Research On Virtual Network Embedding Algorithm In Cloud Data Center

In recent years,the applications of cloud computing have been experiencing a rapid increasing.Cloud computing integrates infrastructure resources into a resource sharing pool using virtualization technology,which allows users to configure and access on demand.With the increasing popularity of various cloud computing services,the need for computing and storage has increased significantly,which brings the rapid development of cloud data centers.Cloud data center pools the underlying infrastructure resources by cloud computing,and allocates resources in the pool to virtual network requests by virtual network embedding technology,to meet the resource requirements and Quality of Service(Qo S)requirements of virtual network requests.The explosive growth of applications has brought a great challenge to cloud data centers,that is,cloud data centers must efficiently utilize the underlying physical resources at a reasonable cost to provide satisfactory services for virtual network requests.However,the resource utilization of cloud data centers is low,multiple tenants/users coexist,and heterogeneous applications cause load imbalance.Therefore,a key challenge for cloud data centers is to achieve fast,flexible,and efficient resource allocation in data center networks with heterogeneous applications and multiple tenants/users.In view of this challenge,this thesis conducts the following research:1.This thesis studies a proximal virtual network embedding method for cloud data center based on load balancing.Since multi-dimensional load balancing was not considered in the existing virtual network embedding algorithms for cloud data center,the problem may happen that some dimensions of resources become unavailable due to the exhaustion of some other dimensions of resources,which leads to low resource utilization.In this thesis,a specific algorithm is designed for the Fat-Tree topology,which is the most widely used topology in current data center.First defines the related measures for the multi-dimensional load balancing,and then the virtual network embedding problem of cloud data center is modeled as an optimization problem aiming at minimizing bandwidth resource consumption and maximizing load balance,based on which a heuristic virtual network embedding algorithm is proposed for cloud data center.This algorithm embeds virtual nodes proximally,that is,the virtual nodes of a virtual network request are embeds within the same rack if possible,otherwise extends to the same pod,and then the global data center.In the embedding process,computing and bandwidth resources are considered,and the balance between the two dimensions of resource is realized in the embedding domain,which alleviates the resource waste caused by single-dimension resource exhaustion,and improves the request acceptance rate,resource utilization and the balance index of the data center.2.This thesis builds a virtual network embedding simulation platform for cloud data center.Considering that the software simulation results are too ideal,this thesis builds a virtual network embedding simulation platform based on network simulation software GNS3,virtual network simulation tool Mininet and SDN controller Floodlight.Firstly,the overall architecture of the platform is given and the function implementation of each module is presented.Then,construction steps are descripted in detail,following which the performance test experiments of virtual network embedding algorithms is carried out in a small Fat-Tree network on this platform.In the experiments,a K8 s cluster is built on the server nodes to manage containers in a unified manner,where the containers are deployed on the server nodes with the help of Docker to simulate the operation of computing tasks.Iperf streams are used to simulate the operation of communication tasks on the links,and Floodlight controller is used to deliver flow table to execute routing decisions.Performance analysis tools are used to monitor the running status of nodes and links,and multiple groups of virtual network embedding results are tested to verify the performance of the proposed virtual network embedding algorithm.