Distribution-Aware Evaluation of LAN, WAN-IPsec, and SD-WAN Architectures for Real-Time Enterprise Applications

Abstract

Real-time enterprise applications increasingly depend on network architectures that can sustain predictable performance under dynamic traffic conditions. While Software-Defined Wide Area Networks (SD-WAN) are widely adopted to improve flexibility and resilience, their benefits relative to traditional WAN-IPsec overlays and local-area networks (LANs) remain insufficiently characterized from a distributional and experience-oriented perspective. This paper presents a controlled, comparative evaluation of three enterprise network architectures—LAN, WAN with IPsec overlay, and Hybrid SD-WAN—under varying traffic loads. The analysis combines Quality of Service (QoS) measurements, formal statistical validation, Quality of Experience (QoE) modeling, and machine learning–based prediction. Rather than focusing on average performance, the study emphasizes variability and tail behavior of delay and jitter, which are critical for real-time services. Experimental results show that LAN environments provide consistently stable performance across load conditions, whereas WAN-IPsec overlays exhibit pronounced delay and jitter tail expansion under congestion. Hybrid SD-WAN significantly mitigates this variability by reducing dispersion and high-percentile delay, even when average throughput gains are modest. Statistical analysis confirms significant architecture–load interaction effects for delay and jitter, while QoE evaluation demonstrates that stability, rather than throughput, dominates perceived service quality. Furthermore, non-linear machine learning models accurately predict QoE from observable network features, with jitter and packet loss emerging as the most influential predictors. These findings highlight the necessity of distribution-aware evaluation and experience-driven control for designing and operating real-time enterprise networks.