Computation Offloading via Sinkhorn’s Matrix Scaling for Edge Services

Abstract

Mobile-edge computing (or MEC) has aroused a wide attention at the 5th generation mobile networks (5G) communication era. Different edge servers (such as cloudlets, micro data centers, and base stations) have been proposed to support the MEC architecture paradigm. For the diverse loading capacities of different edge servers, computation offloading for the edge services loaded in neighboring edge servers is desired for assuring the overall serve performance as well as the Quality of Experience of users for MEC applications. To dynamically balance the computation load for neighboring edge servers, we model the optimal edge services computation offloading destination determination issue as an optimal transport distances problem in this article. We propose computation offloading via Sinkhorn's matrix scaling (COSIMS) to determine the optimal offloading destination. Experimental evaluations conducted on real-world edge computing data sets indicate that COSIMS can guarantee that the neighboring edge servers cooperatively provide effective services to mobile users with least extra communication hops.