|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ЦЭМИ РАН |
||
MARL-LDB: Multi-Agent Reinforcement Learning Load Balancing Algorithm in a Distributed Computing Systemстатья Исследовательская статья
- Авторы: Chupakhin Andrei, Kazantaev Alexey
- Сборник: 2022 International Conference on Modern Network Technologies (MoNeTec), Moscow, Russian Federation
- Год издания: 2022
- Издательство: IEEE Computer Society
- Местоположение издательства: United States
- Первая страница: 1
- Последняя страница: 8
- DOI: 10.1109/monetec55448.2022.9960774
- Аннотация: The load balancing problem in a heterogeneous distributed computing system is considered. The problem is mathematically formulated as a stochastic game. As a multi-agent system, the heterogeneous distributed computing system can be viewed as a system with trainable agents at each computational node. There are some computational nodes that receive tasks according to a pre-established distribution (Uniform, Poisson). Tasks must be distributed among computational nodes in such a way that the average load on each computational node's resources (CPU, RAM, Storage) is the same. Based on the Proximal Policy Optimization (PPO) algorithm, a multi-agent reinforcement learning algorithm has been developed to solve the stochastic game. As a result of the experimental results, the proposed algorithm is capable of solving the formulated problem and is stable under various scenarios of computing system changes.
- Добавил в систему: Чупахин Андрей Андреевич