| Keywords [eng] |
discrete event simulation, queue-based systems, AnyLogic, adaptive routing, capacity scaling, performance analysis, service-oriented systems, cost-efficiency, probability distribution, time in system. |
| Abstract [eng] |
This thesis investigates the performance analysis of queue-based service systems using discrete event simulation. The research focuses on modeling a banking service environment with ATM and teller service channels in the AnyLogic simulation platform. The initial model, developed in the earlier stages of the research, compared system behavior under normal and peak load conditions and demonstrated that increasing ATM capacity reduces queue congestion. However, this initial analysis was identified as too simple for meaningful performance evaluation. In the final stage, the model was extended to include multiple customer service types with probability-based assignment, service-type-based routing, adaptive routing logic that redirects flexible customers based on ATM congestion, lost-customer overflow logic, time-in-system measurement, and normalized cost-efficiency analysis. Four main simulation scenarios were designed and compared: baseline static routing, capacity scaling with two ATMs, and adaptive routing with two different threshold values. The results showed that capacity scaling produced the best overall performance in terms of served customers, lost customers, time in system, and cost per served customer. The adaptive routing experiments demonstrated that dynamic customer redirection is technically feasible but sensitive to threshold selection. A lower threshold caused excessive redirection and increased customer loss, while a higher threshold improved stability but did not outperform capacity scaling. A refined adaptive rule with teller-side congestion awareness reduced unnecessary redirection but had limited benefit due to high teller utilization. The time-in-system data was exported and analyzed using Python with the matplotlib and scipy libraries, producing probability density and cumulative probability curves for each scenario. An extreme arrival rate test confirmed that the system operates near its maximum throughput under default conditions. A planned resource unavailability scenario could not be implemented due to a limitation of the AnyLogic Personal Learning Edition and is suggested as future work. |
