Joint Job Partitioning and Collaborative Computation Offloading in Multi-User Wireless Networks

Computation offloading to nearby devices offers resources-constrained mobile devices the opportunity to reduce local computation load and energy consumption when supporting applications with intensive computation load. In this paper, we consider joint job partitioning and collaborative computation offloading in a multi-user wireless network, aiming at minimizing the energy consumption of the overall network under the completion time constraint of each application. The problem is formulated as a binary integer linear programming problem and then transformed into a weighted bipartite graph matching problem. The Kuhn-Munkres algorithm is firstly adopted to obtain the optimal partition and scheduling. Three distributed scheduling algorithms are then introduced. The first one is referred to as stable matching, where each mobile node chooses matching collaborator or requester based on its own interests. Next an asynchronous greedy algorithm is proposed, where every requester-collaborator pair cooperatively reduces their total energy consumption. Finally, based on the greedy algorithm, a modified version, referred to as Maximum Differential Energy Matching (MDEM) algorithm is devised, which minimizes the overall network energy cost by relaxing the stability criterion in order to benefit energy consumption of all nodes. Simulation results show that the MDEM algorithm achieves near-optimal performance in terms of total energy consumption for the network.