Shark-Inspired Algorithm Tackles Edge-Cloud Task Offloading in Ultra-Dense IoT Networks

A new algorithm inspired by the hunting behavior of white sharks could change how billions of Internet of Things devices decide where to send their computing work. In a study published in Cluster Computing, Wei Li and Hongjian Li of Chongqing University of Posts and Telecommunications present an Improved White Shark Optimization (IWSO) algorithm designed to solve one of the most persistent problems in modern networked computing: how to offload computation tasks between edge servers and the cloud quickly, efficiently, and at massive scale. The research addresses a bottleneck that has grown acute as latency-sensitive applications, from augmented reality to industrial automation, place ever-heavier demands on mobile edge computing infrastructure.

The core challenge the researchers tackle is known as task offloading in Mobile Edge Computing, or MEC. When a smartphone, sensor, or connected machine generates a computational task, the system must decide whether to execute it locally on the device, on a nearby edge server, or remotely in the cloud. Each option involves a trade-off: local execution conserves network bandwidth but drains the device battery and may be too slow; edge execution reduces delay but competes with other users for limited resources; cloud execution offers raw power but introduces transmission latency. The combined cost, measured as a balance of execution delay and energy consumption, is what users ultimately experience, and minimizing it across thousands of simultaneous users is a formidable optimization problem.

Existing approaches to this problem have leaned heavily on metaheuristic algorithms, a family of search techniques that explore large solution spaces where exact methods would be computationally infeasible. Popular examples include particle swarm optimization, the grey wolf optimizer, and genetic algorithms. But as the authors note, these methods frequently suffer from premature convergence, meaning the search settles on a mediocre solution too early; poor diversity in the initial population, which limits the range of solutions explored; and an inability to scale effectively in highly dynamic, ultra-dense network environments where thousands of devices compete simultaneously. These weaknesses translate directly into higher costs and slower responses for end users.

To overcome these limitations, Li and Li built their framework on the White Shark Optimization algorithm, a bio-inspired metaheuristic introduced in 2022 that models the way great white sharks locate and capture prey. In the standard algorithm, candidate solutions move through the search space much like sharks navigating toward food, guided by positional updates that mimic the predators’ sensing and movement patterns. While effective for general global optimization problems, the original formulation still leaves room for improvement when applied to the fast-moving, multi-dimensional decision space of edge-cloud task offloading, where network conditions and user demands shift continuously.

The improved version integrates three key mechanisms, each targeting a specific weakness of the base algorithm. The first is CIRCLE chaotic mapping, which is used for population initialization. In metaheuristic search, the starting positions of candidate solutions matter enormously: a well-spread initial population covers more of the search space and reduces the risk of missing promising regions. Chaotic maps generate sequences that are deterministic yet pseudo-random and spread across the entire feasible range, giving the algorithm a more diverse starting set than conventional random initialization and helping it avoid clustering around unrepresentative points from the outset.

The second mechanism is an adaptive non-linear weight factor that dynamically balances global exploration and local exploitation. Exploration refers to the algorithm’s tendency to search broadly across the solution space, while exploitation refers to its focus on refining solutions in promising neighborhoods. Too much exploration wastes computational effort wandering aimlessly; too much exploitation causes the search to collapse prematurely around a local optimum. By adjusting the weighting non-linearly over the course of the run, the IWSO shifts emphasis from broad searching early on to fine-tuned refinement later, a transition that is crucial in the non-stationary environment of edge-cloud systems.

The third enhancement is Fitness-Distance Balance-based Cauchy reverse learning, a mechanism that helps the algorithm escape local optima. Opposition-based or reverse learning generates candidate solutions that are, in a defined sense, mirror images of existing ones, giving the search a second chance to find better regions it might otherwise never visit. By selecting the reference solution using the Fitness-Distance Balance criterion, which weighs both the quality of a solution and its distance from the best-known one, the method avoids biasing the search toward solutions that are merely close to the current best. The Cauchy mutation component introduces large, heavy-tailed jumps that can carry the search out of stagnant basins of attraction.

Extensive experimental simulations reported in the study demonstrate that the proposed approach significantly improves both convergence accuracy and convergence speed compared with existing White Shark Optimization variants and other standard metaheuristics. In practical terms, this means the algorithm reaches better offloading decisions in fewer iterations, an important property for edge systems where scheduling decisions must be made in near real time. The improved search translates directly into a substantially lower overall user-perceived cost, because the framework optimally balances execution delay against energy consumption when assigning tasks across the edge-cloud continuum.

Perhaps the most striking result concerns scalability. The IWSO exhibits exceptional robustness and optimization performance in ultra-dense network scenarios containing up to 1,000 User Equipments, successfully overcoming the bottlenecks that limit traditional scheduling schemes. This matters because the density of connected devices is rising relentlessly: smart factories, connected vehicles, dense urban sensor deployments, and massive IoT installations all push networks toward configurations where hundreds or thousands of devices must be scheduled simultaneously. An optimizer that degrades gracefully at that scale, rather than collapsing into premature convergence, is a meaningful step toward practical edge-cloud orchestration.

The work, published in Cluster Computing as volume 29, article number 749, arrives amid a broader wave of research into intelligent resource management for edge-assisted IoT applications, including hybrid bio-inspired optimizers, reinforcement learning-based offloading, and multi-objective scheduling frameworks. What distinguishes this contribution is the systematic combination of chaos-based initialization, adaptive weighting, and fitness-distance-guided reverse learning within a single joint offloading framework for an edge-cloud collaborative architecture. As 5G and future 6G networks multiply the number of edge nodes and the diversity of connected devices, algorithms of this kind, capable of balancing delay, energy, and scale simultaneously, are likely to become essential components of the software layer that keeps the responsive, energy-aware computing promised by edge-cloud integration within reach.

Subject of Research: Bio-inspired metaheuristic optimization for task offloading in mobile edge-cloud computing

Article Title: An improved white shark optimization algorithm for edge-cloud integrated computation offloading

Article References: An improved white shark optimization algorithm for edge-cloud integrated computation offloading. (n.d.). https://doi.org/10.1007/s10586-026-06541-8

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06541-8

Keywords: mobile edge computing, task offloading, white shark optimization, metaheuristics, edge-cloud collaboration, Internet of Things, chaotic mapping, fitness-distance balance, ultra-dense networks, energy efficiency, latency, Cluster Computing