Green Cloud Broker: On-line Dynamic Virtual Machine Placement Across Multiple Cloud Providers
Federico Larumbe, Brunilde Sansò · 2016
Cloud computing controls the increasing energy consumption of applications by consolidating them in shared servers through virtualization. This technique can be greatly improved and complemented by choosing an optimal data center for each Virtual Machine (VM). That dynamic optimization problem was stated as a Mixed Integer Linear Programming (MILP) model that minimizes operational expenditures, while respecting constraints on Quality of Service (QoS), power consumption, and CO2emissions. Geographically distributed users experience varying response times depending on where users and VMs are located. Users closer to VMs experience a shorter response time, thus distributing VMs close to users improves the QoS. On the other hand, the increasing energy consumption of the cloud raised concerns about the impact on CO2emissions and global warming. Placing VMs in data centers that use green energy sources is an important way to mitigate this problem. A comprehensive optimization modelling framework and an efficient tabu search heuristic were developed to handle applications with dynamic demands in real time. Test cases had more than 1,000 nodes, 650 applications, and 6,500 VMs. Results report that the communication delay is reduced up to 6 times, 30% of power consumption is saved, and the CO2emissions can be reduced up to 60 times. The framework also allows to carefully assess the trade-offs between delay, cost, CO2emissions and power consumption.