The Research On Network Traffic Matrix Measurement Method Based On SDN

The traffic matrix represents the size of the flow between any two nodes in the network. The traffic matrix is the key input information for many network management tasks, such as traffic engineering, network planning, network performance diagnosis analysis and traffic accounting and so on. Due to the importance of traffic matrix, the traffic matrix measurement is widely concerned by many researchers at home and abroad in recent years. In the actual network, it is very difficult to directly measure the traffic matrix because of the limitation of network measurement resource and capacity. Therefore, the current traffic matrix measurement method commonly used by the research community is based on a few measurements that are easy to measure directly(eg. the link load) or can be directly obtained(eg. business routing) to estimate the network traffic matrix. However, due to the information that can be directly measured or obtained is very few in traditional networks, and the number of flows is huge, there are a lot of errors in estimated traffic matrix.On the other hand, Software Defined Network(SDN) has aroused widespread attention in academia and industry. SDN network enables the separation of the control plane from the data plane. The centralized control plane is running on the network controller, while the data plane is scattered in various devices. The network traffic matrix measurement benefits from the separation strucure of SDN. First of all, the centralized control plane has a global network view and can be unified deployment of network resource. Secondly, the data plane distributed on the network devices provides several counters for flow statistics that we can provide more input for the traffic matrix estimation according to. This paper will mainly study how to better measure traffic matrix using the features provided by SDN network.Aiming at the generalized traffic matrix, its each row represents the size of a flow between two nodes in different periods and its each column represents the size of all flows in a period. The generalized traffic matrix has the temporal and spatial correlation. Therefore, we can only directly measure the size of a small part of flows, and then use the matrix completion technique to estimate the size of other flows. However, which flows to measure directly has a great impact on the accuracy of the final estimated traffic matrix. In order to reduce the inherent complexity in the design of the optimal observation matrix which represents a set of flows that need to be measured directly, this paper presents a method based on random search to design the optimal observation matrix. The random search method uses genetic algorithm and particle swarm algorithm, whose optimization goal is the error of the final estimated traffic matrix, to search and design the optimal observation matrix.On the other hand, the traffic matrix can be estimated by using the counters in the flow table entries of SDN switches. The counter in a flow table entry of the SDN switch represents the sum of the size of all flows that can match this entry. Due to a limited number of TCAM entries, which flows to aggregate by the flow table entries of SDN switches is a great impact on the accuracy of the estimated traffic matrix. Therefore, it is very careful to design the flow table entries. The traffic measurement rule is defined as the matching rule that a TCAM entry can match which flows according to, namely a TCAM entry corresponds to a rule. This paper presents two methods(MLRF and LFF) to design the traffic measurement rules. Both methods satisfy the flow aggregation constraints(determined by associated routing policies) and do not change the routing of flows. They both have low complexity.