BEYOND STATIC TRAFFIC English narrated research explainer Illustrative animation, not experimental footage. 00:00:00 Beyond Static Traffic Welcome to Beyond Static Traffic, a dynamic and phase-aware framework for software-defined networking research. A network can slow down or change its behavior without an attack taking place. This project asks how controlled experiments can turn those changes into traceable observations for intrusion detection dataset construction. 00:00:22 Inside the SDN testbed The framework combines Mininet, Open vSwitch, and the Ryu controller, communicating through OpenFlow version one point three. Mininet creates virtual hosts and links. Open vSwitch forwards packets, while Ryu provides programmable control. Line and star topologies let the experiments explore different network structures inside an isolated testbed. 00:00:47 Three operational phases Each experiment moves through stable, constrained, and recovery phases. Stable conditions provide a reference. The constrained phase applies controlled changes to network settings. Recovery restores the reference configuration. Verified phase events preserve the timing context, so a change in network conditions is not automatically labeled as malicious activity. 00:01:13 Start with legitimate traffic The framework describes six controlled workloads. Two represent legitimate activity. B zero zero generates continuous U D P and I C M P communication. B zero one produces authorized U D P bursts. These benign bursts are important: increased traffic intensity alone does not establish a malicious attack. 00:01:37 Bounded TCP probing A zero one introduces bounded T C P reconnaissance across a limited port range. The animation illustrates probes moving toward several service ports within the controlled testbed. Generator-side logs provide evidence that traffic was generated. They do not independently establish attack success or malicious intent. 00:01:59 Controlled UDP increase A zero two creates a controlled increase in U D P traffic. Here, the denser packet animation represents the workload concept, not recorded experimental packet timing. The dataset must keep workload execution evidence separate from network constraints, and from any claim that an attack successfully disrupted a service. 00:02:22 TCP connection churn A zero three repeatedly establishes and tears down T C P connections. Requests and responses move in both directions as connections open and close. The paper reports endpoint-level verification for this scenario, while independent malicious ground-truth adjudication remains incomplete. 00:02:43 Low-rate service probing A zero four sends low-rate T C P probes toward a controlled service. Unlike a large traffic burst, this workload produces sparse activity. Endpoint-level verification adds execution evidence, but synthetic probing labels still do not demonstrate detection of independently verified real-world or previously unseen malicious attacks. 00:03:06 Bring the evidence together The framework collects OpenFlow flow statistics, packet and byte counters, round-trip-time samples, U D P measurements, and timestamped experiment events. These sources are aligned into observation windows. Run identifiers preserve the connection to each experiment, and missing measurements remain missing rather than being treated as zero traffic. 00:03:31 Label the context carefully Phase labels describe verified network conditions. Activity labels describe recorded workload intervals. A window that crosses a phase boundary is marked as a transition. Mixed activity and partial observations retain their own context. This prevents operational changes or incomplete telemetry from being mistaken for evidence of malicious behavior. 00:03:56 The H14 dataset The H fourteen dataset contains two hundred and sixteen observations across twenty columns. Training, validation, and test each contain seventy-two observations from eight distinct runs, with no run identifiers shared between partitions. The supplied export contains four scenario values. That dataset coverage is narrower than the framework's six described workloads. 00:04:22 Prospective model evaluation H twenty-three evaluates five frozen models on one hundred and forty-four intervals from sixteen additional runs, without retraining or reselection. Gradient Boosting achieved eighty-nine point five eight percent balanced accuracy. Random Forest had the highest recall, at eighty-seven point five percent. These are synthetic probing results, and statistical significance between models was not established. 00:04:50 A measurable network response In H twenty-eight, twelve runs produced thirty-six phase-specific network conditions. Mean round-trip time rose from about six point zero five milliseconds to thirty-two point eight eight under constraint. T C P bulk goodput fell from about nine point three to one point nine megabits per second. Recovery measurements returned close to baseline, demonstrating a measurable response to the configured network phases. 00:05:18 What the evidence supports The study supports controlled S D N traffic generation, traceable dataset construction, synthetic-workload evaluation, and measured network-fidelity changes. It does not yet establish independently verified malicious attack detection or generalization to unseen attack families. Explore the website's actual observations, original graphs, and downloadable evidence packages to trace each result back to its source.