DEPU framework accelerates game-theoretic queue analysis by 10x
No closed-form formulas needed—researchers merge evolutionary dynamics with discrete event simulation for queueing systems.
A new paper from Vincent Knight and Geraint Palmer-Liyu (arXiv:2606.28100) presents DEPU (Discrete Event Population Updates), a framework that unlocks evolutionary game theory analysis for queueing systems that lack closed-form payoff expressions. Traditional behavioral queueing research requires explicit cost or utility formulas, severely limiting the models that can be studied. DEPU removes this constraint by coupling a single long discrete event simulation (DES) run directly to a population update rule—either the replicator dynamics (DERD) or a Moran process (DEMR) for finite populations.
The authors demonstrate DEPU on a multi-server jockeying model (where customers can switch queues), a system for which no analytic fitness function exists. Compared to the standard approach of nesting many short simulations inside an evolutionary loop, DEPU achieves comparable precision tens of times faster. Because each operating point requires only one simulation run, systematic parameter sweeps become computationally feasible. This opens the door to analyzing strategic emergent behavior (e.g., herding, load balancing) in realistic queueing networks that were previously intractable, with applications in call centers, cloud computing, and traffic management.
- DEPU couples a single long discrete event simulation directly to evolutionary update rules, removing the need for closed-form payoff expressions.
- Two implementations: DERD for replicator dynamics (continuous population) and DEMR for Moran-style copying events (finite population).
- On a multi-server jockeying model, DEPU reaches comparable precision tens of times faster than nested simulation, enabling large parameter sweeps.
Why It Matters
Makes evolutionary game theory practical for any simulatable queueing system, accelerating research on strategic behavior in real-world networks.