Effects of network delays on swarming in a multi-agent security system

In testing our multi-agent cyber security system, Ant-Based Cyber Defense (ABCD) we have found that pheromone efficacy, swarm formation, and cyber-security properties of a multi-agent system can be affected by non-uniform delay times experienced by the ant agents as they traverse on large-scale, real-world networks. This led to tests where we artificially induced delay on large sections of a network to experiment with swarms under different delay conditions. As predicted by queuing theory, we found that more agents gathered in grid sections with longer inter-node propagation/transmission delays and that the population of agents in the sections with longer delays was proportional to the delay times in that network section relative to the rest of the network. We verified the observed operational behavior via both simulation and queuing theoretic analysis and found that there may be no closed-form solution to the steady state of a system without static transition probabilities. In ABCD, transition probabilities are dynamic, arising from non-deterministic pheromone deposition rates. Without careful tuning of multi-agent security systems to a given network's delay characteristics the security provided may be significantly less than desired because agents may not be distributed as needed or required by the designers. We have used the findings in this study to discover parameters to help make this tuning possible.