A case for fully decentralized dynamic VM consolidation in clouds

摘要: One way to conserve energy in cloud data centers is transition idle servers into a power saving state during periods of low utilization. Dynamic virtual machine (VM) consolidation (VMC) algorithms are proposed create times by periodically repacking VMs on the least number physical machines (PMs). Existing works mostly apply VMC top centralized, hierarchical, or ring-based system topologies which result poor scalability and/or packing efficiency with increasing PMs and VMs. In this paper, we propose novel fully decentralized dynamic schema based an unstructured peer-to-peer (P2P) network PMs. The validated using three well known algorithms: First-Fit Decreasing (FFD), Sercon, V-MAN, migration-cost aware ACO-based algorithm. Extensive experiments performed Grid'5000 testbed show that once integrated our schema, traditional achieve global very close centralized system. Moreover, remains scalable Finally, algorithm outperforms FFD Sercon released requires less migrations than V-MAN.