Number of cloud data centers which consists of hundreds of hosts has increased tremendously around the world due to increase in demands for cloud services. It is expected energy consumption of data centers will reach 139.8 billion Kwh by 2020. Many algorithms are proposed to reduce energy consumption as well as service level agreement violationby minimizing the number of active hosts.Current proposed algorithms do not consider data center architecture, the physical position of hosts, and energy consumption of numerous switches that are in data centers. In this paper, a novel hierarchical cloud resource management is proposed that not only minimizes the number of hosts but also aggregates virtual machines on a limited subset of data center racks and modules to minimize energy consumption. Experimental results with Cloudsim show that our proposed algorithm reduces energy consumption up to 26% and service level agreement violation up to 96%. KEYWORDS data center architecture, dynamic consolidation, power efficiency, resource management, virtualization 1 | INTRODUCTIONIn recent years, cloud computing has attracted attention from both the academic and industrial communities. Cloud computing services are pay-as-you-go and are easily accessible via internet around the world. 1,2 The infrastructure for cloud services are cloud data centers that are composed of hundreds of computational hosts (servers), tens of high speed switches, and other network devices, all of which consume a significant amount of energy. Number of cloud data centers has increased tremendously around the world due to increase in demands for cloud services.Data center energy consumption in the world was 61 billion kWh in 2006 and 100 billion kWh in 2011, and it is expected to reach 139.8 billion kWh by 2020. [3][4][5] Hence, data center energy consumption is of great concern because it increases carbon dioxide (CO 2 ) emission, which has a catastrophic environmental impact. 6 Servers, storage, switches, and other network devices are the main sources of energy consumption in data centers. These devices consume energy near maximum power levels even when they are idle or not fully utilized.Cloud services are provided through virtualization, 7-9 ie, assigning a virtual machine (VM) for each cloud service. 10 Virtualization technology allows allocation and execution of multiple VMs on a host that leads to an increase in host utilization that brings about energy efficiency in a cloud data center. Virtualization technology also provides live migration VMs from 1 host to another in case a host is overutilized or underutilized. There will be a waste of energy if servers are underutilized, because servers use around 30% of their peak power consumption while they are in idle mode 70% of time. 11To tackle energy consumption of underutilized hosts, a common solution is VM consolidation. 9,12 If a host is underutilized or has no work to do, VM consolidation if possible, tries to live migrate its VMs to another host and switch it to sleep mode or turns it off to con...