A Survey on the Use of Partitioning in IoT-Edge-AI Applications

Centralized clouds processing the large amount of data generated by Internet-of-Things (IoT) can lead to unacceptable latencies for the end user. Against this backdrop, Edge Computing (EC) is an emerging paradigm that can address the shortcomings of traditional centralized Cloud Computing (CC). Its use is associated with improved performance, productivity, and security. Some of its use cases include smart grids, healthcare Augmented Reality (AR)/Virtual Reality (VR). EC uses servers strategically placed near end users, reducing latency and proving to be particularly well-suited for time-sensitive IoT applications. It is expected to play a pivotal role in 6G and Industry 5.0. Within the IoT-edge environment, AI plays an important role in automating decision and control, including but not limited to resource allocation activities, drawing inferences from large volumes of data, and enabling powerful security mechanisms. The use cases in the IoT-Edge-cloud environment tend to be complex resulting in large AI models, big datasets, and complex computations. This has led to researchers proposing techniques that partition data, tasks, models, or hybrid to achieve speed, efficiency, and accuracy of processing. This survey comprehensively explores the IoT-Edge-AI environment, application cases, and the partitioning techniques used. We categorize partitioning techniques and compare their performance. The survey concludes by identifying open research challenges in this domain.