Online Function Scheduling for Dual-Heterogeneous Serverless Vehicular Edge Computing

Vehicular service providers bear a heavy burden of scalability and service load balancing problems in traditional edge computing systems. By abstracting the service computing granularity to a higher function level, the nascent serverless edge computing is envisioned to leave the burdensome scheduling and management issues to computing vendors. Nevertheless, the traditional terrestrial-based edge system's features, such as heterogeneity, scattered deployments, and environment dynamics, significantly complicate the function scheduling process for computing vendors. Specifically, inhomogeneous load distribution caused by these features leads to edge servers' concurrent performance bottlenecks and low resource efficiency issues. Further, it is crucial to develop an efficient scheduling scheme for the serverless vehicular edge computing network with heterogeneous function requirements/edge configurations and without prior environment evolution knowledge. To this end, we propose a dual-heterogeneous serverless vehicular edge computing network to facilitate the function scheduling within two-scale ranges. We formulate the joint function scheduling problem considering the closely coupled request dispatching, function placement, and computation resource allocation as a cooperative partially observable Markov game. Then, a model-free multi-agent reinforcement learning is introduced to acquire an online scheduling strategy. Simulation experiments show that the proposed approach can guarantee more completed function requests than other baselines.