Beyond the Mean: Topology-Explicit and Logical-Dependency-Aware Tail Risk Analysis in Edge Collaborative Network

Edge computing applications are commonly executed through multi-node, multi-hop collaboration. In such systems, user experience is often dictated by rare but severe events such as tail-latency spikes and request blocking. However, mean centric metrics conceal the amplification effects of correlation and topology, leading to systematic underestimation of performance degradation in extreme scenes. We propose an egress side tail-risk analysis framework for edge collaborative networks that is topology-explicit, capturing both correlation and logic dependencies. It supports tail-calibrated provisioning in bursty and correlated scenes, where mean-centric metrics systematically underestimate end-to-end risk and required capacity. In the node view, we map correlated arrivals to the minimum capacity required for a tail constraint. In the network view, we propagate equivalent statistics along routing paths and logical dependencies, then define the System Blocking Indicator (SBI) to quantify end to-end tail risk and egress demand. In the design view, we invert SBI to characterize the budget–reliability tradeoff and compute the maximum reachable tail strength under an egress budget. Experiments on synthetic topologies and trace-driven workloads show that our method meets system-level tail constraints while providing more robust and interpretable capacity planning than mean-centric baselines.