Distributed Algorithm for Energy Efficient Joint Cloud and Edge Computing with Splittable Tasks

The considered hierarchical multi-level offloading scenario consists of multiple mobiles units (MUs), an access point (AP) with attached cloudlet for mobile edge computing (MEC) and a cloud server. Each user has an arbitrarily splittable task and three possible options for the computation of fractions of this task, which are local computation, offloading to the cloudlet and offloading to the cloud server. We decompose a non-linear central energy minimization problem into subproblems and propose a distributed algorithm that separates the allocation of shared communication and computation resources by the AP from the offloading decisions by the MUs. The AP assigns fractions of the shared bandwidth of the radio access channel, the shared backhaul transmission link to the cloud server and the shared computation frequency at the cloudlet according to offloading decisions of the MUs by solving closed-form expressions which are derived in this paper. Given the available resources, each MU solves a linear optimization problem to calculate the optimal fractions of its task to be computed locally or offloaded. In numerical simulations, the algorithm is proven to be stable and reaching results close to the optimal policy.