Evaluating and simulating end‐to‐end latency in 5G environments

This thesis presents a model for estimating worst-case end-to-end latency in 5G vehicular offloading scenarios using Deterministic Network Calculus (DNC). By decomposing the 5G network into functional layers and service components in the Radio Access Network (RAN) and in the Core Network, the work defines arrival and service curves that model per-layer delay contributions under deployment-specific conditions. Three representative scenarios are analyzed: (1) MEC-based edge offloading near the Core edge (MEC@CN), (2) centralized offloading via a long-haul core and Internet path, and (3) ultra-low latency local breakout through a UPF at the gNB site (MEC@gNB). Each scenario is instantiated with realistic parameter values and compared against empirical and simulated results reported in existing literature. The model captures how RAN dominates the delay budget in most cases and consistently identifies the RLC layer as the bottleneck due to its segmentation and reordering behavior. Despite its conservative nature, the DNC model provides valuable insights into architectural trade-offs and delay bounds for safety-critical vehicular applications. Limitations, including the difficulty of modeling mobility-triggered handovers and dynamic traffic patterns, are discussed. The work concludes with a critical discussion of the model's relevance, limitations, and opportunities for future extension toward hybrid, stochastic, and empirically tuned frameworks.