Multidimensional auction for task allocation using computation offloading in fifth generation networks

Abstract

Smartphones, Wireless Sensor Networks (WSNs) and Internet of Things (IoT) suffer from battery lifetime and computing performance. Computation offloading is a significant solution to overcome these issues. In this paper, a distributed framework for offloading a task of an allocator among nearby devices in a heterogeneous network is proposed. Specifically, the allocator holds a multidimensional auction among the nearby devices for allocating its task based on speeds and costs of the nearby devices to minimize the task execution time. The proposed framework is beneficial for all the participants as the task execution time of the allocator is minimized, and the allocator offers its resources for any selected devices at any time by the same used amount. The proposed framework is simulated using real values for the allocator task size and the speeds of the nearby devices taking into account the differences between different network scenarios (i.e. 4G, Wi-Fi and ZigBee). Simulation results demonstrate that the proposed framework provides a 352x speedup over the local execution, which demonstrates the enormous reduction in the task execution time of the allocator. The proposed framework performs very well in heterogeneous networks, which is an objective of the fifth generation (5G) networks. In real scenarios, nearby devices or allocator may leave the cell or turn off before completing the task execution. To deal with devices mobility, a mobility approach is introduced, which holds the auction multiple times rather than holding it only once. Fixed and dynamic period approaches are used to determine the period of auctions. To maintain stability, a stability model is introduced to deal with devices mobility and turnoff in which the allocator offloads its task to stationary devices with the highest battery level. Stability model improves speedup by a factor of 5.5 with respect to game theory framework, whereas, the proposed model improves speedup by a factor of 1.5 with respect to stability model.