit processes and makes the end results easy to interpret. Also the approach allows us to give a version of the classical heavy-traffic limit theorem for the G/GI/∞ queue which, in particular, reconciles the limits obtained earlier by Iglehart and Borovkov. This revised version was published online in June 2006 with corrections to the Cover Date.     

Optimality of routing and servicing in dependent parallel processing systems

In a system of dependent, parallel processing service stations, when is it optimal to route customers to the shortest queue and to devote auxiliary capacity to serve the longest queue? We show that this RSQ/SLQ policy is optimal for a wide class of Markovian systems, where the arrival and service rates at the stations, which may depend on the numbers of customers at all the stations, satisfy certain symmetry and monotonicity conditions. Under this policy, the queue lengths will be stochastically smaller in the weak submajorization ordering than the queue lengths under any other policy. Furthermore, this policy minimizes standard discounted and average cost functionals over finite and infinite horizons.