<![CDATA[Software-Defined Networking (SDN) separates the forwarding layer from a programmable control layer, enabling traffic management from a single logical control point. However, when the controller forwards along purely topological shortest paths, it cannot exploit the path diversity of multi-rooted data-centre fabrics, so flows are concentrated on a subset of links while equally short alternatives stay idle. This work introduces an Adaptive Path Routing (APR) application for the Ryu controller. At each monitoring tick APR queries the OpenFlow statistics interface and selects end-to-end paths that minimize a normalized composite cost combining residual bandwidth, one-way delay, and loss. APR and the default shortest-path-first (SPF) baseline were implemented on the same controller and evaluated in Mininet on a k=4 fat-tree (20 switches, 16 hosts; links shaped to 10 Mbps and 2 ms) across four workloads, each repeated ten times. Against SPF, APR more than doubles aggregate TCP goodput under light load (9.28 to 18.93 Mbps, +104.1%, p<0.01) by spreading flows across the four core switches through load-aware path placement, while under saturation it matches the baseline and, owing to a 30% hysteresis threshold, never increases jitter or packet loss. These results show that live-metric adaptive path selection yields large throughput gains where path diversity can be exploited and behaves as a safe drop-in replacement for shortest-path forwarding otherwise.]]>
Copyrights © 2026
