Optimal graph partitioning for time-sensitive flow scheduling towards digital twin networks

Abstract

The growing maturity of Digital Twin (DT) technology represents a quantum leap towards the realisation of Industry 4.0 and beyond. DT refers to virtual representations (in a virtual space) of physical objects/processes/systems (in a physical space), where information is regularly exchanged between them to enable real-time remote control and monitoring. DT will significantly improve product life-cycles and will transform industries such as smart manufacturing, smart transportation, and so forth. Digital Twin Networks (DTNs) are envisaged to be the norm where multiple DTs are logically connected to their respective physical objects, forming a many-to-many communication relationship. Strict real-time communication for bi-directional data flows is required in DTNs for DTs to accurately reflect the changes in the physical objects, and vice-versa. One potential candidate to achieve real-time data transmission is Time Sensitive Networking (TSN). The IEEE 802.1Q working group has developed a set of TSN standards to facilitate data transmissions that require low-latency, high availability and reliability. In TSN, the Central Network Controller (CNC) computes the schedule of frame transmission. However, the computational time required can exponentially increase as the number of nodes and data flows increases (already typical in industrial environments and will increase exponentially with DTNs). In this paper, we propose a novel technique using multi-level graph partitioning theory with Integer Linear Programming (ILP) to facilitate TSN scheduling. Our results demonstrated significant improvements in computational time and the success rate of scheduled data flows in complex networks where there are up to 100 nodes and 350 data flows.