Research On Techniques Of Performance Guarantee For Software Defined Network Function Virtualization System

The development of network function virtualization and software defined networking technology has caused more and more communication services to be transferred to operator's data centers,forming the software defined network function virtualization system,which also makes the performance protection of the system become an urgent problem to be solved.On the basis of summarizing existing methods and research results,starting from the sub-parts of the system,the research content is divided into four aspects:resource allocation of virtual network functions,request allocation of the control plane,link failure detection of the data plane,and transmission path tracking of the data plane.The main content of this thesis are summarized as follows.In terms of resource allocation of virtual network functions,we propose a resource allocation scheme based on service chain instances.In the process of constructing service chain instances,this scheme can guarantee the resource utilization,throughput and processing delay of service chains.In order to embed the constructed service chain instances into the servers of the data center,we define a convex optimization problem.The optimization goal of this problem is to minimize the average transmission delay after the service chain instances are embedded.In order to improve the usability of the solution,when embedding the service chain instances,we adopt the alternating direction method of multipliers to distribute the process of embedding to multiple servers for parallel execution,which improves the efficiency of embedding.We built a fat-tree topology for real experiments.Compared with the existing solution,our solution can significantly improve the service delay satisfaction rate,and at the same time can improve the resource utilization and throughput.In terms of request allocation for the control plane,we propose a request allocation scheme based on the queue mechanism.By constructing hierarchical queues,different requests are scheduled in the corresponding queues.In order to schedule the requests to appropriate controllers for processing,we propose a scheduling strategy based on the network pressure maximization.In order to improve the usability of the scheme,we weigh the performance and cost of the scheduling strategy.We conducted simulation experiments on fat-tree and VL2 topologies.Compared with the existing solution,our solution can increase the throughput of requests,while reducing the response delay of control plane.In terms of link fault detection for the data plane,we propose a link fault detection scheme based on the detection matrix optimization.This scheme uses only part of paths to cover all links of data plane,and only maintains probes on this part of paths,thereby significantly reducing the overhead of maintaining probes.On this basis,we propose two quality indicators of detection paths to support the selection of detection paths towards a specific quality.We conducted simulation experiments on fat-tree and VL2 topologies.Compared with the existing solution,our solution can significantly reduce the time required for a single detection,can reduce the bandwidth usage,and can reduce the server load.In terms of transmission path tracking for the data plane,we propose a transmission path tracking scheme based on the probabilistic sampling.This scheme only installs 3 rules on each switch,and only takes up the space size of double VLAN in each packet header.Based on the symmetry of data center topology,we propose a path reconstruction method that can quickly converge.We conducted simulation experiments on the fat-tree topology.Compared with the existing solution,our solution can significantly reduce the path reconstruction time,and at the same time can reduce the demand on switch flow entries and packet header space.