3rd TAGRA

Multiservice software-defined networks (SDNs) must deliver multimedia traffic with different service requirements while maintaining acceptable delays and efficient use of limited channel capacity. This study addresses the challenge of distributing traffic flows when transmission patterns are variable and self-similar, making timely delivery difficult to assess. We develop a stochastic mathematical model of a converged SDN to evaluate probabilistic and temporal performance characteristics and formulate flow distribution in terms of network efficiency and the probability of timely message delivery. The model accounts for channel capacity and traffic self-similarity. To support intelligent traffic analysis, we classify traffic flows using the Random Forest algorithm implemented in Python's scikit-learn library. The resulting traffic classes are considered alongside the model's performance indicators to inform routing decisions. The analysis establishes a framework for relating traffic characteristics and channel constraints to the evaluation of delivery timeliness and flow allocation. It also shows how machine-learning-based traffic classification can be incorporated into the construction of SDN routing tables and network-layer packet management. These findings provide a methodological basis for jointly considering statistical traffic behavior, classification, and routing optimization rather than treating them as separate tasks. The proposed approach can support adaptive management of multiservice telecommunication networks and the delivery of multimedia services under changing traffic conditions. Further experiments are needed to quantify improvements in delay, delivery probability, and resource utilization across a range of traffic loads and service classes.