MAGNETIC: Multi-Agent Machine Learning-Based Approach for Energy Efficient Dynamic Consolidation in Data Centers

Improving the energy efficiency of data centers while guaranteeing Quality Service (QoS), together with detecting performance variability servers caused by either hardware or software failures, are two major challenges for efficient resource management large-scale cloud infrastructures. Previous works in area dynamic Virtual Machine (VM) consolidation mostly focused on addressing challenge, but fall short proposing comprehensive, scalable, and low-overhead approaches that jointly tackle variability. Moreover, they usually assume over-simplistic power models, fail to accurately consider all delay costs associated VM migration host mode transition. These assumptions no longer valid modern executing heterogeneous workloads lead unrealistic inefficient results. In this paper, we propose a centralized-distributed failure-aware strategy minimize consumption centers. Our approach selects most adequate frequency each during runtime using distributed multi-agent Learning (ML) based strategy, migrates VMs accordingly centralized heuristic. Multi-AGent machine learNing-based Energy efficienT dynamIc Consolidation (MAGNETIC) is implemented modified version CloudSim simulator, considers overheads transition migration, evaluated traces collected from various running real utilization logs center Results show how our reduces up 15 percent compared other state-of-the-art (SoA), same QoS reducing number migrations transitions 86 90 percent, respectively. it shows better scalability than approaches, taking less 0.7 time overhead execute 1,500 VMs. Finally, solution capable due automatically migrating failing hosts draining them workload.