Reinforcement Learning-based Dynamic Service Placement in Vehicular Networks

DRLD-SP: A Deep Reinforcement Learning-based Dynamic Service Placement in Edge-Enabled Internet of Vehicles

“…With this goal, we proposed an RL-based dynamic service placement approach in [16]. The work in [16] uses a classic model-free Q-learning algorithm that optimizes a certain objective such as minimizing resource usage or minimizing the delay.…”

“…With this goal, we proposed an RL-based dynamic service placement approach in [16]. The work in [16] uses a classic model-free Q-learning algorithm that optimizes a certain objective such as minimizing resource usage or minimizing the delay. In this work, we proposed a single objective function that minimizes the maximum of both edge resource usage and service delay, and controls the relative importance of resource usage vs. service delay by using a parameter 𝛼.…”

“…In our previous work in [16], we addressed the problem of service placement in vehicular networks using delay or edge resource usage as the objective function. We proposed a reinforcement learning-based (Q-learning) dynamic service placement framework to find the optimal placement of services at the edge servers while considering the vehicle's mobility and dynamics in the requests for different types of services.…”

“…We proposed a reinforcement learning-based (Q-learning) dynamic service placement framework to find the optimal placement of services at the edge servers while considering the vehicle's mobility and dynamics in the requests for different types of services. Our work in [16] doesn't consider the increasing demand from vehicles and only performs one-to-one placement (i.e placement of only one service at one edge node). In addition, it maintains a look-up table (Q-table) to keep a record of quality decisions.…”

“…In addition, it maintains a look-up table (Q-table) to keep a record of quality decisions. Different from [16], in this work, we propose deep reinforcement learning based on-demand DRLD-SP system of many-to-one placement where the number of service instances varies based on the requirement from vehicles. It has the ability to scale up or scale down the usage of resources at the edge with changing service demands.…”