Cloudlet deployment and task offloading in mobile edge computing using variable-length whale and differential evolution optimization and analytical hierarchical process for decision-making

Mobile edge computing (MEC) is a well-known technique to support delay-sensitive applications at the edge of mobile networks. MEC has shown its potential in real-world computation but is still not fully mature. MEC's main feature is pushing computing resources to the network edges. In MEC environment, cloudlets that represent a relatively powerful computing resource can be collocated with the base station to enable good coverage of computing service due to the high demand and random distribution of users. The problem of Cloudlet Deployment and Task Offloading (CDTO) involves deploying a set of cloudlets in an environment and assigning user tasks to optimize various metrics, including energy consumption, quality of service (QoS) and cost. Typically, approaches deal with them separately, which might cause sub-optimality. Furthermore, assuming the fixed location of the cloudlets will limit the dynamic adaptability of the problem. Enabling more optimality and adaptability to the dynamic nature of CDTO, we propose a novel Variable-Length multi-objective Whale optimization Integrated with Differential Evolution designated as VL-WIDE for joint cloudlet deployment and tasks offloading. Unlike the existing optimization algorithm, VL-WIDE features the capability of searching different lengths of solutions to cover the variable number of cloudlets for deployment. It provides an application-oriented solutions repair operator for repairing non-valid solutions and assuring that all solutions are generated in the feasible region. Furthermore, it enables non-dominated evaluation of solutions based on four objectives using crowding distance for selection. The proposed algorithm with its variable length solution encoding enables moving the cloudlets among pre-defined locations, adding or removing them in order to increase the quality of service according to the change in the user density caused by user mobility. VL-WIDE was also integrated with the solution selection model based on the Analytical Hierarchical Process (AHP) that considers decision-maker preference for the optimized objectives. Comparing this developed algorithm with other algorithms shows its superiority in multi-objective optimization (MOO) evaluation metrics. VL-WIDE has accomplished a higher median value for the domination over state-of-the-art algorithms with a higher number of non- dominated solutions value than all other benchmarks. Three hundred scenarios involving various parameters related to base stations, cloudlets, users, and wireless communications were generated. Additionally, a simulator is used to evaluate the proposed methodology under different deployment scenarios and network conditions. The simulator provides a realistic environment to test the system, and the results are compared with the benchmarks. The improvement percentage in terms of hyper-volume, delta-metric, and the number of non-dominated solutions are (8%), (5%), and (6%), respectively, over the baseline approach. Furthermore, the AHP VL-WIDE solutions were more fulfilling to the desire of the decision-maker compared with other algorithm.

Text FK 2023 1 IR.pdf Download (3MB)

Actions (login required)

View Item