Machine Learning to Control Network Powered by Computing Infrastructure

Abstract

Machine learning (ML) methods are applied to optimal resource control for Network Powered by Computing Infrastructure (NPC)—a new generation computing infrastructure. The relation between the proposed computing infrastructure and the GRID concept is considered. It is shown how ML methods applied to computing infrastructure control make it possible to solve the problems of computing infrastructure control that did not allow the GRID concept to be implemented in full force. As an example, the application of multi-agent optimization methods with reinforcement learning for network resource management is considered. It is shown that multi-agent ML methods increase the speed of distribution of transport flows and ensure optimal NPC network channel load based on uniform load balancing; moreover, such control of network resources is more effective than a centralized approach.