End-to-end control and optimization in information and calculating networks

Abstract

Distributed systems and networks of computers assume the presence of multiple nodes interacting with each other through a packet switching network. Each node generates data packets, consisting of a certain number of bytes of useful information and bytes of service information, such as: node addresses, data type, checksum, etc. A packet switching network consists of a set of switching equipment; during each data transfer between nodes, a path is chosen that includes a certain amount of this equipment. The article describes a distributed information and computing system, controlled by all nodes from a single control center. A method for optimal control of such a system by the criterion of minimum data delivery time has been developed. Given that the delivery time of packets depends on the number of switching equipment that they pass, various types of network interference, internal and external noise in the monitoring channels and inside the control object are analyzed. It is shown that in the presence of random delays in the delivery of data and noise in the observations, end-to-end system management should be considered as an asymptotic approximation. It is proposed to solve this problem by means of linearised recurrent control minimizing the functional of delaying the delivery of packets in a system with arising stochasticities. This approach opens up good prospects for minimizing data delivery delays in modern distributed systems with remote spaced elements. And taking into account the unified control center, it makes it possible to create an intelligent subsystem for analyzing requirements in the data packet transfer system to eliminate downtime in the data delivery environment.