Open Service for Networking Systems of AI: A Case Study in Adaptation Optimization

Under the architecture of Networking Systems of AI (NSAI), we propose a three-tier edge computing platform called Open Service for NSAI (OpenSAI), which can jointly configure the application execution and compute-storage-network resources, for online application-network adaption, to ensure the efficiency and reliability of the services. In this research, we investigate a specific task-device adaptation optimization problem in OpenSAI, which recommend application tasks to distributed computing devices. We evaluate the performance of various optimization and search algorithms in recommending optimal task-device combinations, which show significant differences among the algorithms in terms of recommendation runtime costs and task adaptability. Overall, the hybrid recommendation of SCORE Algorithm, Particle Swarm Optimization, and Item-Item Collaborative Filtering demonstrates the best performance in both time efficiency and fitness scores across multiple scenarios.