Estimating Bandwidth and Analyzing Worst-Case Delay Bounds Using Network Calculus

This paper proposes a bandwidth estimation framework for Network-on-Chip systems using network calculus, addressing the challenges due to nonlinear behavior and dynamic traffic. However, most existing approaches struggle to predict bandwidth effectively when network traffic packets are affected by possible nonlinearity in the network infrastructure. Our approach models available bandwidth as a service curve and employs min-plus algebra-specifically, convolution and deconvolution operations-to derive end-to-end predictions from individual link characteristics. We conduct a worst-case delay analysis under varying traffic loads and channel rates, demonstrating the robustness of our approach across heterogeneous Network-on-chip platforms. Unlike conventional methods that are constrained by buffer sizes or flow serialization issues, our framework integrates passive measurement techniques and probabilistic method to capture bandwidth variations more accurately. The proposed solution enhances prediction fidelity in systems with bursty and synthetic non-uniform traffic, offering a scalable and analytically grounded alternative to existing bandwidth estimation and worst-case delay bound techniques.