Game-theoretic method of decentralized load balancing in microservice architectures

Abstract

The paper presents a game-theoretic method of decentralized load balancing in microservice architectures, aimed at improving the efficiency of request distribution among service instances without using centralized controllers. The main idea is to represent the balancing process as a non-cooperative potential game in which each microservice is considered an autonomous agent seeking to minimize its own cost function. Unlike traditional algorithms such as Round Robin and Least Connections, the proposed approach is based on adaptive adjustment of agent strategies depending on the current state of the system, which ensures the achievement of Nash equilibrium and a stable load distribution.

The mathematical model of game-theoretic method of decentralized load balancing takes into account the intensity of the request flow, the throughput of each node, and the quadratic component of the cost associated with resource overload. To optimize the decision-making process, a stochastic Softmax-update dynamics are used, which approximate the gradient descent of the potential function. This allows the system to gradually balance the load even in the presence of asynchronous updates and communication delays between nodes. It has been proven that the process converges to a stationary state in polynomial time, ensuring scalability and predictable behavior in large distributed environments.

An experimental study conducted in the SimPy simulation environment demonstrated that the proposed method significantly outperforms classical algorithms in key metrics. Under peak load, the Game-Theoretic Load Balancer algorithm reduced the average system response time compared to the Round Robin and Least Connections algorithms. The standard deviation of processor utilization decreased, indicating a more balanced workload distribution and the absence of overloaded computing nodes.

The obtained results confirm the analytical stability, convergence, and practical effectiveness of the game-theoretic approach. The developed method ensures adaptive self-regulation of the system, minimizes the risk of overload, and increases the reliability of microservice architectures. Future research should focus on integrating game-theoretic method of decentralized load balancing with cloud orchestrators and extending the model to multi-level games in hybrid computing environments.