Abstract:
Offloading and resource utilization in vehicular networks and smart cities has been an
important problem due to the excessive load on the vehicles despite the availability of
multiple resources. These resources are located at the cloud or fog end of the network. Smart vehicles produce a generous number of tasks due to the multiple duties they
perform on the road. The tasks that are more CPU-intensive and require real-time results
quickly should not wait in the queue to be executed. An efficient offloading technique is
required for this purpose that can utilize the resources of the network efficiently while
ensuring task execution at a lower waiting time and higher efficiency. There are many
existing offloading techniques that have been implemented to solve this problem but none
of these techniques have attempted to solve the problem by making the system learn from
its behavior. Hence, in our proposed framework, we have introduced an intelligent
system of offloading that generated rewards based on certain parameters for each entity
included in the offloading decision. Multi-armed bandit is a deep-learning reinforcement
algorithm that is implemented on the fog federation. Fog nodes act as both the agent and
the arms of the bandit where rewards are assigned to each arm based on different
parameters in different variants of the algorithm. The task is offloaded to the highest
reward generating fog node after running the algorithm. We have also implemented the
network without the multi-armed bandit algorithm and compared the results of 6 variants
of the system. The aim of this research is to prove that offloading and resource utilization
can be improved if the system acts intelligently by learning from its past behavior and
using that knowledge to make efficient
