Research On Data Center Network Structure And Its Scheduling Optimization Algorithm

In order to adapt to the development and demand of emerging application model,the data center network is going through tremendous changes,which are reflected not only in the improvement of network size,scalability and the controlling of cost,but also in other aspects,such as the resource scheduling strategies etc.Network topology and resource scheduling algorithm are the focus of research on modern data center,the existing network structures have the advantages of good scalability and fault-tolerance,however,there is a problem that scalability of these structures is limited by the number of network interface card(NIC)ports in servers,and these structures can not take into account the performance and economic costs.Secondly,with the rapid increase of the amount of users and data center size,the existing resource scheduling algorithms can not effectively balance efficiency and load balancing when dealing with user's multiple request.Aiming at the above problems,this thesis specifically includes the following several aspects:(1)Aiming at the problem that the scalability for the data center network is limited by the number of network ports in servers,a high extensible data center network structure is proposed,and is called HRN(Hyper Ring Network),where servers have a small amount of ports.HRN can build a large-scale data center network structure which is based on low-ports servers and switches by the hierarchical recursive way.This thesis firstly defines the encoding rules and construction rules of HRN,then gets and proves its topological properties according to its way of construction;Then,in order to ensure communication performance of data center,this thesis designs the routing strategy based on HRN structure,including shortest-path routing,parallel-path routing and fault-tolerant routing;Finally,this thesis describes the simulation work,simulating HRN topology and routing algorithms through experiments,and compares with other network architectures on topological performance and reliability.Experimental results show that the HRN structure is highly extensible which can achieve a large-scale topology by a lower percentage of servers and switches,and effectively reduce costs while efficient routing algorithms make the structure can provide good throughput and high reliability.(2)As a shared platform for managing and scheduling resources,the data center must have an efficient resource scheduling policy.In order to solve the problem of loadimbalance in the process of scheduling,combined the above-mentioned structure,a resource scheduling optimization algorithm for data center that is based on Max-Min Algorithm and Ant Colony Algorithm is proposed:1 This thesis designs scheduling optimization goals based on data center resource scheduling model,and then formulate resource constrain function according to the user request time constraints,reliability,communication bandwidth and other indicators;2 Aiming at the deficiency when solving resource scheduling with Ant Colony algorithm,this thesis improves the algorithm transition probability and that in the updating methods of local and global pheromone.And for the problem of uneven load in resources,the load adjustment factors is proposed and included in updating the pheromone to ensure the load balance in the process of scheduling;3 Considering the convergence of the ant colony algorithm is slow in the early iterations,so a hybrid algorithm is proposed that the Max-Min algorithm is combined with the improved Ant Colony algorithm.The Max-Min algorithm is used at the beginning of the algorithm due to it has the advantages of high processing efficiency and good effect in load balance,then the pheromone distribution of the ant colony algorithm is initialized with the optimal solution of Max-Min;4 The algorithm is simulated on the CloudSim platform and compared with Max-Min and Ant Colony algorithm,experimental results show that proposed algorithm can speed up the overall scheduling speed,and ensure the load balance in the data center,at the same time complete the scheduling in the shorter time.