Research On The Key Issues About Efficient Resource Management In The Data Center

Cloud computing, a newly developing distributed computing paradigm, has been strongly supporting the continued development of various network based applications and services. Data centers(DCs), as the mainstream resource-provisioning frameworks adopted by the cloud computing, provide a flexible and stable underlying resource environment for the cloud computing running efficiently. With the flexible and efficient resource management in the data center, the underlying physical resources can be allocated in demand, which can considerably improve the usage of physical resources.Furthermore, as tenants only need to pay for the number of resources they used, cloud computing dramatically accelerates the development of applications and services towards the directions of becoming large-scale and widely diverse.In cloud computing, with the virtualization technology, the underlying distributed computing and networking resources are virtualized and abstracted as a resource allocation pool, in which resources are centrally managed by cloud providers(CPs) and allocated to tenants. In this situation, a significant issue the CP should seriously consider is how to manage the resource pool to improve the utilization of underlying physical resources while guaranteeing tenants’ Service Level Agreement(SLA) requirements. As efficient resource management scheme can typically not only improve the utilization of resources and reduce the energy consumption of data centers, but also increase the revenues that cloud providers can obtain, how to design an efficient resource management scheme has been a hot topic attracting a considerable attentions from industries and academics.In Chapter 1, the concepts of cloud computing, its developing history and the related technologies of virtualization are first introduced. Then, key issues about the-state-of-art resource management schemes in the data center are briefly presented.In Chapter 2, a joint resource scheduling and allocating issue is studied to improve the effectiveness of resource allocation and increase the expected revenue the CP might obtain. In this study, the admission control and the placement for each virtual machine(VM) request are jointly considered. As separately considering these two issues may typically result in the expected revenue degraded and the global optimization solution hard to be obtained, a joint optimization policy is proposed to combine these two problems.To tackle this joint optimization problem, the Markov Decision Progress(MDP) modeling framework is used to model this issue to estimate the potential impact caused by the action currently made to the expected revenue that might be obtained in the future. Based on this estimation, better actions can be done for coming VM requests towards globally optimizing the allocation of resources and increasing the revenue to be obtained. Due to the enormous complexity of solving the MDP model directly, an approximate solution is proposed by integrating one or more specifically given VM placement scheme(s) into a sample-based approximate dynamic programming solution to speed up solving this MDP model while ensuring its accuracy.In Chapter 3, the VM migration issue is studied in the situation where the number of resources required by each VM constantly varies over time. In order to reduce the number of VM migrations due to the bursts of increased resource requirements between VMs, a Service Level Agreement(SLA) based VM migration triggering scheme is proposed. By monitoring the resource usage of VMs during the past period, it can effectively reduce the number of VM migrations frequently triggered by transient resource competitions. Based on this migration triggering scheme, a resource usage correlation based VM migration scheme is also proposed to choose the most appropriate VMs that should be migrated as well as their migration order, and the servers to which VMs are migrated. By combining these two schemes, two VM migration cases are also studied which are respectively towards guaranteeing VMs’ SLA requirements and saving the energy consumption in DCs.In Chapter 4, the responding delay performance preserved VM placement issue is studied. Firstly, an in-depth analysis on the relationship between the VM’s responding delay and the number of resources allocated is carried out via the theoretical method and verified via experiments. Then, the exact number of resources that should be allocated to preserve the responding delay after the VMs are split is also studied. Based on this analysis, a novel splitting based VM resource allocation and placement scheme is proposed with the aim of reducing the number of resource fragments while preserving VMs’ responding delay. In this scheme, VMs that have large number of resource required may be split into multiple VMs having smaller resource requirements. With this method, the smaller resource allocation granularity can be achieved, which can substantially benefit to improving the flexibility of resource allocation and reducing the resource fragments.In Chapter 5, the flexible deployment issue for virtualized network function chains(vNFCs) is studied and a novel vNFC deployment scheme based on the multi-to-one instantiation for virtualized network functions(vNFs) is proposed to increase the utilization of the underlying physical resources. In this scheme, multiple instantiations of a single vNF might be intentionally instantiated to spread the workload and the resource requirement of each vNF by fully exploiting the advantages of vNFs that could be easily deployed in a decentralized manner as well as the flexibility of virtual resources that could be allocated in demand. Thus, each vNF can be widely spread and deployed across the entire underlying network, and easily scaled in and out as the tenant’s demand varies.Then, a more fine-grained resource allocation scheme is proposed to carefully choose the most appropriate resource management scheme including the number of resources allocated and the placement for each instantiation of each vNF. Hence, the purpose of improving the utilization of physical resources can be achieved.In Chapter 6, the author concluded all the research in this dissertation and presented the directions for the future research.