Sandeep Juneja

Effective bandwidth and fast simulation of ATM intree networks

Abstract We consider the efficient estimation, via simulation, of very low buffer overflow probabilities in certain acyclic ATM queueing networks. We apply the theory of effective bandwidths and Markov additive processes to derive an asymptotically optimal simulation scheme for estimating such probabilities for a single queue with multiple independent sources, each of which may be either a Markov modulated process or an autoregressive process. This result extends earlier work on queues with either independent arrivals or with a single Markov modulated arrival source. The results are the extended to estimating loss probabilities for intree networks of such queues. Experimental results show that the method can provide many orders of magnitude reduction in variance in complex queueing systems that are not amenable to analysis.

A large deviations perspective on ordinal optimization

We consider the problem of optimal allocation of computing budget to maximize the probability of correct selection in the ordinal optimization setting. This problem has been studied in the literature in an approximate mathematical framework under the assumption that the underlying random variables have a Gaussian distribution. We use the large deviations theory to develop a mathematically rigorous framework for determining the optimal allocation of computing resources even when the underlying variables have general, nonGaussian distributions. Further, in a simple setting we show that when there exists an indifference zone, quick stopping rules may be developed that exploit the exponential decay rates of the probability of false selection. In practice, the distributions of the underlying variables are estimated from generated samples leading to performance degradation due to estimation errors. On a positive note, we show that the corresponding estimates of optimal allocations converge to their true values as the number of samples used for estimation increases to infinity.

Nested Simulation in Portfolio Risk Measurement

Risk measurement for derivative portfolios almost invariably calls for nested simulation. In the outer step, one draws realizations of all risk factors up to the horizon, and in the inner step, one reprices each instrument in the portfolio at the horizon conditional on the drawn risk factors. Practitioners may perceive the computational burden of such nested schemes to be unacceptable and adopt a variety of second-best pricing techniques to avoid the inner simulation. In this paper, we question whether such short cuts are necessary. We show that a relatively small number of trials in the inner step can yield accurate estimates, and we analyze how a fixed computational budget may be allocated to the inner and the outer step to minimize the mean square error of the resultant estimator. Finally, we introduce a jackknife procedure for bias reduction.

Portfolio Credit Risk with Extremal Dependence: Asymptotic Analysis and Efficient Simulation

We consider the risk of a portfolio comprising loans, bonds, and financial instruments that are subject to possible default. In particular, we are interested in performance measures such as the probability that the portfolio incurs large losses over a fixed time horizon, and the expected excess loss given that large losses are incurred during this horizon. Contrary to the normal copula that is commonly used in practice (e.g., in the CreditMetrics system), we assume a portfolio dependence structure that is semiparametric, does not hinge solely on correlation, and supports extremal dependence among obligors. A particular instance within the proposed class of models is the so-called t-copula model that is derived from the multivariate Student t distribution and hence generalizes the normal copula model. The size of the portfolio, the heterogeneous mix of obligors, and the fact that default events are rare and mutually dependent make it quite complicated to calculate portfolio credit risk either by means of exact analysis or naive Monte Carlo simulation. The main contributions of this paper are twofold. We first derive sharp asymptotics for portfolio credit risk that illustrate the implications of extremal dependence among obligors. Using this as a stepping stone, we develop importance-sampling algorithms that are shown to be asymptotically optimal and can be used to efficiently compute portfolio credit risk via Monte Carlo simulation.

Adaptive Importance Sampling Technique for Markov Chains Using Stochastic Approximation

For a discrete-time finite-state Markov chain, we develop an adaptive importance sampling scheme to estimate the expected total cost before hitting a set of terminal states. This scheme updates the change of measure at every transition using constant or decreasing step-size stochastic approximation. The updates are shown to concentrate asymptotically in a neighborhood of the desired zero-variance estimator. Through simulation experiments on simple Markovian queues, we observe that the proposed technique performs very well in estimating performance measures related to rare events associated with queue lengths exceeding prescribed thresholds. We include performance comparisons of the proposed algorithm with existing adaptive importance sampling algorithms on some examples. We also discuss the extension of the technique to estimate the infinite horizon expected discounted cost and the expected average cost.

Stream-Packing: Resource Allocation in Web Server Farms with a QoS Guarantee

Current web server farms have simple resource allocation models. One model used is to dedicate a server or a group of servers for each customer. Another model partitions physical servers into logical servers and assigns one to each customer. Yet another model allows customers to be active on multiple servers using load-balancing techniques. The ability to handle peak loads while minimizing cost of resources required on the farm is a subject of ongoing research.We improve resource utilization through sharing. Customer load is expressed as a multidimensional probability distribution. Each customer is assigned to a server so as to minimize the total number of servers needed to host all the customers. We use the notion of complementarity of customers in simple heuristics for this stochastic vector-packing problem. The proposed method generates a resource allocation plan while guaranteeing a QoS to each customer. Simulation results justify our scheme.

The concert queueing game: to wait or to be late

We introduce the concert (or cafeteria) queueing problem: A finite but large number of customers arrive into a queueing system that starts service at a specified opening time. Each customer is free to choose her arrival time (before or after opening time), and is interested in early service completion with minimal wait. These goals are captured by a cost function which is additive and linear in the waiting time and service completion time, with coefficients that may be class dependent. We consider a fluid model of this system, which is motivated as the fluid-scale limit of the stochastic system. In the fluid setting, we explicitly identify the unique Nash-equilibrium arrival profile for each class of customers. Our structural results imply that, in equilibrium, the arrival rate is increasing up until the closing time where all customers are served. Furthermore, the waiting queue is maximal at the opening time, and monotonically decreases thereafter. In the simple single class setting, we show that the price of anarchy (PoA, the efficiency loss relative to the socially optimal solution) is exactly two, while in the multi-class setting we develop tight upper and lower bounds on the PoA. In addition, we consider several mechanisms that may be used to reduce the PoA. The proposed model may explain queueing phenomena in diverse settings that involve a pre-assigned opening time.

Splitting-based importance-sampling algorithm for fast simulation of Markov reliability models with general repair-policies

Markov chains with small transition probabilities occur while modeling the reliability of systems where the individual components are highly reliable and quickly repairable. Complex inter-component dependencies can exist and the state space involved can be huge, making these models analytically and numerically intractable. Naive simulation is also difficult because the event of interest (system failure) is rare, so that a prohibitively large amount of computation is needed to obtain samples of these events. An earlier paper (Juneja et al., 2001) proposed an importance sampling scheme that provides large efficiency increases over naive simulation for a very general class of models including reliability models with general repair policies such as deferred and group repairs. However, there is a statistical penalty associated with this scheme when the corresponding Markov chain has high probability cycles as may be the case with reliability models with general repair policies. This paper develops a splitting-based importance-sampling technique that avoids this statistical penalty by splitting paths at high probability cycles and thus achieves bounded relative-error in a stronger sense than in previous attempts.

Estimating tail probabilities of heavy tailed distributions with asymptotically zero relative error

Abstract Efficient estimation of tail probabilities involving heavy tailed random variables is amongst the most challenging problems in Monte-Carlo simulation. In the last few years, applied probabilists have achieved considerable success in developing efficient algorithms for some such simple but fundamental tail probabilities. Usually, unbiased importance sampling estimators of such tail probabilities are developed and it is proved that these estimators are asymptotically efficient or even possess the desirable bounded relative error property. In this paper, as an illustration, we consider a simple tail probability involving geometric sums of heavy tailed random variables. This is useful in estimating the probability of large delays in M/G/1 queues. In this setting we develop an unbiased estimator whose relative error decreases to zero asymptotically. The key idea is to decompose the probability of interest into a known dominant component and an unknown small component. Simulation then focuses on estimating the latter ‘residual’ probability. Here we show that the existing conditioning methods or importance sampling methods are not effective in estimating the residual probability while an appropriate combination of the two estimates it with bounded relative error. As a further illustration of the proposed ideas, we apply them to develop an estimator for the probability of large delays in stochastic activity networks that has an asymptotically zero relative error.

Fast simulation of Markovian reliability/availability models with general repair policies

Markovian models of highly reliable systems are considered. An importance sampling based variance reduction technique known as failure biasing has been found to be very useful in the fast Monte Carlo simulation of such systems. The authors show by examples that existing failure biasing heuristics break down for systems which involve more general repair/recovery policies that are common in practice. This motivated a detailed look at the theory of failure biasing from a different perspective than what has been done before, i.e., the effect of failure biasing on sample paths of the Markov chain that involve cycles. This cycling perspective is used to give a much simpler proof of the established fact that existing failure biasing heuristics produce an order of magnitude increase in simulation efficiency over standard simulation, for a class of Markovian systems with simple repair policies. This approach allows the development of theory and efficient heuristics for systems with the more general repair policies.<<ETX>>