Parallel scheduling of multiclass M/M/m queues: Approximate and heavy-traffic optimization of achievable performance

We address the problem of scheduling a multiclass $M/M/m$ queue with Bernoulli feedback on $m$ parallel servers to minimize time-average linear holding costs. We analyze the performance of a heuristic priority-index rule, which extends Klimov's optimal solution to the single-server case: servers select preemptively customers with larger Klimov indices. We present closed-form suboptimality bounds (approximate optimality) for Klimov's rule, which imply that its suboptimality gap is uniformly bounded above with respect to (i) external arrival rates, as long as they stay within system capacity;and (ii) the number of servers. It follows that its relativesuboptimality gap vanishes in a heavy-traffic limit, as external arrival rates approach system capacity (heavy-traffic optimality). We obtain simpler expressions for the special no-feedback case, where the heuristic reduces to the classical $c \mu$ rule. Our analysis is based on comparing the expected cost of Klimov's ruleto the value of a strong linear programming (LP) relaxation of the system's region of achievable performance of mean queue lengths. In order to obtain this relaxation, we derive and exploit a new set ofwork decomposition laws for the parallel-server system. We further report on the results of a computational study on the quality of the $c \mu$ rule for parallel scheduling.

Dynamic perceptual mapping

Perceptual maps have been used for decades by market researchers to illuminatethem about the similarity between brands in terms of a set of attributes, to position consumersrelative to brands in terms of their preferences, or to study how demographic and psychometricvariables relate to consumer choice. Invariably these maps are two-dimensional and static. Aswe enter the era of electronic publishing, the possibilities for dynamic graphics are opening up.We demonstrate the usefulness of introducing motion into perceptual maps through fourexamples. The first example shows how a perceptual map can be viewed in three dimensions,and the second one moves between two analyses of the data that were collected according todifferent protocols. In a third example we move from the best view of the data at the individuallevel to one which focuses on between-group differences in aggregated data. A final exampleconsiders the case when several demographic variables or market segments are available foreach respondent, showing an animation with increasingly detailed demographic comparisons.These examples of dynamic maps use several data sets from marketing and social scienceresearch.

A dynamic choice model of hybrid behavior in the attribute-space

This paper presents a dynamic choice model in the attributespace considering rational consumers that discount the future. In lightof the evidence of several state-dependence patterns, the model isfurther extended by considering a utility function that allows for thedifferent types of behavior described in the literature: pure inertia,pure variety seeking and hybrid. The model presents a stationaryconsumption pattern that can be inertial, where the consumer only buysone product, or a variety-seeking one, where the consumer buys severalproducts simultane-ously. Under the inverted-U marginal utilityassumption, the consumer behaves inertial among the existing brands forseveral periods, and eventually, once the stationary levels areapproached, the consumer turns to a variety-seeking behavior. An empiricalanalysis is run using a scanner database for fabric softener andsignificant evidence of hybrid behavior for most attributes is found,which supports the functional form considered in the theory.

On certain greedoid polyhedra, partially indexable scheduling problems and extended restless bandit allocation indices

We present a polyhedral framework for establishing general structural properties on optimal solutions of stochastic scheduling problems, where multiple job classes vie for service resources: the existence of an optimal priority policy in a given family, characterized by a greedoid(whose feasible class subsets may receive higher priority), where optimal priorities are determined by class-ranking indices, under restricted linear performance objectives (partial indexability). This framework extends that of Bertsimas and Niño-Mora (1996), which explained the optimality of priority-index policies under all linear objectives (general indexability). We show that, if performance measures satisfy partial conservation laws (with respect to the greedoid), which extend previous generalized conservation laws, then theproblem admits a strong LP relaxation over a so-called extended greedoid polytope, which has strong structural and algorithmic properties. We present an adaptive-greedy algorithm (which extends Klimov's) taking as input the linear objective coefficients, which (1) determines whether the optimal LP solution is achievable by a policy in the given family; and (2) if so, computes a set of class-ranking indices that characterize optimal priority policies in the family. In the special case of project scheduling, we show that, under additional conditions, the optimal indices can be computed separately for each project (index decomposition). We further apply the framework to the important restless bandit model (two-action Markov decision chains), obtaining new index policies, that extend Whittle's (1988), and simple sufficient conditions for their validity. These results highlight the power of polyhedral methods (the so-called achievable region approach) in dynamic and stochastic optimization.

Revenue management under general discrete choice model of consumer behavior

Customer choice behavior, such as 'buy-up' and 'buy-down', is an importantphe-nomenon in a wide range of industries. Yet there are few models ormethodologies available to exploit this phenomenon within yield managementsystems. We make some progress on filling this void. Specifically, wedevelop a model of yield management in which the buyers' behavior ismodeled explicitly using a multi-nomial logit model of demand. Thecontrol problem is to decide which subset of fare classes to offer ateach point in time. The set of open fare classes then affects the purchaseprobabilities for each class. We formulate a dynamic program todetermine the optimal control policy and show that it reduces to a dynamicnested allocation policy. Thus, the optimal choice-based policy caneasily be implemented in reservation systems that use nested allocationcontrols. We also develop an estimation procedure for our model based onthe expectation-maximization (EM) method that jointly estimates arrivalrates and choice model parameters when no-purchase outcomes areunobservable. Numerical results show that this combined optimization-estimation approach may significantly improve revenue performancerelative to traditional leg-based models that do not account for choicebehavior.

Dynamic adverse selection and debt

This paper argues that the strategic use of debt favours the revelationof information in dynamic adverse selection problems. Our argument is basedon the idea that debt is a credible commitment to end long term relationships.Consequently, debt encourages a privately informed party to disclose itsinformation at early stages of a relationship. We illustrate our pointwith the financing decision of a monopolist selling a good to a buyerwhose valuation is private information. A high level of (renegotiable)debt, by increasing the scope for liquidation, may induce the highvaluation buyer to buy early at a high price and thus increase themonopolist's expected payoff. By affecting the buyer's strategy, it mayreduce the probability of excessive liquidation. We investigate theconsequences of good durability and we examine the way debt mayalleviate the ratchet effect.

Dynamic graphics of parametrically linked multivariate methods used in compositional data analysis

Many multivariate methods that are apparently distinct can be linked by introducing oneor more parameters in their definition. Methods that can be linked in this way arecorrespondence analysis, unweighted or weighted logratio analysis (the latter alsoknown as "spectral mapping"), nonsymmetric correspondence analysis, principalcomponent analysis (with and without logarithmic transformation of the data) andmultidimensional scaling. In this presentation I will show how several of thesemethods, which are frequently used in compositional data analysis, may be linkedthrough parametrizations such as power transformations, linear transformations andconvex linear combinations. Since the methods of interest here all lead to visual mapsof data, a "movie" can be made where where the linking parameter is allowed to vary insmall steps: the results are recalculated "frame by frame" and one can see the smoothchange from one method to another. Several of these "movies" will be shown, giving adeeper insight into the similarities and differences between these methods.

Implications of dynamic trading for insurance markets

We study the interaction between insurance and capital markets within singlebut general framework.We show that capital markets greatly enhance the risksharing capacity of insurance markets and the scope of risks that areinsurable because efficiency does not depend on the number of agents atrisk, nor on risks being independent, nor on the preferences and endowmentsof agents at risk being the same. We show that agents share risks by buyingfull coverage for their individual risks and provide insurance capitalthrough stock markets.We show that aggregate risk enters private insuranceas positive loading on insurance prices and despite that agents will buyfull coverage. The loading is determined by the risk premium of investorsin the stock market and hence does not depend on the agent s willingnessto pay. Agents provide insurance capital by trading an equally weightedportfolio of insurance company shares and riskless asset. We are able toconstruct agents optimal trading strategies explicitly and for verygeneral preferences.

Dynamic higher order expectations

In models where privately informed agents interact, agents may need to formhigher order expectations, i.e. expectations of other agents' expectations. This paper develops a tractable framework for solving and analyzing linear dynamic rational expectationsmodels in which privately informed agents form higher order expectations. The frameworkis used to demonstrate that the well-known problem of the infinite regress of expectationsidentified by Townsend (1983) can be approximated to an arbitrary accuracy with a finitedimensional representation under quite general conditions. The paper is constructive andpresents a fixed point algorithm for finding an accurate solution and provides weak conditions that ensure that a fixed point exists. To help intuition, Singleton's (1987) asset pricingmodel with disparately informed traders is used as a vehicle for the paper.

Optimization of multiclass queueing networks with changeover times via the achievable region approach: Part I, the single-station case

We address the performance optimization problem in a single-stationmulticlass queueing network with changeover times by means of theachievable region approach. This approach seeks to obtainperformance bounds and scheduling policies from the solution of amathematical program over a relaxation of the system's performanceregion. Relaxed formulations (including linear, convex, nonconvexand positive semidefinite constraints) of this region are developedby formulating equilibrium relations satisfied by the system, withthe help of Palm calculus. Our contributions include: (1) newconstraints formulating equilibrium relations on server dynamics;(2) a flow conservation interpretation of the constraintspreviously derived by the potential function method; (3) newpositive semidefinite constraints; (4) new work decomposition lawsfor single-station multiclass queueing networks, which yield newconvex constraints; (5) a unified buffer occupancy method ofperformance analysis obtained from the constraints; (6) heuristicscheduling policies from the solution of the relaxations.

Restless bandits, linear programming relaxations and a primal-dual index heuristic

We develop a mathematical programming approach for the classicalPSPACE - hard restless bandit problem in stochastic optimization.We introduce a hierarchy of n (where n is the number of bandits)increasingly stronger linear programming relaxations, the lastof which is exact and corresponds to the (exponential size)formulation of the problem as a Markov decision chain, while theother relaxations provide bounds and are efficiently computed. Wealso propose a priority-index heuristic scheduling policy fromthe solution to the first-order relaxation, where the indices aredefined in terms of optimal dual variables. In this way wepropose a policy and a suboptimality guarantee. We report resultsof computational experiments that suggest that the proposedheuristic policy is nearly optimal. Moreover, the second-orderrelaxation is found to provide strong bounds on the optimalvalue.

Optimization of multiclass queueing networks with changeover times via the achievable region method: Part II, the multi-station case

We address the problem of scheduling a multi-station multiclassqueueing network (MQNET) with server changeover times to minimizesteady-state mean job holding costs. We present new lower boundson the best achievable cost that emerge as the values ofmathematical programming problems (linear, semidefinite, andconvex) over relaxed formulations of the system's achievableperformance region. The constraints on achievable performancedefining these formulations are obtained by formulatingsystem's equilibrium relations. Our contributions include: (1) aflow conservation interpretation and closed formulae for theconstraints previously derived by the potential function method;(2) new work decomposition laws for MQNETs; (3) new constraints(linear, convex, and semidefinite) on the performance region offirst and second moments of queue lengths for MQNETs; (4) a fastbound for a MQNET with N customer classes computed in N steps; (5)two heuristic scheduling policies: a priority-index policy, anda policy extracted from the solution of a linear programmingrelaxation.

Insurance with frequent trading: A dynamic analysis of efficient insurance markets

This paper extends existing insurance results on the type of insurance contracts needed for insurance market efficiency toa dynamic setting. It introduces continuosly open markets that allow for more efficient asset allocation. It alsoeliminates the role of preferences and endowments in the classification of risks, which is done primarily in terms of the actuarial properties of the underlying riskprocess. The paper further extends insurability to include correlated and catstrophic events. Under these very general conditions the paper defines a condition that determines whether a small number of standard insurance contracts (together with aggregate assets) suffice to complete markets or one needs to introduce such assets as mutual insurance.

Dynamic optimal law enforcement with learning

We incorporate the process of enforcement learning by assuming that the agency's current marginal cost is a decreasing function of its past experience of detecting and convicting. The agency accumulates data and information (on criminals, on opportunities of crime) enhancing the ability to apprehend in the future at a lower marginal cost.We focus on the impact of enforcement learning on optimal stationary compliance rules. In particular, we show that the optimal stationary fine could be less-than-maximal and the optimal stationary probability of detection could be higher-than-otherwise.

Predicting the potential for lightning activity in mediterranean storms based on the WRF model dynamic and microphysical fields

A new parameter is introduced: the lightning potential index (LPI), which is a measure of the potential for charge generation and separation that leads to lightning flashes in convective thunderstorms. The LPI is calculated within the charge separation region of clouds between 0 C and 20 C, where the noninductive mechanism involving collisions of ice and graupel particles in the presence of supercooled water is most effective. As shown in several case studies using the Weather Research and Forecasting (WRF) model with explicit microphysics, the LPI is highly correlated with observed lightning. It is suggested that the LPI may be a useful parameter for predicting lightning as well as a tool for improving weather forecasting of convective storms and heavy rainfall.