Speaker: WANG Tan (Associate Professor,University of Science and Technology of China)

Date & Time: Monday, 14 September 2026, from 10:00 to 11:30 AM (Beijing Time)

Location: Tongji Building A308

ABSTRACT

Large-scale queueing networks are foundational to modern service and operational systems, yet conventional discrete-event simulation (DES) becomes computationally prohibitive as network size and complexity grow. In this talk, we present a fast simulation framework for large-scale queueing networks via Euler approximation and high-performance computing architectures. We develop discrete-time approximation schemes that support time-varying dynamics, Markovian settings, and complex non-Markovian features (such as non-stationary arrivals and general service time distributions). Theoretically, we establish the stochastic bounding properties and first-order accuracy of the proposed schemes, proving that approximation errors remain bounded over time and exhibit diminishing asymptotic relative error as the system scales up. Drawing inspiration from modern AI model parallelization, we further propose a layer-wise parallel computing architecture combined with vectorized computation. Numerical experiments demonstrate that our framework achieves orders-of-magnitude speedups over DES while maintaining high accuracy. The proposed framework holds significant promise for simulating complex, large-scale discrete systems in practice.