Learning-powered migration of social digital twins at the network edge

• Published in: Computer communications Vol. 226-227; p. 107918
• Main Authors: Chukhno, Olga, Chukhno, Nadezhda, Araniti, Giuseppe, Campolo, Claudia, Iera, Antonio, Molinaro, Antonella
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.10.2024
• Subjects: Digital twin; Edge computing; Machine learning; Migration; Social Internet of Things; Social Internet of Things; Digital twin; Migration; Machine learning; Edge computing
• ISSN: 0140-3664; 1873-703X
• DOI: 10.1016/j.comcom.2024.07.019

Digital Twins (DTs), which are paired to Internet of Things (IoT) devices to represent them and augment their capabilities, are gaining ground as a promising technology to enable a wide variety of applications in the sixth-generation (6G) ecosystem, ranging from autonomous driving to extended reality and metaverse. In particular, “social” IoT (SIoT) devices, which are devices capable to establish social relationships with other devices, can be coupled with their virtual counterparts, i.e., social DTS (SDTs), to improve service discovery enabled by browsing the social network of friend devices. However, the mobility of SIoT devices (e.g., smartphones, wearables, vehicular on board units, etc.) may require frequent changes in the corresponding SDT placement in the edge domain to maintain a low latency between the physical device and its digital replica. Triggering SDT relocation at the right time is a critical task, because an incorrect choice could lead to either increased delays or a waste of network resources. This work proposes a learning-powered social-aware orchestration that predicts the mobility of SIoT devices to make more judicious migration decisions and efficiently move the paired SDTs accordingly, while ensuring the minimization of both intra-twin and inter-twin communication latencies. Different machine learning (ML) and deep learning (DL) algorithms are used for SIoT device mobility prediction and compared in terms of a wide set of meaningful metrics in order to identify the model that achieves the best trade-off between prediction accuracy and inference times under different scenarios. Simulation results showcase the improvements of the proposal in terms of reduced network overhead (by up to a factor of 3) and intra-twin and inter-twin communication latency (by up to 10%) compared to a more traditional solution, which activates the relocation of the DTs at fixed time intervals following periodic optimizations.