Application-aware sdn network resource scheduling method based on deep learning

Abstract

An application-aware SDN network resource scheduling method based on deep learning, the content of which is: based on the network characteristics of the SDN network, a deep neural network DNN is deployed on the virtual network function (VNF) located on the data plane, and the DNN processes the application data flow forwarded by the switch. Learning and classification, and reporting the classification results to the SDN controller, the SDN controller schedules network resources according to the classification results, generates routing information that meets the network resource requirements of the application data flow, and sends the routing information to the switch, the present invention The method greatly improves the flexibility and robustness of the system, realizes the reasonable scheduling of network resources according to the resource requirements of applications, and thus improves the service quality of the network.

Description

technical field [0001] The invention relates to an SDN network resource scheduling method based on deep learning to realize application perception, which belongs to the field of information technology, in particular to the field of SDN network technology. Background technique [0002] There are various access devices in the Internet of Things, and different access devices may carry different types of network applications, and these network applications have different requirements for network resources. For example, for audio over the Internet (VOIP), such applications have high requirements on network delay, and we should try to allocate low-latency paths for such applications. For video surveillance applications, it requires both a low-latency network path and a large enough bandwidth to realize real-time transmission of video data. In the IoT scenario, the key to realizing on-demand scheduling of the network is to obtain specific types of traffic information. [0003] Tr...

AI Technical Summary

Problems solved by technology

[0003] Traditional traffic scheduling mostly classifies traffic based on datagram header information. The Internet Engineering Task Force (IETF) assigns fixed port numbers to some standard protocols, and the network traffic can be divided into Different application categories, but there are many problems in the classification based on port numbers: First, with the increase of network applications, many application layer protocols do not allocate special port numbers, and these application protocols cannot be distinguished by port numbers; Some application layer protocols may carry many different types of application content, and different application content has completely different requirements for the network.

DPI technology solves the problem of traditional traffic identification based on packet header fields to a certain extent, but it also has many problems: (1) Poor scalability: due to the lag of this method for traffic identification of new P2P applications, that is, when the signature database is not upgraded New applications cannot be detected before, and the load characteristics of the new application must be found before the application can be effectively detected

This point becomes the bottleneck limiting the method

(2) Lack of encrypted data analysis function: some application loads are encrypted for transmission, which hides the application protocol and data characteristics, so deep packet inspection technology has very limited detection capabilities for encrypted applications

(3) High cost: due to the need to complete operations such as protocol analysis and restoration and feature matching, the calculation and storage overhead is large, and the performance of the traffic detection algorithm is low

The more complex the load characteristics, the higher the detection cost and the worse the algorithm performance

Images