PROJECT TITLE :

QoS Driven Task Offloading with Statistical Guarantee in Mobile Edge Computing

ABSTRACT:

Popular mobile applications, such as augmented reality, typically offload the work they need to do on their devices to resource-rich edge servers when using mobile edge computing. When a large number of mobile users compete for limited Communication and computation resources, the user experience can be adversely affected to a significant degree. Guaranteeing the Quality of Service (QoS) for the applications being offloaded is the primary technical challenge associated with task offloading. The current work on task offloading focuses on deterministic QoS (delay) guarantee, which means that tasks have to be finished before the specified deadline with one hundred percent accuracy. In spite of this, it is impractical to impose a deterministic QoS guarantee for tasks when offloading to edge servers because of the high levels of variability that are present in the wireless environment. In this paper, we focus on task offloading with statistical quality of service guarantee (tasks are allowed to complete before a given deadline with a probability above the given threshold). This type of offloading can save even more energy because there is no longer a quality of service requirement. In particular, we begin by putting forward a statistical computation model as well as a statistical transmission model in order to quantify the correlation between the statistical quality of service guarantee and task offloading strategy. After that, we turn the problem of task offloading into a mixed-integer, non-linear programming issue with a statistical delay constraint. We begin by transforming the statistical delay constraint into the constraints on the number of CPU cycles, and then we move on to the delay exponent. With the help of convex optimization theory and the Gibbs sampling method, we propose an algorithm that will provide the statistical quality of service guarantee for tasks. The findings of the experiment indicate that the proposed algorithm performs better than the three baselines.