No abstract

Virtual machine (VM) consolidation in data centres is a technique that is used to ensure minimum use of physical servers (hosts) leading to better utilization of computing resources and energy savings. To achieve these goals, this technique requires that the estimated VM size is on the basis of application workload resource demands so as to maximize resources utilization, not only at host-level but also at VM-level. This is challenging especially in Infrastructure as a Service (IaaS) public clouds where customers select VM sizes set beforehand by the Cloud Service Providers (CSPs) without the knowledge of the amount of resources their applications need. More often, the resources are overprovisioned and thus go to waste, yet these resources consume power and are paid for by the customers. In this paper, we propose a technique for determining fixed VM sizes, which satisfy application workload resource demands. Because of the dynamic nature of cloud workloads, we show that any resource demands that exceed fixed VM resources can be addressed via statistical multiplexing. The proposed technique is evaluated using VM usage data obtained from a production data centre consisting of 49 hosts and 520 VMs. The evaluations show that the proposed technique reduces energy consumption, memory wastage and CPU wastage by at least 40%, 61% and 41% respectively.

Datacenters are becoming the indispensable infrastructure for supporting the services offered by cloud computing. Unfortunately, datacenters consume a lot of energy, which currently stands at 3% of global electrical energy consumption. Consequently, cloud service providers (CSP) experience high operating costs (in terms of electricity bills), which is, in turn, passed to the cloud users. In addition, there is an increased emission of carbon dioxide to the environment. Before one embarks on addressing the problem of energy wastage in a datacenter, it is important to understand the causes of energy wastage in datacenter servers. In this paper, we elaborate on the concept of cloud computing and virtualization. Later, we present a survey of the main causes of energy wastage in datacenter servers as well as proposed solutions to address the problem.

Virtualization has enabled cloud computing to deliver computing capabilities using limited computer hardware. Server virtualization provides capabilities to run multiple virtual machines (VMs) independently in a shared host leading to efficient utilization of server resources. Unfortunately, VMs experience interference from each other as a result of sharing common hardware. The performance interference arises from VMs having to compete for the hypervisor capacity and as a result of resource contention, which happens when resource demands exceed the allocated resources. From this viewpoint, any VM allocation policy needs to take into account VM performance interference before VM placement. Therefore, understanding how to measure performance interference is crucial. In this paper, we propose a simple experimental approach that can be used to measure performance interference in Infrastructure as a Service (IaaS) cloud during VM consolidation.

Cloud computing has gained a lot of interest from both small and big academic and commercial organizations because of its success in delivering service on a pay-as-you-go basis. Moreover, many users (organizations) can share server computing resources, which is made possible by virtualization. However, the amount of energy consumed by cloud data centres is a major concern. One of the major causes of energy wastage is the inefficient utilization of resources. For instance, in IaaS public clouds, users select Virtual Machine (VM) sizes set beforehand by the Cloud Service Providers (CSPs) without the knowledge of the kind of workloads to be executed in the VM. More often, the users overprovision the resources, which go to waste. Additionally, the CSPs do not have control over the types of applications that are executed and thus VM consolidation is performed blindly. There have been efforts to address the problem of energy consumption by efficient resource utilization through VM allocation and migration. However, these techniques lack collection and analysis of active real cloud traces from the IaaS cloud. This paper proposes an architecture for VM consolidation through VM profiling and analysis of VM resource usage and resource usage patterns, and a VM allocation policy. We have implemented our policy on CloudSim Plus cloud simulator and results show that it outperforms Worst Fit, Best Fit and First Fit VM allocation algorithms. Energy consumption is reduced through efficient consolidation that is informed by VM resource consumption.