Research On The Key Technologies Of Dynamic Resource Adaptation For Smarter Cloud Networking

In recent years, with the growing popularity of high-speed Internet and corporate IP connections, continually advanced by the academic and industrial community, cloud computing has become a global platform for large application deployment, data ag-gression and intensive communications. Arising from the rapid development of cloud computing and the evolution of Internet architecture, a cloud networking solution is needed to distribute the benefits of cloud computing more deeply into the network, and provide dynamic resource adaptation features at infrastructure and service lev-els. However, the traditional Internet architecture-based resource dynamic adaptation cloud networking confronts grim challenges, for example, the "rigid" and "still" de-fects of the current Internet caused by double bindings of service/location and iden-tifier/location, high energy consumption, poor scalability, low network utility and in-efficiency of service resolution. Even worse, these problems have begun to hinder the further development of cloud computing. Therefore, to break through the limitations of current cloud computing and realize smarter cloud networking, this dissertation pro-poses the key technologies of dynamic resource adaption, including dynamic resource management in data centers, dynamic service adaptation and enterprise application migration. The main contributions are summarized as follows:1. Performing in-depth research and analysis on the high energy consumption, low resource utilization, virtual machine online migration problem and poor scala-bility of cloud data centers, this dissertation proposes the virtual machine-based resource management mechanisms for cloud networking, including energy-aware virtual machine placement algorithm and multi-objective migration approaches. In the placement algorithm, with the design goal of energy-saving, application awareness and network utility maximization, a theoretically analysis model of data center virtual machine placed or replaced to the physical machine is de-veloped. Due to peak-load of applications, the performance of hosting virtual machine decreases sharply when overloading occurs. To solve this problem and provide dynamic resource management in cloud data centers, migration policy based on multi-objective algorithm is designed to achieve traffic reduction and maximize network utility.2. The cloud service trend towards diversification, pervasive, ubiquitous brings sig-nificant challenges to the binding between service and network location in current Internet. Besides, due to the host-centric design principle, domain name system can not support dynamic service adaptation in cloud computing environments. By performing in-depth research and analysis on these issues, this dissertation presents a novel naming system for cloud networking, providing universal nam-ing and resolution mechanism in which unique and persistence service identifier is defined and corresponding mapping schemes, service identifier to location and ser-vice identifier to endpoint identifier, are designed to decouple service providers from locations. Both in two service-oriented naming systems proposed in this dissertation, SIDMAP and HNRS, two-step name resolution is implemented to support dynamic service adaptation mechanisms such as service migration and composition, intelligent name resolution. According to analysis and simulation results, these two types of naming systems perform low resolution latency with good arrangement and scalability, significantly improving the availability of cloud service or data.3. Performing in-depth research and analysis on the Churn and scalability issues of cloud networking naming systems which is based on large-scale DHT system such as Chord or hierarchical Chord, this dissertation brings forward a survival analysis model for mapping node behaviors. Based on runtime data traces from real-world DHT system, a semi-parametric model called CPSCox is proposed as the function of three dominant risk factors explored by K-M method. Applying this model to analyze node’s session length, inter-session length and remaining uptime, some useful conclusions are achieved. The design principles of cloud networking naming systems are built upon these theoretical conclusions.4. Existing cloud migration approaches often suffer from the migration complexities of determining the set of components migrated to the cloud since global network information is needed to make migration decision. Besides, few performance modelings of migrating enterprise applications into cloud analyze both migration benefits and costs systematically. By partially migrating enterprise applications into cloud, a hybrid cloud architecture is achieved. Based on convex optimization theory, migration benefits and costs analytical models are deduced to determine the optimal migrated components set of an enterprise application. The migration policies proposed are analytically shown to be capable of moving an enterprise application between local data center and remote cloud data center in a way that the cost of service provision is reduced and data security is achieved by hosting intensive data on-premise. Such an approach is suitable for widespread of cloud computing.