Performance Modeling of Representative Load Sharing Schemes for Clustered Servers in Multiaccess Edge Computing

Due to their limited functionality, ubiquitous connected devices in the Internet of Things rely heavily on the computational and storage resources of the cloud. However, mainstream cloud systems always require high network bandwidth and cannot satisfy the delay requirement of real-time applications. Therefore, a new paradigm called multiaccess edge computing has emerged to offload the computation and storage needs of end user devices to the edge cloud servers located in the radio access networks of 5G mobile networks. In this paper, we study and compare three load sharing schemes, namely, no sharing, random sharing, and least loaded sharing, which exploit the collaboration between clustered servers in different degrees. We develop computationally efficient analytical models to evaluate the performance of these schemes. These models are validated by simulation, and then used to compare the performances of the three load sharing schemes under various system parameters. Comparison results show that the least loaded sharing scheme is most suitable to fully exploit the collaboration between the servers and achieve load balance among them. It contributes to reducing the blocking probability and waiting time experienced by users.