Many researchers focus on offloading issues and challenges to improve energy efficiency and reduce application response time by employing multi-objective offloading frameworks but without considering offloading failures. Edge Computing, due to distributed architecture that contains diverse resource and reliability characteristics, is prone to server and network failures that can postpone or prevent offloading thus affecting the overall system performance. In this study, we propose a novel solution to model the energy consumption of mobile device and application response time assuming the resource and reliability diversity of the Edge Computing system. The model adopts the Markov Decision Process (MDP), which provides a formal framework for capturing stochastic and non-deterministic behavior of Edge offloading. We propose the Energy Efficient and Failure Predictive Edge Offloading (EFPO) framework based on a model checking solution called Value Iteration Algorithm (VIA). EFPO determines the feasible offloading decision policy, which should yield a near-optimal system performance. Evaluation is performed by offloading various mobile applications modeled as Directed Acyclic Graphs (DAG). Failures are emulated from the failure trace data set from Pacific Northwest National Laboratory. Results show that the proposed EFPO framework yields better time performance between 12% - 57% and better energy efficiency between 15% - 51% when comparing to other offloading decision policies from the literature.

Top