Research On Non-tree Like Topology And Optimal Bandwidth Allocation For Virtual Data Center

With the development of the computer industry and Internet, as the runtime environment for information service data center plays an increasingly important role in human lives. Especially in recent years, the rapid development of cloud computing technology makes the data centers become more important as the information infrastructure and brings in virtual data center that becomes the development trend of future data centers. Virtual data center is the data center that conforms the physical resources within hosts together to form a shared virtual resource pool by virtualization technology, which includes the virtualization of software, operating systems, storage and management, can achieve flexible and efficient use of resources. In the aspect of construction topology, due to the high energy consumption, building cost and low efficiency the traditional tree like structure has already been difficult to meet the scalability and build-cost effective requirements under new situation based cloud computing; In the aspect of virtualization application, the main side of the current virtualization technology in virtual data centers is the perspective of servers that contain virtualization of CPU, memory, hard disk technologies; However, the virtualization technology from the perspective of network resource has not been well researched and has few applicable values.In addition, there are a huge number of virtual machines, which belong to different users. The control and management problem and the competition in physical bandwidth resource of these virtual machines should be addressed urgently. Therefore, it’s necessary to design more advanced non-tree like physical network topology and virtual machine management and bandwidth allocation mechanisms which is on top of the new topology, to form a system. The system should provide adequate protection and support for the future data centers towards cloud computing and improve the aspect of resource utilization, application delivery, response to business needs, complexity of operational management, etc.This paper introduces the relative research work in non-tree like network topology of future data center and network virtualization technology, then analyses the key issues which should be resolved in current virtual data centers. On this basis, by deeply researching in three closely logical linked key technologies in non-tree like topology, virtual data center network virtualization architecture and virtual network bandwidth allocation algorithm, This paper has gained several innovation contributions as follows:(1) Aiming to the requirements for more flexible and highly scalable capability of "shipping-container" data centers, a Hypercube based modular network architecture named MCube for data centers with appropriate addressing mechanism and basic server-centric routing protocol is proposed. MCube leverages commodity mini-switches to reduce building cost and can provides strong scalability, only needs incremental deployment to scale out without changes to existing network equipment. There are many multi paths between any servers in the MCube network. It achieves higher network capacity and fault tolerance.(2) Aiming to improve the pool compatibility of the new non-tree like topology, a mirror-fat-tree interconnecting structure together with appropriate address assignment and multi path routing mechanism is proposed. The structure can achieves high compatibility to the upper applications, reducing the building cost and improving the network capacity. In the structure, each server is able to communicate with others at full bandwidth of its network port even in full network load. Due to the novel multi path mechanism, mirror-fat-tree achieves one step ahead in traffic distribution than traditional approach. It not only enhances fault tolerance but also greatly accelerates typical traffic patterns in data centers.(3) Due to the multi tenancy mechanism, there are great amount of virtual machines in current virtual data center, which can lead to control and management problems. A network virtualization architecture based programmable switches for data center by leveraging Internet network virtualization technology is proposed. The proposed architecture divides the whole data centre system into substrate hardware providers and service providers, solves the control and management problems. On top of the virtualization architecture, an active targeting mechanism is proposed to support re-addressing in virtual machines migration, which can achieve faster response time than traditional DNS based realization.(4) On top of the network virtualization architecture, two dynamic bandwidth allocation algorithms are proposed from the view of optimizing congestion of each virtual network, one based optimization decomposition, another base particle swarm intelligence algorithm. The algorithms can solve bandwidth competition among virtual machines resided on the same physical host. Experimental results show that the congestion based virtual bandwidth allocation algorithms has good convergence, can achieve dynamic and efficient allocation of network bandwidth resources(5) From the point of view of the commercial operation mode in cloud computing environment, two game based bandwidth allocation models for virtual networks are proposed. One of the two model base single layer non-cooperate game theory, in which the hardware owner of the data center announce a price to drive all virtual networks to a Nash Equilibrium, then can achieve maximal revenue of virtual networks; Another model base double layer Stackelberg game which has a leader-follower structure, provides more flexible pricing mechanism and achieves maximal revenue of both the hardware owner of the data center and the virtual Network tenants. Theoretical proof and experimental results show that the two models can achieve optimal solutions and a fair and efficient allocation of network resources.The above contributions are of great significance to build low-cost, high performance and scalability network architecture for virtual data centers, also be important to solve the control and management problems of the large number of internal virtual machines belong to different users in current virtual data centers. The integration of them can offer an effective technical means to researching of future data center solutions under could computing environment.