Computing Centric Network — Next Generation Computing Infrastructure

The architecture of next-generation computing infrastructure is presented, enabling the implementation of "computing on demand" services. A comparison is made with the GRID computing concept, highlighting the key challenges in its implementation. The necessity of applying machine learning methods to overcome these limitations through automated optimal decision-making is demonstrated. As a primary solution, a combination of software-defined networking (SDN), service virtualization, multi-agent optimization, and reinforcement learning is proposed. The novelty of this approach lies in self-organizing resource distribution without centralized control. Experiments show that the proposed method reduces task allocation time by $20-30$ compared to traditional approaches and ensures uniform channel utilization (load variation coefficient $< 15$). Thus, the effectiveness of ML-based decentralized management is confirmed.