Software Defined Wireless Architecture and Drones Accelerate Edge Network Data Migration

Computer systems researchers have designed and demonstrated a dual-band software-defined wireless networking framework coordinated with mobile unmanned aerial vehicles to accelerate data migration across distributed edge in-network storage systems. As continuous sensor streams, high-resolution cameras, and mobile devices generate massive data volumes across urban cellular environments and temporary field deployments, conventional centralized cloud offloading encounters severe bandwidth constraints, prolonged transmission delays, and heightened data vulnerability. Although edge storage servers mitigate latency by processing and holding records locally, conventional fixed servers create severe forwarding bottlenecks, accelerate battery depletion in neighboring relay devices, and lack physical adaptability during network failures or post-disaster rescue scenarios. To overcome these infrastructural boundaries, modern network architectures increasingly incorporate in-network storage, integrating processing, data relaying, and persistent caching directly within intermediate switches. However, because internal network nodes possess strictly finite local storage capacities, accumulated data must be migrated rapidly to release local capacity without triggering transmission bottlenecks.

Addressing the inherent rigidity of fixed migration repositories, the investigative team introduced an agile network architecture wherein multiple autonomous unmanned aerial vehicles serve as mobile migration destinations rather than relying on stationary ground storage infrastructure. In conventional aerial data collection paradigms, an individual drone typically follows a preplanned sequential trajectory, approaching distributed sensor locations one by one to download stored telemetry onboard. This sequential traversal imposes enormous flight-time overhead, demands prolonged battery endurance, necessitates precise localized hovering, and requires solving an NP-hard traveling salesman trajectory problem that restricts mission duration. In contrast, the researchers formulated an alternative single-point ground access scheme wherein unmanned aerial vehicles dynamically select designated ground access nodes based on prevailing channel states. Stored files distributed across decentralized network nodes are aggregated along multi-hop ground links and transferred to the designated aerial vehicles through local wireless access points, balancing transmission loads without requiring sequential physical flyovers.

To coordinate the distributed network components effectively, the architecture decouples control functionality from forwarding pipelines through a software-defined wireless networking paradigm supervised by a centralized controller. Traditional distributed wireless topologies tightly couple control management with localized packet forwarding, preventing individual switches from acquiring comprehensive visibility into dynamic link qualities, instantaneous node storage loads, and aerial coordinates. Under the demonstrated design, the controller executes four integrated management modules: a topology awareness module that periodically tracks switch status and link adjacencies via OpenFlow requests and link discovery mechanisms; a network measurement engine monitoring queuing delays, loss rates, and port statistics; an instruction module that distributes flow tables and migration commands via standard communication protocols; and a dedicated vehicle coordination module. This aerial management module guides drones toward target access sectors, verifies wireless associations, and coordinates file transfers.

A critical engineering feature of the platform is an out-of-band communication framework designed to prevent link loops and eliminate broadcast storms that frequently disrupt in-band wireless software-defined networks. The team divided wireless transmissions across distinct spectral bands, dedicating the 2.4 gigahertz frequency band to out-of-band control signaling and reserving the 5 gigahertz frequency band for high-throughput data payload routing. Within the data plane, switches connect to the primary router through 2.4 gigahertz relay bridging, automatically instantiating wireless wide-area network interfaces for uplink control traffic while bridging local area connections across unified subnets. The researchers leveraged the greater physical penetration and coverage of the 2.4 gigahertz spectrum to guarantee stable transport for critical control signaling, whereas the broader 5 gigahertz spectrum supported an IEEE 802.11s wireless mesh topology capable of handling intensive multi-hop data migration flows without direct channel contention.

At the mathematical core of the system, the researchers formulated the parallel data migration task as an NP-hard combinatorial optimization problem characterized by many-to-many communication routing and multi-variable coupling. When multiple decentralized source nodes transmit files concurrently toward several mobile aerial carriers, competing traffic flows inevitably contest shared bottleneck links, causing severe queuing delays and packet drops. The mathematical model explicitly allocates residual link bandwidth in proportion to respective file sizes, ensuring that larger files receive proportional bandwidth allocations across contested paths. Because the overall migration mission terminates only after the final distributed file reaches its intended destination, the objective function seeks to minimize the maximum completion time across all concurrent transfers, accounting for bottleneck link capacities, cumulative path latencies, and multiplicative link packet loss probabilities across dynamically selected multi-hop paths.

To resolve the severe computational complexity arising from the simultaneous coupling of aerial destination choices, ground access node selections, and end-to-end transmission routes, the authors developed a two-stage algorithmic decomposition. The first optimization stage determines destination vehicle assignments and ground access node selections by applying an enhanced spectral clustering and k-means partitioning technique. To characterize inter-node communication quality realistically under mutual multi-flow interference, the algorithm constructs an average comprehensive network state score across the shortest candidate paths between every node pair, assigning normalized weights of 0.7 to residual bandwidth, 0.2 to latency utility, and 0.1 to successful transmission rates. Node pairs exceeding a quality threshold of 0.25 establish virtual connectivity edges within a reconstructed similarity graph. The system then computes a normalized graph Laplacian, derives low-dimensional spectral embedding vectors from its smallest eigenvalues, and executes a customized k-means clustering process enhanced with simulated annealing to partition the ground nodes into balanced spatial operational regions.

The second optimization stage resolves the many-to-many routing challenge by implementing an improved whale optimization algorithm, an adaptive swarm intelligence method that establishes optimal forwarding paths for all active data files. Classical heuristic routing approaches such as dynamic programming approximations or minimum-cost forest algorithms rely primarily on static topological assumptions and frequently become trapped in local performance optima. To overcome these limitations, the authors integrated a piecewise linear chaotic map with a control parameter of 0.2 to generate uniformly distributed initial candidate path combinations. The algorithm continuously alternates between bubble-net spiral encircling and wide-area exploratory search behaviors, guided by a nonlinear trigonometric convergence factor and a sigmoid-based adaptive inertia weight designed to balance global exploration and fine localized exploitation across successive iterations.

The experimental validation was conducted on a fully functional hardware testbed comprising an N5105 soft router functioning as the centralized controller, ten storage-forwarding integrated switches, ten standard forwarding switches, and two autonomous quadrotor aerial vehicles. The controller hardware incorporated an Intel quad-core processor and dual-channel memory running OpenFlow management protocols. The storage-forwarding switches utilized dual-band wireless routers equipped with 1.3 gigahertz main controllers, one gigabyte of memory, eight gigabytes of embedded flash memory, and one terabyte solid-state drives mounted to support in-network persistence. Standard forwarding nodes utilized dual-band routers operating OpenWrt firmware with virtual Open vSwitch bridging, while custom-built quadrotor drones carried integrated wireless routing modules connected directly to flight microcontrollers over serial interfaces for automated flight maneuvering.

Systematic empirical evaluations were conducted across ten-, fourteen-, and twenty-node mesh topologies, evaluating migration batches ranging between 200 and 400 megabytes per active storage node over ten repeated experimental trials. Parameter sensitivity investigations revealed that evaluating four candidate paths during the average comprehensive network state calculation produced significantly shorter and more consistent migration completion durations than single-path formulations, confirming that multi-path metrics effectively capture dynamic link contention. Furthermore, calibrating the clustering load-balancing trade-off parameter to a moderate value of 0.5 achieved optimal network partitioning; lower values concentrated excessive data loads onto single aerial targets and caused local link saturation, whereas higher values dispersed related nodes across divergent regions and degraded transmission path quality.

Comparative benchmarking demonstrated that the proposed two-stage optimization architecture consistently outperformed baseline configurations and alternative heuristic algorithms across every evaluated network scale. Ablation trials comparing the framework against shortest-hop forwarding and clustering without swarm path optimization confirmed that both regional spectral partitioning and swarm path selection provide indispensable latency reductions. In direct comparisons with random assignment, single-stage global whale optimization, and adaptive hyper-heuristic search algorithms, the two-stage model maintained substantially lower and more consistent maximum migration completion times as network scale increased. The researchers noted that future investigations will focus on coordinating multiple software-defined controllers and explicitly incorporating drone battery depletion rates into path planning to sustain large-scale, long-duration operational deployments.

Subject of Research: Edge in-network storage data migration using software-defined wireless networking and cooperative unmanned aerial vehicles

Article Title: A data migration method for edge in-network storage leveraging software-defined wireless networking and multi-UAV coordination

Article References: Ye, M., Huang, G., Jiang, Q., Huang, Y., Wang, Y., & Liang, S. (2026). A data migration method for edge in-network storage leveraging software-defined wireless networking and multi-UAV coordination. Discover Computing, 29(1), Article 885. https://doi.org/10.1007/s10791-026-10847-x

Image Credits: AI Generated

DOI: 10.1007/s10791-026-10847-x

Keywords: edge computing, in-network storage, data migration, software-defined wireless networking, unmanned aerial vehicles, spectral clustering, whale optimization algorithm, k-means clustering, wireless mesh networks, out-of-band management, many-to-many routing, network load balancing, OpenFlow, link bandwidth allocation, combinatorial optimization, quadrotor drones