WSC 2008

WSC 2008 Final Abstracts

General Applications Track

Wednesday 8:30:00 AM 10:00:00 AM
Business Services Simulation

Chair: Sugato Bagchi (IBM Research)

Constructing Business Simulations with Service Patterns
Richard Lam (IBM T.J. Watson Research Center)

Typically, system dynamics-based simulations of business processes are constructed in an ad hoc manner, with a modeler creating low-level components and defining inter-relationships one-by-one. As the numbers of process components or variables grow, the resulting model becomes ever more difficult to manage and extend. This paper discusses the definition and use of high-level business patterns that have predefined system dynamics sub-models. These patterns enable rapid construction of arbitrarily complex system dynamics models of business processes through abstraction, reuse, and unification of sub-model elements.

Managing Workforce Resource Actions with Multiple Feedback Control Schemes
Young M. Lee, Lianjun An, and Daniel Connors (IBM T.J. Watson Research Center)

Demand disturbances in service businesses are typically managed by resource actions such as hiring, releasing and cross training of the workforce. The magnitudes of resource actions are often decided by estimating the discrepancy between the demand for services and the supply of workforce. However, naive feedback control of the resource actions by policies that equate the discrepancy to the control action can produce undesirable effects such as oscillation between hiring and releasing of workforce, and amplified oscillation through the stages of the service processes. Effective combination of multiple feedback control schemes can produce desirable policies of workforce resource actions. In this work, we study application of control theoretic principles in managing resource actions to see how various feedback control schemes can improve costs, utilization and stability of workforce.

Modeling and Simulation of E-Mail Social Networks: A New Stochastic Agent-Based Approach
Fabian Menges, Giuseppe Narzisi, and Bud Mishra (Courant Institute of Mathematical Sciences, New York University)

Understanding how the structure of a network evolves over time is one of the most interesting and complex topics in the field of social networks. In our attempt to model the dynamics of such systems, we explore an agent-based approach to model growth of email-based social networks, in which individuals establish, maintain and allow atrophy of links through contact-lists and emails. The model is based on the idea of common neighbors, but also on a detailed specialization of the classical preferential attachment theory, thus capturing a deeper understanding of the topology of inter-node connections. In our event-based simulation that schedules the agents' actions over time, the proposed model is amenable to significant efficiency improvements through an application of the Gillespie stochastic simulation schemes. Computer simulation results are used to validate the model by showing that its unique features endow it with ability to simulate real-world email networks with convincing realism.

Wednesday 10:30:00 AM 12:00:00 PM
Technological Enhancements

Chair: Rachel Johnson (Arizona State University)

Generating Artificial Populations Using a Multi-Level Fuzzy Inference Engine
Carlos Ramon Garcia-Alonso and Gabriel Maria Perez-Alcala (ETEA)

The design of complex artificial populations is the first step in simulating evolution during the time span of socio-economic variables as the family income. In this paper, a new hybrid model based on Monte-Carlo simulation and fuzzy inference is described to design environmental conditions, the basic socioeconomic structure and to determine the causes for mortality in an artificial population. The model is based on three main databases that describe the characteristics of the environment, individuals and mechanisms (mortality). These expert-based characteristics guide the simulation model which has a fuzzy inference engine to evaluate fuzzy dependence relationships. These relationships have been formulated to automatically determine complex environmental and individual characteristics as well as mechanism parameters, and they are based on expert knowledge. An artificial population has been designed with satisfactory results when critical design factors are carefully adjusted.

Phrase Based Browsing for Simulation Traces of Network Protocols
Nathan Jay Schmidt and Peter Kemper (College of William and Mary)

Most discrete event simulation frameworks are able to output simulation runs as a trace. The Network Simulator 2 (NS2) is a prominent example that does so to decouple generation of dynamic behavior from its evaluation. If a modeler is interested in the specific details and confronted with lengthy traces from simulation runs, support is needed to identify relevant pieces of information. In this paper, we present a new phrase-based browser that has its roots in information retrieval, language acquisition and text compression which is refined to work with trace data derived from simulation models. The browser is a new navigation feature of Traviando, a trace visualizer and analyzer for simulation traces. The browsing technique allows a modeler to investigate particular patterns seen in a trace, that may be of interest due to their frequent or rare occurrence. We demonstrate how this approach applies to traces generated with NS2.

New Approaches for Inference of Unobservable Queues
Yun Bae Kim and Jinsoo Park (Sungkyunkwan University)

Many inference methods of queueing systems have been developed on the basis of Larson's QIE (queue inference engine) with the assumption of homogeneous Poisson arrivals. It inferred the queueing systems with starting and ending times of service. However, the arrival processes are becoming complex lately, so there are some limits to apply the method. Our study introduces new methods of queue inference which can find the internal behaviors of queueing systems with only external observations, arrival and depar-ture time. This study deals with general GI/G/c queueing systems: (a) FCFC (first come first served), (b) LCFS (last come first served), (c) RSS (random selection for service). The accurate inferences were obtained from FCFS and LCFS systems, and the approximate solutions from RSS systems.