Optimizing Cloud providers revenues via energy efficient server allocation

Service level agreements-driven management of distributed applications in cloud computing environments

“…Mazzucco and Dyachuk [105] propose an approach for allocating VMs to servers by considering energy efficiency aspects by controlling the number of running servers in the datacenter. However, they do not consider the case when a server could host multiple VMs.…”

“…Formula 6.2 determines the uptime value used in formula 6.1 for calculating the penalty value For a given server, it is possible to model its energy consumption as a linear dependency of the CPU utilization [105]. In case of multi-core/CPU servers the average CPU utilization of all cores can be used.…”

“…In case of multi-core/CPU servers the average CPU utilization of all cores can be used. As a consequence of the fact that the idle power consumption is almost 60% of the one at full load [105], keeping servers in idle state or at low utilization would produce low revenues due to the fact that the server will consume almost as much energy as running with high load, but will generate only low revenues, if at all. Formula 6.3 describes the energy costs calculation, as described in [105].…”

“…In a cloud environment, usually the VMs experience low CPU utilization, with 30% average [105] and usually having daily, weekly and monthly seasonality [64]. This helps to predict the resource utilization and do a better allocation by taking into account the quantity of resources that a VM will use, in fact enabling using the virtual machine live migration as a load balancing method [150].…”

“…Given that the average CPU utilization of a VM hosted in a data center is usually around 30% [105] it is common to overcommit the CPUs of the physical hosts with factors between 1.5 and 3 [105,109]. For example, OpenNebula [113] allows allocating virtual CPUs to a fraction of the available physical CPUs, using support of Xen and KVM [38] hypervisors.…”