Online Dependent Task Offloading by Application Partitioning in Edge Intelligence for Internet of Vehicles

The Internet of Vehicles offers a comprehensive perception of environment, which enhance transportation efficiency. To handle the large amount of collected data, distributed edge intelligence is a promising paradigm in which the edge server share data and computing resources with each other, providing low-latency services for local devices. However, offloading the computing-intensive application fully to one edge server might lead to a large latency as the computing resource of edge servers are usually limited. To solve this problem and elevate Quality of Service (QoS) to new heights, existing methodologies merely partition applications into modules, overlooking the crucial fact that these modules harbor distinct input requirements, posing a pivotal challenge in scheduling optimization. In this article, we study dependent task offloading by partitioning applications and dividing modules into two categories: 1) stateful modules and 2) statelss modules. The stateful modules necessitate the incorporation of previous calculation results, while stateless modules operate independently. We subsequently frame this intricate dependent task offloading challenge as an optimization problem, boldly acknowledging its NP-hard nature. Considering this, we unveil an innovative online collaborative dependent task offloading (OCDTO) algorithm, grounded in a two-layer collaborative edge computing architecture. This algorithm meticulously minimizes the make-span, redefining the benchmarks for efficiency. Our rigorous experimentation not only validates but also showcases the superiority of our approach, consistently achieving the lowest average system cost compared to the state-of-the-art, which verifies the effectiveness of our proposed approach in latency-sensitive and computing-intensive scenarios.