The problem of load balancing in a distributed network is considered. The network operates in discrete time under the influence of noise in communication channels. The node's load is formulated as a function of the node's operation parameters and the value of this function is exchanged locally among the node's neighbors. Each node, based on comparison of its load and the neighbors' loads, exchanges the tasks with its neighbors so that the load is averaged across the network. The step-size of the load adjustment procedure is chosen via fast gradient like Nesterov algorithm to achieve faster convergence. The algorithm divergence upper bound is estimated. The simulation results suggest faster convergence of the proposed algorithm.

Multi-agent management strategies rely either on the classic client-server architecture or complex inter-agent communications. Distributed control algorithms allow you to balance the load. In this paper, an innovative technology for organizing a distributed structure of servers is considered in such a way as to reduce the computational load on the nearest servers while maintaining an adequate response time to client requests, regardless of the geographical location of both clients and servers in the network. The proposed model allows to reduce the impact of random interference on short-term tasks in the system and decrease the summary duration of all long-term tasks, improving the quality of the augmented reality service provided in general.

In the paper a multi-agent network system of different computing nodes is considered. A problem of load balancing in the network is addressed. The problem is formulated as consensus achievement problem and solved via local voting protocol. Agents exchange information about their states in presence of noise in communication channels. At certain moment network system topology changes and new step size of control protocol is chosen to meet new conditions. Step size adjustment is done by stochastic approximation type algorithm. Analytically obtained optimal step size values are given. Simulation example demonstrating step size adjustment is provided.