Resource Optimization For Cloud Services In Mobile Network

Since the 1980s, the rapid development of mobile communication technology has brought a huge change to the people’s daily life. By the 1990s, with the popularity of the Internet, the development of mobile system has turn to a new direction, i.e., the MobileInternet (MI). The advancing networking and hardware technologies have contributed to the decrease of server and bandwidth cost. Therefore, plenty of Cloud applications have emerged. The integration of MI and Cloud techonology are natural and inevitable. The Internet — especialy the MI, cannot be described without the aid of Cloud Computing (CC). In general, all services in MI can be regard as Cloud services. "Cloud" has become the key feature and trend of the future MI.The offloading mechanism determines the two notable features of Cloud services, i.e., the interactivity and interchangeability. The mobile device sends task specification (including the input data) to the Cloud server, then the server perform the task accordingly and sends output results back to the mobile device. In this process, the mobile user achieve computing/storage resources by offer corresponding communication resources. We can conclude that the resource optimization for mobile Cloud services can be focused on two aspects, i.e., the computing/storage resources and the communication resources, or the cooperation between them. In this dissertation, we first make architectural design and necessary assumptions for the mobile Cloud system, then the involved resources optimization problems are proposed and analyzed. The major contributions of this dissertation include:1. Architectural design for mobile Cloud systemsWe combine the CC system and mobile communication system into a whole. By introducing the Cloudlet concept into the radio access network (RAN) nodes, the "Cloud" can be pull closer to the RAN form the backbone network, and the resource of mobile devices becomes a part of the Cloud. Form the aspect of the mobile user, the available Cloud resources in the network can be divided into three layer, i.e., the remote Cloud (the upper layer of the three-layered architecture, located at the backbone network and can be reached by wired and wireless link), local Cloud (the middle layer of the three-layered architecture, located at the local service area and can be reached directly by wireless link) and the terminal Cloud (the lower layer of the three-layered architecture, consist of the resources of mobile terminals and can be reached by D2D link). Such architechture eliminates the boundary between "Cloud" and "terminal". The communication and computing ability of the whole network can be regarded as some kinds of "services", which can make the "Everthing as a Service" concept into realization.2. Transmission optimization for local Cloud servicesThe interaction between Cloud server and mobile terminal will cause bi-directional (uplink and downlink) data transmission in the network, and the uplink traffic and the downlink traffic are highly correlated. In the three-layered Cloud architecture, the local Cloud occupies an important position. Local Cloud services can avoide the backbone transmission, therefore can achieve the highest sensitive services responses. This part of optimization focuses on the communication resources in the network, the objective of which is to minimize the roundtrip delay of service data flows. The queuening theory is used for model formulation of the service traffic. According to different service arrival characteristics, we proposed different control scheme to jointly allocate the uplink and downlink resources in the system. Numerical and simulation results indicate that the proposed algorithm can significantly improve the service sensitivity and the resource utilization of the system.3. Load scheduling for terminal Cloud servicesThe terminal Cloud is the lower layer of the three-layered Cloud architecture and consists of the scattered resources in the terminal devices. By the aids of rich sensing functions, the smart terminal devices are perfect platform for users to express emotions and share information. Terminal Cloud computing is a new computing mode which is similar with Mobile Grid Computing and Mobile Social Computing. We consider the scenario in which the mobile devices can share their idle computing resources to realize parallel processing. A Stackelberg Pricing Game-based divisible load scheduling algorithm is proposed to fully utilize the idle resources of each mobile devices. Simulation results indicate that the proposed algorithm can significantly improve the time gain of the system with perfect convergence.4. Resource coorperation for cross-layer Cloud servicesWe introduce the "Follow Me Cloud" concept into distributed cloud systems, which enables the service migration between federal resource pools. Therefore, the mobile users can always get the optimal quality of service (QoS) when make cross-area motion. This mechanism involves the coorperation of more than one layer of Cloud resources. In this dissertation, we proposed a novel resource control algorithm for resource coorperation in a distributed cloud system. The optimization objective is to maximize the utility of the overall system. The six-directional random walk mobility model is adopted to describe the users’movement. Then a semi-Markov Decision Process-based model is formulated to help resource controller make access control (for new service requests and service migration requests) and resource allocation decision (for accessed service requests). Numerical results indicate that the proposed algorithm can significantly improve the overall system utility, balance the network loads and improve user experice quality.