DRL-Based Long-Term Resource Planning for Task Offloading Policies in Multiserver Edge Computing Networks

Multi-access edge computing (MEC) has been regarded as one of the essential technologies for mobile networks, by providing resources and services close to users, thereby, avoiding extra energy consumption fitting low-latency ultra-reliable requirements emerging 5G applications. Task offloading policy plays a pivotal role in handling requests maximizing network performance. Most recently developed solutions are designed instant rewards, therefore, neglecting long-term resource optimization at edge, which fail deliver optimized performance when significant increase appears. In this paper, with objective benefits on delay consumption, task policies proposed firstly avoid over-distribution through deep reinforcement learning (DRL) based reservation server cooperation, secondly maximize average reward utilization reserved an optimization-based joint consisting decision, transmission power allocation distribution. The DRL-based is evaluated simulated multi-server network. Compared previous solutions, algorithms achieve higher more reliable overall rewards. Of implemented three algorithms, fully cooperative multi-agent DRL accounts cooperation between servers, achieving 70.5% reduction variance 13.4% rewards over 500 continuous operations. Resource balanced help networks handle explosive growth computing-intensive applications future.