Abstract: A topology-aware energy consumption optimization algorithm is designed for the high energy consumption problem in data center networks. According to the properties of multi-dimensional orthogonality and single-dimensional full mesh in the generalized hypercube, the algorithm optimizes the location of virtual machines in servers and puts forward a multi-dimensional best fit decreasing strategy to utilize multi-dimensional resources fully. With the consideration of resource balance of servers, communication cost of virtual machines and the resource consumption in migration process, the algorithm utilizes resource requirement prediction model and migration cost formula of virtual machines to satisfy the performance and energy consumption requirement of the system and relieve the congestion of links. Finally, the algorithm transforms the energy consumption optimization problem into optimized allocation of virtual machines in servers. The experimental results show that the system energy consumption and congestion are decreasing in the large scale in comparison with the other algorithms.

Key words: data center networks, energy consumption optimization, topology-aware, multi-dimensional best fit decreasing, prediction model, migration cost, congestion control

摘要： 针对数据中心网络中高能耗的问题，提出了一种拓扑感知型能耗优化算法。算法首先根据广义超立方体拓扑多维正交和单维全连接的结构特性，优化虚拟机的部署位置，进而提出多维最佳适应策略来充分利用服务器各维资源。然后利用虚拟机资源需求预测模型并结合迁移代价公式，均衡考虑服务器资源使用代价、虚拟机通信代价和迁移资源消耗，在合理迁移虚拟机以满足系统性能的前提下，降低了网络的能耗并且缓解了网络链路的拥塞。最终将网络的能耗优化问题转化成虚拟机在服务器上的优化配置问题。实验结果表明，与其他三种算法比较，算法在降低系统能耗和减少拥塞方面获得了良好的效果。

关键词: 数据中心网络, 能耗优化, 拓扑感知, 多维最佳适应, 预测模型, 迁移代价, 拥塞控制