Prof Hassan Rajaei Ph.D. – Bowling Green State University; Department of Computer Science

Objectives

Simulation is a powerful tool for teaching students about the techniques as well as providing deeper understanding of courses such as networking, operating systems, operational research, just to name a few. Simulation is a well-known technique for evaluating what-if scenarios and decision making in industry, defense, finance, and many others. Students quickly realize these values and want to learn how to master this technique.

Teaching simulation techniques often requires attractive problem assignments. Real-world has numerous examples that excite students to study and motivate them focusing on their learning objectives. Further, it challenges them to develop models to reflect the reality. Clear examples can teach students how to collect data, develop the base model, improve it to advanced models, analyze the obtained results, and think about the usability of their simulation results. These learning outcomes can clearly demonstrate valuable educational objectives.

A simulation tool like ProModel has numerous example models in its library, but the educational objectives can be best achieved through step-by-step experimental development of useful samples. ProModel can be a great help by exploring the details of similar examples.

This article, presents an example where a group of students developed a simulation model for the Bowling Green State University (BGSU) Students Union Cafeteria. Managing a university dining hall often exhibits challenges for the food services located in it. This study focused on reducing the average waiting time of the diners, while increasing overall efficiency of the services.

Simulating the Nest Cafeteria

This project focused on finding solutions for the Falcon’s Nest Cafeteria to increase the efficiency and decrease the average time of the customer spent in the system.

Overview of the Nest: Students cafeteria at BGSU functions as an important part of the University’s dining service. This cafeteria serves thousands of students every day. During the rush-hours of lunch and dinner, this place gets really congested with long queues contributing to long waiting times. In this simulation, the Nest model consists of five main components: Customers, Servers, Locations, Queues, and Cashiers.

Using ProModel: This tool was selected for multiple reasons: a) the availability; b) the course used the tool and trained students; c) the tool supports discrete-event systems; d) large number of library models; e) statistical analysis and output results; f) animations.

Problems Encountered: The main problem faced was lack of statistics and accurate information. Other barriers included project time limit and lack of deeper familiarity of ProModel.

Possible Solutions: Based on primary analysis, two potential solutions were feasible:

1) Increase the attractiveness of other food stations which have lower waiting time;

2) Increase number of food servers.

Three approaches to reach the goals:

a) Ask the SME to provide all data and statistics;

b) Make a very detailed model over the actual system;

c) Combination of (a) & (b) methods.

Approach c was adopted for the study.

Simulation Models

Four models were developed: 1) base, 2) intermediate, 3) advanced, 4) final

Base Model: The base model had very basic setups with one food station and one cashier. The objective was to test the station service and the customers’ arrival, and their flow in the system.

Intermediate Model: All food stations were added according to the Nest along with the logic for entities to move through the system with a shared queue.

 Advanced Model: The advanced model includes all queues targeting to obtain realistic statistics using several scenarios (Figure 1).

Figure 1:  The Advanced Model improved from the intermediate one

Final Model: After developing three scenarios, obtaining good confidence, making sure they were on the right track, students moved towards developing the final model shown in Figure 2. It was implemented using a time schedule to simulate the rush hour and normal operating hours.

Figure 2:  Final simulation model for the Falcon’s Nest Cafeteria

Results and Analysis

In this simulation, students first aimed to find an ideal solution to demonstrate how to reduce the waiting time. It turned out that such a scenario would need more implementation time. Instead, students focused on two solutions:

1) To make other food stations more attractive;

2) Adding additional workers to the top three food stations. Test cases were developed for each solution.

The result shown in Figure 3 demonstrate a reduction in the average time compared to the baseline, except Case 3. The figure suggests an 11.1% decrease in average time spent in the Nest.

Figure 3: Solution 1 demonstrating reduction of Average Time in the system

Next method focused on improving the waiting time by adding food runners to 3 populated stations. This method was simulated and tested with 4 scenarios, and was compared with the baseline.

Figure 4: Solution 2, advocating one additional worker at each food station

As was expected, by adding a food-runner to each station the average time of the customers would decrease, however, certain stations would benefit most. If case 4 is adopted, there would be a 12.6% reduction in time spent by customers. If only 1 food runner is added, then the result yields only to 6.1% decrease in average time spent in the system by customers.

Concluding Remarks

This article presents an example of a real-world case study conducted by a group of students as a term project in a simulation techniques course shared by senior undergraduate students as well as graduate students. An important result of this study demonstrates how deeply the students were engaged in their learning objectives of the course. In a short period of time, they conducted a complete case study including: observation, gathering data, analyzing the problem at hand, developing models, confirming with the subject matter expert, documenting, and delivering the results. The full article is published in ASEE 2017 Annual Conference.

Professor Hassan Rajaei Ph.D.  

Hassan Rajaei is a Professor of Computer Science at Bowling Green State University, Ohio.  His research interests include Distributed Systems & IoT, Cloud Computing, High Performance Computing (HPC), Computer Simulation, Distributed Simulation, with applications focus on communications & wireless networks. Dr. Rajaei has been active in simulation conferences (e.g. SCS SprintSim, WSC) as organizer as well as research contributor. Dr. Rajaei received his Ph.D. from Royal Institute of Technologies, KTH, Stockholm, Sweden and he holds a MSEE from Univ. of Utah.

Dr. Farhed Moeeni – Prof. of Computer & Information Technology, Arkansas State University

GUEST BLOGGER – Dr. Farhed Moeeni

In an earlier blog I discussed some of the issues related to modeling customers’ behavior; such as impatience, especially the balking behavior.  Customers’ impatience can lead to certain behaviors such as balking, reneging and jockeying.  Ignoring the case of balking at arrival caused by queue capacity, balking happens when customers assess the waiting line, develop some opinion about the required waiting time (likely by scanning the queue length) and decide whether to join the queue or not. In short, balking happens when the person’s tolerance for waiting is less than the anticipated waiting time at arrival.  Jockeying happens when two or more service channels provide identical services, independently and with distinct waiting lines.  A customer who is presently waiting in one of the queues chooses to leave the current line and join another queue in the hopes of less waiting time.

Reneging, however happens after a person joins a queue.  In other words, the initial anticipated delay was tolerable.  But later, as time passes, the person becomes impatient and abandons the queue without receiving the service because the prospect of being served within the initially anticipated time diminishes or the predicted remaining waiting time exceeds the customer’s tolerance threshold for further waiting.  Other incidents such as receiving an urgent call may also cause abandoning of the queue.  However, this or similar scenarios are not caused by impatience.

Researchers believe the tolerance threshold is not a constant and can change over the course of the waiting period.  For example, studies indicate that the probability of reneging decreases as the person continues to wait, but only for a certain amount of waiting time after which the customer starts losing patience and the probability of reneging increases.  Another interesting conjecture proposed in the literature is that as the number of people behind a waiting customer increases, the likelihood of the customer reneging decreases.  The decision to renege is also influenced by other factors, for example, personal differences, prior experiences with similar situations, the critical nature of the service to be received, attributes and atmosphere of the waiting area’s surrounding (service-scape), and whether or not an alternative date or time for the same service or a similar service is available. Thus, the decision process for reneging is more complex than balking and modeling and simulating reneging is also more complex.

Obviously, it is not possible to directly characterize the complex behavioral aspects of customers into the simulation model.  Each reneging incident is the result of complex interactions among many factors and the result of a thought process that eventually causes a customer to become impatient and abandon the queue. In other words, reneging cases are not random events; however, from the perspective of a simulation modeler or an observer who is watching the queue, the incidents may look like a random process.

In discrete-event simulation, the modeler has a number of mechanisms or tools available. For example, the modeler may apply some probability distribution to characterize the uncertain behavior of customers or formulate some rules to trigger reneging during the simulation runs.  In addition to probability distributions, the modeler may also identify and apply a number of observable predictor variables to fine tune the reneging behavior.  These variables may include the waiting time of each customer in queue before reneging incidents, the position of each customer in queue just before reneging, the number of customers in queue behind the reneging customer at the moment of abandoning the queue, etc.

Incorporating the above mechanisms is not too complex from the modeling and coding perspective. On the other hand, it is difficult to collect relevant data from the field if the required information is not available. A number of obstacles can contribute to these difficulties. First, sufficient observations are needed to estimate the probability distributions, the shapes and parameters with adequate precision.  Unfortunately, in many instances, reneging does not happen often enough to yield sufficient observational data from the field within a reasonable amount of time.

Second, accurate data collection for events such as reneging may need sophisticated data collection procedures because each customer should be tracked while in the system and the values of relevant state variables along with other important information such as the “waiting time in queue before reneging” and “the position of the customer in queue” before abandoning the queue should also be recorded.   Only the information related to those customers who renege is needed. However, those who eventually renege are not known in advance or at the time of joining the queue; thus everyone should be tracked.  Such elaborate data collection system may be costly to implement, intrusive to customers, or may need consent and collaboration from people.  The latter may also influence the behavior and contaminate the collected data.

I discussed some of the important issues surrounding customer’s reneging behavior in simulation.  The focus of the discussion has so far been on physical queues or waiting lines in which customers should be physically present to receive the service.  Some of the complex characteristics and thought processes associated with physical queues may or may not apply to virtual queuing systems such as call centers, web servers, e-commerce sites and the like.  One simplifying advantage of modeling human impatience in virtual queues is that much of the data needed to model reneging behavior may readily be collected electronically without facing the obstacles explained above. Virtual queuing systems have unique characteristics and deserve a  separate discussion.

Check -out more information about Dr. Moeeni.  If you would like more information about ProModel Simulation and simulation studies about queues check out the ProModel services industry information. Contact ProModel to learn more.