Graph-neural-network-based delay estimation for communication networks with heterogeneous scheduling policies

Modeling communication networks to predict performance such as delay and jitter is important for evaluating optimizing them. In recent years, neural have been used do this, which may advantages over existing models, example from queueing theory. One of these RouteNet, based on graph networks. However, it simplified assumptions. key simplification the restriction a single scheduling policy, describes how packets different flows are prioritized transmission. this paper we propose solution that supports multiple policies (Strict Priority, Deficit Round Robin, Weighted Fair Queueing) can handle mixed in network. Our RouteNet architecture part "Graph Neural Network Challenge". We achieved mean absolute percentage error under 1% with our extended model evaluation data set challenge. This takes neural-network-based estimation one step closer practical use.