Comparing Actor-Critic and Neuroevolution Approaches for Traffic Offloading in FaaS-powered Edge Systems

In a computing context characterized by a complex and interconnected network of heterogeneous devices, which generate enormous amounts of data requiring exchange and near-real-time processing, the collaboration between Edge Computing and Function as a Service (FaaS) models holds significant potential to enhance the flexibility, cost-effectiveness, and responsiveness of applications. However, traditional FaaS encounters challenges in distributed edge environments due to dynamic traffic demands and resource limitations. Effective methodologies must be developed to address the load management issue, which involves determining the allocation of incoming requests to each node and deciding whether to process them locally, reject them, or offload them to neighboring nodes with available resources. This paper investigates and compares various approaches for managing incoming requests in a Decentralized FaaS environment. On the one hand, it considers Actor-Critic Reinforcement Learning algorithms, namely Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC). On the other hand, it examines the NeuroEvolution of Augmenting Topologies (NEAT) method. Experimental validation underscores the promising results of PPO, which ensures an average rejection rate of less than 4%.