Design Of Virtual Network Resource Allocation Framework Based On Software-defined

With the rapid development and popularization of cloud computing,more and more application services are developed and migrated to the cloud platform.Virtualization technology is the foundation for cloud computing to achieve efficient use of resources and build a high-availability and highly scalable cloud platform.The flourishing of cloud computing also poses more requirements and challenges for virtualization technology.At present,the virtualization technology of processors and memory is relatively mature,and I/O virtualization technology and the allocation mechanism of virtual network resources in data center virtual machines is relatively lagging.The network is an extremely important resource in the data center,and its performance is often the key to the performance of the entire system.Network I/O virtualization and live migration technologies are effective means to solve data center network contention and improve system performance.However,although the workload of data center has the characteristics of complexity and demand variability,the attributes of network workload are seldom perceived and utilized by resource scheduler in practice.Failure to dynamically schedule network resources according to the changes of workload will inevitably lead to the failure to achieve optimal throughput and performance in the allocation of network resources.It is an urgent need to design a load-aware scheduling framework based on network I/O virtualization and live migration technology to allocate network resources.Based on the in-depth study and analysis of dynamic SR-IOV technology and live migration technology,this paper proposes a virtual network resource allocation framework based on software-defined.The main contributions of this paper are as follows:(1)Improved dynamic SR-IOV technology: Dynamic SR-IOV technology is an effective method for optimizing resource allocation in data center virtual networks.However,dynamic SR-IOV is a highly restricted technology because it can only be optimized for a single physical server network,and it is not possible to schedule or allocate network resources in the data center from a global perspective.This is likely to lead to hotspots in computing and communications,which will lead to the increase in the overall cost of services and performance reduction in data centers.This paper analyzes and studies the working principle of dynamic SR-IOV technology,and improves the dynamic SR-IOV technology to compensate for the shortcomings of dynamic SR-IOV technology by combining live migration technology.(2)Proposed and designed the Sova system: This paper proposes and designs the Sova system,an optimization framework for data center virtual network allocation.It integrates virtual machine live migration technology into dynamic SR-IOV by means of software-defined,in order to improve the overall network performance and resource utilization of data center.Depending on the resource usage of the physical server and the workload characteristics of the virtual machine,Sova can adaptively optimize network allocation between different servers using dynamic SR-IOV or live migration.Therefore,Sova needs to monitor the network traffic as well as the workload characteristics in the cluster,whereby the network attributes of the physical server and the virtual machine can be obtained on the fly to direct the design and implementation of the coordination mechanism between the two technologies.(3)Implemented the Sova system based on the Xen virtualization environment: This paper implements Sova system based on Xen virtualization environment,and implements the communication mechanism between the same host virtual machine on the basis of XenStore.It also modifies XenMon and uses the way to let the virtual machine run daemon process to realize the collection of virtual machine data.In addition,this paper uses the hotplug technology of PCIe devices to minimize the adverse effects of the dynamic allocation and removal of VF on running virtual machines,and uses activate backup mode of Bonding Driver technology to achieve dynamic switching between multiple NICs,while ensuring the uninterrupted network of virtual machines.(4)Validation and analysis of the effectiveness of the Sova system based on the prototype system: In this paper,seven physical servers are used to build the prototype cluster system of Sova,and the effectiveness of the virtual network optimization framework proposed in this paper is verified on the prototype system.The experimental results show that Sova can combine the advantages of dynamic SR-IOV and live migration,and has better performance than the two technologies when used alone.