Split Computing: Dynamic Partitioning and Reliable Communications in IoT-Edge for 6G Vision

Implementation of Deep Neural Networks (DNNs) in 6G era is expected to get widespread attention in the applications of Internet of Things (IoT). Unfortunately, it is a challenging task to run DNN models in IoT devices due to their limited computation capability. Further, remotely deployed cloud is i...

Full description

Saved in:

Bibliographic Details
Published in:	2021 8th International Conference on Future Internet of Things and Cloud (FiCloud) pp. 233 - 240
Main Authors:	Karjee, Jyotirmoy, Anand, Kartik, Bhargav, Vanamala Narasimha, Naik, Praveen S, Dabbiru, Ramesh Babu Venkat, Srinidhi, N
Format:	Conference Proceeding
Language:	English
Published:	IEEE 01-08-2021
Subjects:	6G mobile communication Bluetooth Cloud computing Computational modeling Edge Computing Internet of Things Reliability Split Computing Switches
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Implementation of Deep Neural Networks (DNNs) in 6G era is expected to get widespread attention in the applications of Internet of Things (IoT). Unfortunately, it is a challenging task to run DNN models in IoT devices due to their limited computation capability. Further, remotely deployed cloud is incompatible to support DNN inferences in IoT platform due to its latency constraints and unreliable connectivity during poor network conditions. To address the problems, we deploy edge devices to the close proximity of IoT devices and introduce the concept of "Split Computing" to execute the DNN inference task among IoT-edge devices. In the context of split computing, we propose two mechanisms that can reduce both computational and communicational overhead by finding a trade-off between them given as follows: (1) Dynamic Split Computation (DSC) mechanism: selects an optimal partition of DNN inference between IoT device and edge to reduce computation latency and computational resources. (2) Reliable Communication Network Switching (RCNS) mechanism: During poor network conditions, this mechanism provides suitable network selection to decide whether to choose Cellular (i.e., 4G/SG/6G), Wi-Fi or Bluetooth network, respectively based on the available bandwidth. To illustrate RCNS mechanism, we propose learning based reliable communication network switching (L-RCNS) and rule based reliable communication network switching (R-RCNS) models, respectively to provide reliable connectivity compared to Cellular/Wi-Fi/Bluetooth in poor network conditions. Based on the real data-set for Cellular, Wi-Fi and Bluetooth collected by Samsung Galaxy S20 device and Raspberry Pi, we conduct extensive experiments to compare performance of the mechanisms with respect to the state of art.
DOI:	10.1109/FiCloud49777.2021.00041