Abstract
Managing agent populations in decentralized, dynamic environments like Mobile Ad Hoc Networks (MANETs) presents a significant challenge. This paper proposes a self-stabilizing method that treats computational tasks as "food" and agents as "consumers." By introducing a logistic growth "food ceiling" to task generation, the system achieves global stability through local rules. Theoretical analysis and simulations demonstrate that this approach maintains optimal population sizes with high precision, independent of network topology.
Introduction
In Multi-Agent Systems (MAS) deployed on Mobile Ad Hoc Networks (MANETs), population control is a critical "Goldilocks" problem. An insufficient number of agents leads to task backlogs, while an excessive population causes network congestion and drains limited battery resources. The decentralized and volatile nature of MANETs—where nodes frequently join, leave, or fail—precludes centralized management. This research explores a decentralized, biologically inspired solution to maintain a stable agent population.
The Synthetic Ecology Model
The proposed system models agents and tasks as a synthetic ecology. In this framework, jobs produced by network nodes are treated as "food points." Agents act as consumers that must "eat" food by completing tasks to maintain their execution thread.
As shown in the flowchart in Figure 1, the agent lifecycle is governed by food availability. If an agent accumulates sufficient food, it reproduces by spawning a replica. Conversely, if an agent fails to consume food within a specific timeframe, it terminates. This local behavior allows the global population to scale naturally with the workload.
Achieving Stability through Logistic Growth
A primary limitation of earlier ecological models was "neutral stability," where any environmental disturbance caused the population to oscillate indefinitely. To address this, the authors applied stability theory from control systems.
The key innovation is the introduction of a "food ceiling" or logistic growth model for task generation. By capping the maximum amount of food a node can hold, the system's dynamics are transformed. Jacobian matrix analysis proves that this modification ensures the system converges to a stable equilibrium point regardless of initial conditions.
The phase portraits in Figure 3 visualize this stability. While the original model (a) shows closed orbits (oscillations), the modified models (b-d) show trajectories spiraling toward a fixed point, representing a stable population size.
Experimental Evaluation and Results
The system was evaluated using a custom discrete event simulator across various network sizes (10 to 35 hosts) and topologies.
Precision and Recovery
In steady-state conditions, the system demonstrated remarkable precision, achieving a performance index between 94.4% and 99.5%. This indicates the population consistently stayed within ±1 agent of the theoretical target. Furthermore, the system showed rapid recovery; when the number of hosts was suddenly reduced (e.g., from 25 to 10), the agent population transitioned to the new target level quickly without long-term oscillations.
Topology Independence
A significant finding was that the equilibrium population is largely independent of network topology. Whether the network was configured as a linear chain or a fully connected graph, the average number of agents remained primarily defined by the ratio of food production ($g$) to consumption ($b$).
Conclusion
By treating decentralized task management as an ecological balancing act, this research provides a robust framework for agent population control. The introduction of a local resource ceiling successfully stabilizes the global system, making it resilient to the inherent volatility of MANETs. Future work will explore heterogeneous agent types and more complex "food chains" to handle specialized tasks and redundant agent cleanup.
Summary updated