| The development of wireless networks has deeply impacts on the life of human society. With the deployment and promotion of LTE/LTE-A networks,mobile internet services have experienced an unprecedented growth, such as instant message, mobile social applications, high definition video services and online web services. Furthermore, the upcoming 5G technologies will bring profound changes on Internet of Things, Industrial Internet, Internet of Vehicles and other communication scenarios. However, the rapid development of wire-less communication systems and the exponentially increasing of mobile ser-vice demands have brought new challenges to wireless networks, which could be summarized as: (1) The challenges between the diversity and dynamics of network scenarios and rigid network architecture. (2) The challenges between the differentiation of service quality of experience and the onefold QoS oriented network management mechanism. (3) The challenges between the coordination among heterogeneous networks and the independent deployment and manage-ment of them. Due to the above challenges, the demand of network scenarios and services are difficult to match with the multi-dimensional wireless network resources. Current wireless network architecture fails to fulfill the services'quality of experience, and network resources are usually unable to adapt to the dynamic changes of service and network scenarios.To deal with the above challenges, this paper proposes the on-demand wireless network management. An on-demand wireless network virtualization architecture is designed, and the testbed platform is developed based on open source projects. Then, the on-demand management of wireless resource and network resources are deeply studied in the paper. The achievements of this research could be used to promote the coordination of heterogeneous networks and reduce wireless network costs. The on-demand management of wireless re-sources could guarantee the services' quality of experience from a micro scope of a single service. And the on-demand management of network resources could promote network and service performance from a macro scope of the whole network services. The major contributions of this paper can concluded as fol-lowings:1. Propose an on-demand architecture based on virtualization and micro-service. The design is carried out from two aspects: the on-demand man-agement of wireless virtualized resources and virtualized network resources.Firstly, on the radio access network (RAN) side, with reference to software defined network (SDN) and network function virtualization (NFV) technolo-gies, virtual resource layer (VRL) and a centralized RAN access controller are introduced. In order to percept the “demand” of services, network situation awareness is deemed as one of the key technologies. Under the virtualization architecture, the new wireless network management model is discussed. And then the feasibility and performance of the proposed architecture are verified by using a semi-physical simulation. Secondly, considering rapid iteration of wire-less services and diversities of network scenarios, micro-service (MS) concept is introduced to networks. On-demand network deployment is implemented through dynamic mapping of micro-service function chains and dynamic re-source allocation among micro-service instances. Lastly, MS communication mechanism, MS discovery mechanism and procedure of on-demand network deployment are carefully designed.2. Implement the testbed platform. According to the virtualization archi-tecture, the testbed platform is implemented from four aspects, including radio access network, core network, user device, and the deployment of networks.Considering the weak control of WLAN networks, software access point based WLAN architecture is designed to promote the network' s control of user de-vices and network resources. Centralized radio resource scheduler is introduced in LTE network to improve the global control of radio resources. And then virtualized WLAN and LTE network are implemented based on open source projects. A Floodlight based RAN controller is designed, while unified wire-less resource management and network status perception are realized on the controller. A virtual switch layer is introduced into user devices, which makes the devices could support multi-connections. Furthermore, Juju and Openstack are used to develop micro-service orchestrators.3. Study services, quality of experience oriented on-demand wireless vir-tual resource management. In this part, the paper firstly studies the perception model of services' quality of experience. According to Natural Scene Statistics(NSS) theory, the distribution of Shearlet coefficients of video data are used as the NSS features, which are more sensitive to wireless distortions. Com-bined with the back propagation neural network (BP-NN), a no-reference video quality assessment model is developed. And the performance is verified on the LIVE video quality database and the EPFL-PoliMI database. For web browsing service, the perception information concept is defined, and impact of perception information and service response delay on quality of experience are carefully studied. And then a Weber-Fechner law based web services' quality of expe-rience perception model is established. Subjective experiments are carried out to verify its performance. Secondly, based on the acquired demand of service quality, the on-demand wireless virtual resource management is studied. Apart from guaranteeing the basic demand of users' services, the network charges for the additional service quality requirements by using a dynamic pricing mecha-nism, which could promote service's quality of experience and also avoid too much occupation of network resources. The on-demand virtual resource man-agement is modeled as a Stackleberg game. Based on the VRL and centralized controller, this game could be solved by using complete information approach.4. Study service function chain based on-demand microservice mapping and virtual resource management. Micro-service based network management could be used to perform on-demand network deployment and adjustment. Micro-service based network deployment consists three major folds: first, composing related micro-services into service function chain. Second, mapping the service function chain onto physical networks. Third, allocating network resource to each micro-service instance according to the service's demand. Due to the mu-tually affects of these three folds, the on-demand network deployment is jointly modeled as a MILP problem from the node mapping perspective, which is an NP-hard problem and difficult to find the optimal solution in polynomial time.In order to get the optimal solution by using CPLEX optimization software, the on-demand network deployment is re-modeled from a link mapping perspec-tive. The results have shown that CPLEX can get the optimal solution in finite time in a small network with 50 nodes or less, however the procedure takes a rather long time. And then, a two-step heuristic method are proposed to solve the optimization problem. Extensive simulations are performed to evaluate the performance, and results have shown that the proposed algorithms can get a so-lution within 1.25 times of the optimal solution with reasonable execution time,which indicates that it can be used for online planning. |
PDF Full Text Request