Data center network flow splicing method based on deep learning

Abstract

The invention discloses a data center network flow splicing method based on deep learning. The method comprises the following steps: initializing a twin neural network model for network flow splicing;selecting representative flow characteristics according to the flow information acquired in a period of time to form a sample, and training the twin neural network model by using the sample; and extracting flow characteristics of flows needing to be spliced, inputting the flow characteristics into the trained twin neural network model, determining flows which can be spliced together, and completing flow splicing. According to the invention, the data center network traffic feature selection method is optimized, the representativeness and robustness of the selected features are improved, the network traffic splicing model is constructed by using the deep learning algorithm, and the traffic splicing precision is improved.

Description

technical field [0001] The invention belongs to the field of data center networks, and in particular relates to a data center network traffic splicing method. Background technique [0002] With the rise of cloud computing, data centers have developed rapidly in recent years, and their number and scale are increasing rapidly. It is estimated that by 2021, the number of hyperscale data centers in the world will reach 628, an increase of 53% compared to 2016. The data center is a service platform with complete equipment (such as access bandwidth, network, computer room environment, etc.), professional management, and many applications. It has massive computing resources and storage resources to provide users with on-demand services. As an important part of the data center, the data center network connects all computing and storage resources, so that the data center can provide users with the services they need, and provide corresponding QoS to ensure the quality of service. The...

AI Technical Summary

Problems solved by technology

There are a large number of network function devices in the current data center network. These network function devices are designed to regulate the traffic in the network to optimize the overall performance of the network. Packets are sent to different links to adjust the link load and prevent network congestion. These network function devices will not only change the five-tuple information of the traffic, but also affect the characteristic distribution of the data packets. For example, when the traffic passes through the tunnel node, the data Packets are encapsulated, so that the size of the data packet will change. For example, the load balancing device will reshape the data packet when adjusting the link load, resulting in changes in the size and number of data packets. When these characteristics change, the active association method makes the same The characteristic distribution of traffic is similar, but after passing through the network function node, these characteristic distributions will be changed, making it impossible to identify

At the same time, due to the existence of network function nodes, different flows have similar feature distributions or feature distributions have greatly changed, resulting in passive association methods that cannot be spliced ​​accurately

[0007](2) The existing traffic splicing method is relatively coarse-grained

The classification-based traffic splicing method mainly divides traffic into different categories through classifiers. Common categories include web application traffic, audio and video traffic, FTP file transfer traffic, email traffic, and network chat traffic. The method generally only divides network traffic into different application categories according to characteristics, but there are still a large number of traffic from different users in the same application category, and it is difficult to distinguish these flows again, that is, it is impossible to determine which tenant the traffic in the same category belongs to , it is impossible to effectively splice traffic and obtain the path of traffic in the network

[0008]Therefore, the existing traffic splicing method applied to the data center network traffic splicing still has great limitations. Unable to meet the high-precision and fine-grained requirements of data center network traffic splicing

Images