On Traffic Routing in Network Powered by Computing Environment: ECMP vs. UCMP vs. Multi-Agent MAROH

The problem of efficient traffic balancing in a data network environment is considered in relation to Network Powered by Computing (NPC)—a new generation of computational infrastructure, where computational resources and data transmission resources form a network—“Network is a Computer”. This problem requires making load balancing decisions in a rapidly changing environment without knowledge all the traffic parameters. For this reason this paper considers multi-agent reinforcement learning approach (MARL) and introduces a novel approach: multi-agent reinforcement learning with hashing (MAROH). This approach is evaluated compared to standard load balancing algorithms: ECMP and UCMP, which shows improved effectiveness of the load balancing.