The Rich Get Richer: Preferential Attachment in the Task Allocation of Cooperative Networked Multiagent Systems With Resource Caching

In networked multiagent systems (NMASs) with resource caching, resource replicas are cached in favor of the agents who accessed such resources most recently and frequently. Task execution in NMASs is described through agents' operations when accessing necessary resources distributed in the networks, and thus, agents with richer experiences executing tasks will have higher access to resources. To optimize tasks' resource access time, we investigate two types of preferential attachments in the task allocation of NMASs with resource caching: history and present preferential attachments, in which an agent has higher access to a resource if that agent has richer history (or present) accessing experiences for that resource. Therefore, agents that were (or are) heavily burdened by tasks may have certain preferential rights to new tasks in the future. Our experiments found that preferential attachment in task allocation can effectively reduce tasks' execution time, particularly when the network context is considered and the number of tasks is high. In addition, we discovered two interesting phenomena: (1) Compromise between preferential attachment and load balancing can achieve better performance than single preferential attachment when there are too many tasks waiting, and (2) the integration of history and present preferential attachments can outperform either history or present preferential attachment alone in task allocation.