MSU scientists have developed an innovative method for managing network traffic, allowing real-time optimization of channel load and minimization of data transfer delays — the MAROH (Multi-Agent Routing using Hashing) method. The new technology has already been presented at the international conference "Mathematics in the Constellation of Sciences."
The developed method uses artificial intelligence (AI) to automatically distribute the load between channels. It is based on multi-agent reinforcement learning — a system where each network element independently makes decisions about data routing, exchanging information with its neighbors. This is similar to how drivers choose the least congested roads during rush hour, only it happens automatically and millions of times faster.
The MAROH (Multi-Agent Routing using Hashing) method shows 1.5–2 times better performance compared to the classic ECMP and UCMP solutions, ensuring stable operation even at maximum load. The principle of operation is simple: each output port of the router is assigned a special weight, automatically adjusted depending on the current load.
Our method proves that multi-agent learning can effectively solve the problem of traffic balancing in complex networks. This is especially important for modern computing systems, where every millisecond of delay is critical.
The technology is ready for use in data centers, cloud services, corporate networks, and Internet service provider systems. In the future, the developers plan to improve the system by introducing graph networks with an attention mechanism, which will make it even more resistant to changes in load and network topology.
The implementation of the technology will improve the speed of networks and significantly reduce the load on equipment, which will lead to energy savings and reduced infrastructure maintenance costs.
Read more on the topic:
Stable Internet Provided on the Entire Route: Moscow — St. Petersburg
Yandex Learns to Recognize Slow Video Loading and Immediately Offer Analogues on Other Sites
MegaFon and Bureau 1440 to Launch Satellite Internet in Remote Regions of Russia
