Article Preview
Top1. Introduction
Cloud computing is an emerging paradigm, which provides services by virtualizing hardware and software resources to satisfy users’ demands. So, business and institutions are eager to build their private cloud systems; otherwise, some famous public clouds, such as Amazon, Google and Microsoft, also provide cloud services. However, single cloud cannot provide unlimited services with limited physical resources; when the amount of service requests increases, single cloud may not have enough resources to provide services, and the federation of multiple clouds may be one possible solution. Therefore, this paper proposes a novel efficient resource allocation mechanism for federated clouds to improve system throughput and reduce resource repacking overhead in the auto-scaling mechanism.
In general, the federated cloud consists of multiple IaaS clouds to enable any participant to buy and sell capacity on demand. When one user suddenly needs a lot of computing services, he can buy the needed capacity from the federated cloud. Even a small-scale IaaS cloud can offer a global service without extending infrastructure. Therefore, the federation of multiple IaaS clouds allows a client to select the best cloud, in terms of cost, flexibility, and availability of resources, to meet a particular requirement and enable further cost-efficiency due to efficient resource allocation.
Obtaining resources from federated clouds is an important issue in dynamic provisioning. In general, when the workload changes, the resource allocation mechanism could adopt vertical/horizontal scaling fashions to repack the required resource into virtual machines. In the vertical scaling approach, the resource allocation mechanism allocates more resources into virtual machines for improving virtual machine’s capability. In the horizontal scaling approach, the resource allocation mechanism allocates more virtual machines for enhancing the virtual cluster’s capability.
When the load of requested services varies, the general resource auto-scaling mechanism usually varies the required resources by increasing or decreasing the number of virtual machines (horizontal elasticity). However, in some cases, the vertical scaling approach by varying the resources in the virtual machine is necessary for reducing the communication overhead. By the way, when the load of services changes, the allocated virtual machines maybe need to be repacked for better performance. However, frequent resource repacking may reduce the system performance. Therefore, this paper proposes a novel efficient resource allocation mechanism for federated clouds to improve system throughput and reduce resource repacking overhead in the auto-scaling mechanism. The proposed work takes the communication overhead into consideration. The proposed approach allocates resources for configuring the virtual cluster. In the meantime, the repacking overhead and communication overhead may be reduced. Our approach includes two phases. The first phase analyzes the execution pattern of applications and the cost/efficiency of resources in the federated clouds. In the second phase, the proposed approach allocates resources according the cost/efficiency and the execution pattern of applications for improving system performance.
In this study, we adopt a federated cloud which consists of the clouds from three universities: National Tsing Hua University (NTHU), National Taichung University of Education (NTCU), and Providence University (PU). This federated cloud applies the GRE tunnel to connect individual clouds into one virtual internet domain. On the federated cloud, this study proposes an H&V resource allocation algorithm to improve system throughput and reduce communication. This study profiles the characteristics of jobs in advance, and then creates virtual clusters among clouds in considering the system performance. Performance for NAS Parallel Benchmarks (for example, Embarrassingly Parallel (EP), Conjugate Gradient (CG), Integer Sort (IS)) is evaluated. Experimental results show that our approach could reduce repacking and communication overhead, and improve the resource utilization and throughput.
This paper is organized as follows. Section 2 discusses related work and motivation. Section 3 presents the proposed methodology. Section 4 presents experimental results. Section 5 analyzes the results and evaluates the performance of this study. The final section gives the conclusions and future works.