Effect of Modeling Uncertainty on Some Routing Problems

In this paper, we consider the routing problem described in Mohanty and Cassandras (Ref. 1). As in Ref. 1, we show that the optimal Bernoulli split to minimize mean time in the system is asymptotically independent of the variance of the service time. We give simple proofs of the results in that paper. We exploit the fact that the optimal split to minimize the mean queueing time is variance independent and the special structure of the Karush–Kuhn–Tucker optimality conditions to derive the optimal solution. Apart from being very straightforward, the proofs also give insight into the reason for the existence of the variance-independent asymptotically optimal routing policy.     

Generalised ‘join the shortest queue’ policies for the dynamic routing of jobs to multi-class queues

Jobs or customers arrive and require service that may be provided at one of several different stations. The associated routing problems concern how customers may be assigned to stations in an optimal manner. Much of the classical literature concerns a single class of customers seeking service from a collection of homogeneous stations. We argue that many contemporary application areas call for the analysis of routing problems in which many classes of customer seek service provided at a collection of diverse stations. This paper is the first to consider routing policies in such complex environments which take appropriate account of the degree of congestion at each service station. A simple and intuitive class of policies emerges from a policy improvement approach. In a numerical study, the policies were close to optimal in all cases.     

Size- and state-aware dispatching problem with queue-specific job sizes

We consider the dispatching problem in a size- and state-aware multi-queue system with Poisson arrivals and queue-specific job sizes. By size- and state-awareness, we mean that the dispatcher knows the size of an arriving job and the remaining service times of the jobs in each queue. By queue-specific job sizes, we mean that the time to process a job may depend on the chosen server. We focus on minimizing the mean sojourn time (i.e., response time) by an MDP approach. First we derive the so-called size-aware relative values of states with respect to the sojourn time in an M/G/1 queue operating under FIFO, LIFO, SPT or SRPT disciplines. For FIFO and LIFO, the size-aware relative values turn out to be insensitive to the form of the job size distribution. The relative values are then exploited in developing efficient dispatching rules in the spirit of the first policy iteration.     

Optimal performance of parallel-server systems with job size prediction errors

Modern communication networks integrate distributed computing architectures, in which customers are processed in parallel. We show how to minimize the waiting time of customer's jobs by leveraging a simple threshold-based job dispatching policy. The optimal policy leverages the SITA routing, which assigns jobs to servers according to the size of the job. Moreover, the optimal policy permits to optimize system performance even when the job size is not known a priori and is estimated by means of error-prone predictors.     

Optimal pricing for tandem queues with finite buffers

We consider optimal pricing for a two-station tandem queueing system with finite buffers, communication blocking, and price-sensitive customers whose arrivals form a homogeneous Poisson process. The service provider quotes prices to incoming customers using either a static or dynamic pricing scheme. There may also be a holding cost for each customer in the system. The objective is to maximize either the discounted profit over an infinite planning horizon or the long-run average profit of the provider. We show that there exists an optimal dynamic policy that exhibits a monotone structure, in which the quoted price is non-decreasing in the queue length at either station and is non-increasing if a customer moves from station 1 to 2, for both the discounted and long-run average problems under certain conditions on the holding costs. We then focus on the long-run average problem and show that the optimal static policy performs as well as the optimal dynamic policy when the buffer size at station 1 becomes large, there are no holding costs, and the arrival rate is either small or large. We learn from numerical results that for systems with small arrival rates and no holding cost, the optimal static policy produces a gain quite close to the optimal gain even when the buffer at station 1 is small. On the other hand, for systems with arrival rates that are not small, there are cases where the optimal dynamic policy performs much better than the optimal static policy.

     

Optimal server scheduling in nonpreemptive finite-population queueing systems

We consider a finite-population queueing system with heterogeneous classes of customers and a single server. For the case of nonpreemptive service, we fully characterize the structure of the server's optimal service policy that minimizes the total average customer waiting costs. We show that the optimal service policy may never serve some classes of customers. For those classes that are served, we show that the optimal service policy is a simple static priority policy. We also derive sufficient conditions that determine the optimal priority sequence.     

Scheduling policies using marked/phantom slot algorithms

We address the problem of schedulingM customer classes in a single-server system, with customers arriving in one ofN arrival streams, as it arises in scheduling transmissions in packet radio networks. In general,NM and a customer from some stream may join one of several classes. We consider a slotted time model where at each scheduling epoch the server (channel) is assigned to a particular class (transmission set) and can serve multiple customers (packets) simultaneously, one from every arrival stream (network node) that can belong to this class. The assignment is based on arandom polling policy: the current time slot is allocated to theith class with probability _i. Our objective is to determine the optimal probabilities by adjusting them on line so as to optimize some overall performance measure. We present an approach based on perturbation analysis techniques, where all customer arrival processes can be arbitrary, and no information about them is required. The basis of this approach is the development of two sensitivity estimators leading to amarked slot and aphantom slot algorithm. The algorithms determine the effect of removing/ adding service slots to an existing schedule on the mean customer waiting times by directly observing the system. The optimal slot assignment probabilities are then used to design adeterministic scheduling policy based on the Golden Ratio policy. Finally, several numerical results based on a simple optimization algorithm are included.This work was supported by the Naval Research Laboratory under contracts N000014-91-J-2025 and N000014-92-J-2017, by the National Science Foundation under grant EID-9212122, and by the Rome Laboratory under contract F30602-94-C-0109.     

Optimal admission policies for a finite queue with bursty arrivals

We consider a service system with a single server, a finite waiting room and two classes of customers with deterministic service time. Primary jobs arrive at random and are admitted whenever there is room in the system. At the beginning of each period, secondary jobs can be admitted from an infinite pool. A revenue is earned upon admission of each job, with the primary jobs bringing a higher contribution than the secondary jobs, the objective being to maximize the total discounted revenue over an infinite horizon. We model the system as a discrete time Markov decision process and show that a monotone admission policy is optimal when the number of primary arrivals has a fixed distribution. Moreover, when the primary arrival distribution varies with time according to a finite state Markov chain, we show that the optimal policy is conditionally monotone and that the numerical computation of an optimal policy, in this case, is substantially more difficult than in the case of stationary arrivals.This research was supported in part by the National Science Foundation, under grant ECS-8803061, while the author was at the University of Arizona.     

Column generation for vehicle routing problems with multiple synchronization constraints

Synchronization of workers and vehicles plays a major role in many industries such as logistics, healthcare or airport ground handling. In this paper, we focus on operational ground handling planning and model it as an archetype of vehicle routing problems with multiple synchronization constraints, coined as “abstract vehicle routing problem with worker and vehicle synchronization” (AVRPWVS). The AVRPWVS deals with routing workers to ground handling jobs such as unloading baggage or refuelling an aircraft, while meeting each job’s time window. Moreover, each job can be performed by a variable number of workers. As airports span vast distances and due to security regulations, workers use vehicles to travel between locations. Furthermore, each vehicle, moved by a driver, can carry several workers. We propose two mathematical multi-commodity flow formulations based on time-space networks to efficiently model five synchronization types including movement and load synchronization. Moreover, we develop a branch-and-price heuristic that employs both conventional variable branching and a novel variable fixing strategy. We demonstrate that the procedure achieves results close to the optimal solution in short time when compared to the two integer models.     

Coordination mechanisms for parallel machine scheduling

We consider coordination mechanisms for the distributed scheduling of n jobs on m parallel machines, where each agent holding a job selects a machine to process his/her own job. Without a central authority to construct a schedule, each agent acts selfishly to minimize his/her own disutility, which is either the completion time of the job or the congestion time (defined as the load of the machine on which the job is scheduled). However, the overall system performance is measured by a central objective which is quite different from the agents’ objective. In the literature, makespan is often considered as the central objective. We, however, investigate problems with other central objectives that minimize the total congestion time, the total completion time, the maximum tardiness, the total tardiness, and the number of tardy jobs. The performance deterioration of the central objective by a lack of central coordination, referred to as the price of anarchy, is typically measured by the maximum ratio of the objective function value of a Nash equilibrium schedule versus that of an optimal, coordinated schedule. In this paper we give bounds for the price of anarchy for the above objectives. For problems with due date related objectives, the price of anarchy may not be defined since the optimal value may be zero. In this case, we consider the maximum difference between the objective function value of an equilibrium schedule and the optimal value. We refer to this metric as the absolute price of anarchy and analyze its lower and upper bounds.     

Optimal control of a single server in a finite-population queueing network

We study the optimal dynamic assignment of a single server to multiple stations in a finite-population queueing network. The objective is to maximize the long-run average reward/throughput. We use sample-path comparisons to identify conditions on the network structure and service time distributions under which the optimal policy is an index policy. This index policy assigns the server to the non-empty station where it takes the shortest amount of time (in some stochastic sense) to complete a job. For example, in a network of multiple parallel stations, the optimal policy assigns the highest priority to the fastest station if service times can be ordered in likelihood ratios. Finally, by means of a numerical study, we test the shortest-expected-remaining-service-time policy on parallel-series networks with three stations and find that this index policy either coincides with the optimal policy or provides a near-optimal performance.     

Admission control with batch arrivals

We consider the problem of dynamic admission control in a multi-class Markovian loss system receiving random batches, where each admitted class-i job demands an exponential service with rate μ, and brings a reward r_i. We show that the optimal admission policy is a sequential threshold policy with monotone thresholds.     

Scheduling in Multiclass Networks with Deterministic Service Times

We consider general feed-forward networks of queues with deterministic service times and arbitrary arrival processes. There are holding costs at each queue, idling may or may not be permitted, and servers may fail. We partially characterize the optimal policy and give conditions under which lower priority should be given to jobs that would be delayed later in the network if they were processed now.     

Analysis of job assignment with batch arrivals among heterogeneous servers

We revisit the problem of job assignment to multiple heterogeneous servers in parallel. The system under consideration, however, has a few unique features. Specifically, repair jobs arrive to the queueing system in batches according to a Poisson process. In addition, servers are heterogeneous and the service time distributions of the individual servers are general. The objective is to optimally assign each job within a batch arrival to minimize the long-run average number of jobs in the entire system. We focus on the class of static assignment policies where jobs are routed to servers upon arrival according to pre-determined probabilities. We solve the model analytically and derive the structural properties of the optimal static assignment. We show that when the traffic is below a certain threshold, it is better to not assign any jobs to slower servers. As traffic increases (either due to an increase in job arrival rate or batch size), more slower servers will be utilized. We give an explicit formula for computing the threshold. Finally we compare and evaluate the performance of the static assignment policy to two dynamic policies, specifically the shortest expected completion policy and the shortest queue policy.     

Asymptotic analysis of the effect of arrival model uncertainties in some optimal routing problems

We study the effect of arrival model uncertainties on the optimal routing in a system of parallel queues. For exponential service time distributions and Bernoulli routing, the optimal mean system delay generally depends on the interarrival time distribution. Any error in modeling the arriving process will cause a model-based optimal routing algorithm to produce a mean system delay higher than the true optimum. In this paper, we present an asymptotic analysis of the behavior of this error under heavy traffic conditions for a general renewal arrival process. An asymptotic analysis of the error in optimal mean delay due to uncertainties in the service time distribution for Poisson arrivals was reported in Ref. 6, where it was shown that, when the first moment of the service time distribution is known, this error in performance vanishes asymptotically as the traffic load approaches the system capacity. In contrast, this paper establishes the somewhat surprising result that, when only the first moment of the arrival distribution is known, the error in optimal mean delay due to uncertainties in the arrival model is unbounded as the traffic approaches the system capacity. However, when both first and second moments are known, the error vanishes asymptotically. Numerical examples corroborating the theoretical results are also presented.This work was supported by the National Science Foundation under Grants ECS-88-01912 and EID-92-12122 and by NASA under Contract NAG 2-595.The authors wish to thank an anonymous referee for pointing out Ref. 20, thus avoiding the need for an explicit proof of convexity of the cost function considered in the paper.     

Structural properties of the optimal resource allocation policy for single-queue systems

This paper studies structural properties of the optimal resource allocation policy for single-queue systems. Jobs arrive at a service facility and are sent one by one to a pool of computing resources for parallel processing. The facility poses a constraint on the maximum expected sojourn time of a job. A central decision maker allocates the servers dynamically to the facility. We consider two models: a limited resource allocation model, where the allocation of resources can only be changed at the start of a new service, and a fully flexible allocation model, where the allocation of resources can also change during a service period. In these two models, the objective is to minimize the average utilization costs whilst satisfying the time constraint. To this end, we cast these optimization problems as Markov decision problems and derive structural properties of the relative value function. We show via dynamic programming that (1) the optimal allocation policy has a work-conservation property, and (2) the optimal number of servers follows a step function with as extreme policy the bang-bang control policy. Moreover, (3) we provide conditions under which the bang-bang control policy takes place. These properties give a full characterization of the optimal policy, which are illustrated by numerical experiments.     

A post-optimization method for the routing and wavelength assignment problem applied to scheduled lightpath demands

We consider here a NP-hard problem related to the Routing and Wavelength Assignment (RWA) problem in optical networks, dealing with Scheduled Lightpath Demands (SLDs). An SLD is a connection demand between two nodes of the network, during a certain time. Given a set of SLDs, we want to assign a lightpath, i.e. a routing path and a wavelength, to each SLD, so that the total number of required wavelengths is minimized. The constraints are the following: a same wavelength must be assigned all along the edges of the routing path of any SLD; at any time, a given wavelength on a given edge of the network cannot be used to satisfy more than one SLD. To solve this problem, we design a post-optimization method improving the solutions provided by a heuristic. The experimental results show that this post-optimization method is quite efficient to reduce the number of necessary wavelengths.     

Optimal Policies for Multi-server Non-preemptive Priority Queues

We consider a multi-server non-preemptive queue with high and low priority customers, and a decision maker who decides when waiting customers may enter service. The goal is to minimize the mean waiting time for high-priority customers while keeping the queue stable. We use a linear programming approach to find and evaluate the performance of an asymptotically optimal policy in the setting of exponential service and inter-arrival times.     

Workforce routing and scheduling for electricity network maintenance with downtime minimization

We investigate a combined routing and scheduling problem for the maintenance of electricity networks. In electricity networks power lines must be regularly maintained to ensure a high quality of service. For safety reasons a power line must be physically disconnected from the network before maintenance work can be performed. After completing maintenance work the power line must be reconnected. Each maintenance job therefore consists of multiple tasks which must be performed at different locations in the network. The goal is to assign each task to a worker and to determine a schedule such that the downtimes of power lines and the travel effort of workers are minimized. For solving this problem, we combine a Large Neighborhood Search meta-heuristic with mathematical programming techniques. The method is evaluated on a large set of test instances which are derived from network data of a German electricity provider.     

Dynamic routing in open queueing networks: Brownian models,cut constraints and resource pooling

We present an introductory review of recent work on the control of open queueing networks. We assume that customers of different types arrive at a network and pass through the system via one of several possible routes; the set of routes available to a customer depends on its type. A route through the network is an ordered set of service stations: a customer queues for service at each station on its route and then leaves the system. The two methods of control we consider are the routing of customers through the network, and the sequencing of service at the stations, and our aim is to minimize the number of customers in the system. We concentrate especially on the insights which can be obtained from heavy traffic analysis, and in particular from Harrison's Brownian network models. Our main conclusion is that in many respects dynamic routingsimplifies the behaviour of networks, and that under good control policies it may well be possible to model the aggregate behaviour of a network quite straightforwardly.Supported by SERC grant GR/F 94194.