On the basis of comprehensive review, this paper deeply analyzes the application of artificial intelligence technology in network diagnosis, and on this basis, focuses on the BP network fault diagnosis model with routing interface as the object. In this paper, the BP neural network diagnosis model of network interface fault is obtained by training 6 physical fault indexes and 9 running state indexes. On this basis, an intelligent network management system is designed and implemented according to the actual application after demand analysis. The system mainly includes four modules: user management, network monitoring, configuration management and fault diagnosis. The user management module is used to record the relevant information of the administrator. The configuration management module can configure network devices through web pages; Fault diagnosis module, as the focus of this paper, adopts BP network fault diagnosis model to realize network fault analysis and diagnosis; The network monitoring module displays the running parameters and status of network equipment.
This paper combines BP neural network model with modern network management technology, which has certain reference significance and value for the research and design of this kind of system.
Keywords: BP neural network; Network failure; SNMP network management
1 Introduction
1. 1 research background and significance of research purpose
China Internet Network Information Center (CNNIC, 20 18) released the 43rd Statistical Report on Internet Development in China as of 20 18 and 12. According to the report, as of 20 18 and 12, the number of netizens in China is 829 million, and it keeps increasing by more than 50 million every year. And this trend will continue in the next few years. The arrival of the 5G era will accelerate the integration of the Internet and other industries, and the network scale will inevitably increase further.
The traditional network management system is based on distributed network application system, which adopts the combination of software and hardware. SNMP protocol is the most widely used network management protocol in the field of network management at present, which unifies the way for various network devices to obtain data, and almost all network equipment manufacturers support this protocol. However, the traditional network management software based on SNMP is mostly based on C/S architecture, which has some shortcomings, such as poor expansibility and flexibility, difficult upgrade and maintenance, which brings some inconvenience to network management. Therefore, the network management system based on three-tier architecture has become a development trend. With the rapid development of Web technology, a "Web Distributed Network Management System" based on B/S (Browser/Server) architecture came into being. It is independent of specific client applications, cross-platform, convenient and easy to use, supports distributed management, and can be dynamically expanded and updated.
This paper will focus on the fault diagnosis model based on BP, and realize an intelligent network management system model with interface fault as the research object. Based on this, an intelligent network management system based on web can be designed and implemented, which can not only monitor network data in real time, but also diagnose interface faults in communication networks based on BP network fault diagnosis model, thus realizing the automation of network fault management to some extent. The system can not only ensure that network equipment provides stable and reliable network services, but also reduce the cost of maintaining network equipment in enterprises.
1.2 research status at home and abroad
Network equipment management refers to various operations and related configurations of various network equipment (such as core layer, tandem layer, access layer routing and switching equipment, servers and computers). The management server (Manager) is used to process network information, and the entity that cooperates with the management server to process and manage network information is called an agent. Managed objects refer to all resource information of devices used to provide or use network services, and various managed objects constitute a management information base. In the actual process of network management, information is exchanged between the management server and the proxy server, and between the proxy server and the managed object through standardized network management protocols (Wang He2015).
Compared with foreign network management systems and products, domestic network management systems and products started late, but with the development of internet technology, network management software developed rapidly, and many excellent network management software were born, which was widely used in the field of network management in China.
1.2. 1 research status abroad
At present, large foreign network service providers have network management systems corresponding to their products. From the initial C/S architecture to the current B/S architecture. Famous ones are: Spectrum of Cabletron Systems, CiscoWorks, OpenView, Tivoli HP, TH NetView of Tivoli Systems. These network management products are all combined with their own products to realize all the functions of network management, but the relatively professional system still adopts C/S architecture.
NetView is the most popular management software in the field of network management. NetView can monitor the network operation data in a distributed real-time manner, automatically obtain the changes of the network topology and generate the network topology. In addition, the system has a powerful historical data backup function, which is convenient for administrators to manage historical data statistics.
OpenView has good compatibility. The software integrates the advantages of various network management software, supports more protocol standards and has strong heterogeneous network management capabilities.
CiscoWorks is a product of Cisco. The software supports remote control of network devices, and administrators manage network devices through remote control terminals, providing functions such as automatic discovery, network data visualization, remote device configuration and fault management. Using the same product can provide better services, so CiscoWorks combined with other products of Cisco platform can provide more detailed services for Cisco equipment.
Cabletron's spectrum is a flexible and extensible network management platform. It adopts the methods of object-oriented and artificial intelligence, and can manage various object entities. It uses inductive model to examine different network objects and events, find their similarities and summarize their essence. At the same time, it also supports automatic device discovery and can manage network and device data in a distributed way.
1.2.2 Domestic research status
With the rapid development of home computers, the scale of network equipment is constantly expanding, and the complexity of topological structure is also increasing. In order to deal with these problems, a large number of excellent network management software came into being. For example, Nanjing Lianchuang OSS integrated network management system, Maipu Masterplan and other network management systems. Huawei's iManager U2000 network management system and the SugarNMS open source network management platform independently developed by Beijing Hezhi Communication have been widely used.
The main feature of Masterplan is that it can realize good fault diagnosis and performance management of network applications, and it is suitable for monitoring and management of servers, network devices and key applications on devices in the network.
SugarNMS has the functions of one-click automatic discovery, visual topology management, network resource management, fault management, log management, payment delivery and so on, and provides two usage modes: C/S and B/S.
IManager U2000 is located in the network element management layer and the network management layer of the telecommunication network. It adopts open, standard and unified northbound integration, which greatly shortens the integration time of OSS. The system takes business as the center and shortens the fault handling time, thus reducing the cost of enterprise fault handling.
In recent years, with the rise of artificial intelligence technology, more and more enterprises began to apply artificial intelligence technology to network management to replace the traditional centralized network management. In order to reduce the cost of enterprise network maintenance and improve the work efficiency of network managers, intelligent and automatic network management system has become the focus of many scholars.
Applicability analysis of 1.3 neural network in network management
The function of network management is to control network resources, monitor the running state of communication network and eliminate network faults. In essence, the management of network resources means that the administrator issues relevant equipment configuration commands to change the state of network equipment in order to meet business requirements, thus ensuring business stability; Monitoring the running state of the network is generally to obtain the running state of the equipment in a specified period or in real time for visualization, which is convenient for administrators to analyze whether the current equipment is running normally. Troubleshooting the network fault means that the administrator can determine the source, category, cause and solution of the fault by analyzing the running data of the network equipment, comparing it with the previous data or analyzing it according to his own experience. Troubleshooting is to analyze the characteristics of network faults found in the previous stage, get the results according to the diagnosis process, and execute specific instructions to restore the normal operation of network equipment (Hong, 20 16).
Neural network has the basic characteristics of parallelism and distributed storage, self-learning and adaptive ability, nonlinear mapping and so on. At present, the most popular neural network model is BP(Back-Propagation) neural network, which is a multi-layer feedforward neural network trained according to error back propagation and belongs to a supervised learning neural network. The model is divided into input layer, hidden layer and output layer. The stimulation of external input samples constantly changes the connection weight, and the output error is reversed layer by layer through the hidden layer in some form, so that the network output keeps approaching the expected output. Its essence is to dynamically adjust the connection weight. BP neural network has outstanding generalization ability and is good at dealing with classification problems.
BP network is a commonly used error processing method, which has been widely used in many fields. Its processing unit has the characteristics of large data volume and simple structure. Based on the physiological research results of the brain, neural network simulates some mechanisms of the brain and forms a very complex nonlinear dynamic system. When dealing with the fuzzy data and signals in the operation of network equipment, it can learn independently and get the required results. The input and output vectors in the model can be classified and connected to adapt to complex transmission and storage processing. Therefore, based on the existing network management technology, this paper will combine BP neural network to solve the network fault problem.
1.4 Main research objectives of this paper
The research purpose of this paper.
Aiming at the problems and shortcomings of traditional network management, the method based on BP neural network is explored, and the fault diagnosis model based on communication network interface is constructed. Communication network interface fault diagnosis model can effectively diagnose interface faults and identify fault types. Promote the existing network management system to be more intelligent. On this basis, the intelligent network management system based on three-tier architecture is analyzed, designed and implemented.
1.4.2 technical route
The research of intelligent network must first determine the development technical route of the system. The main process of research is to build an experimental environment on the basis of consulting relevant scientific research materials. On the premise of ensuring the normal communication of the network, collect the inflow and outflow traffic of each port, record the running state of the equipment and manage the inbound information of the equipment. At the same time, the corresponding faults in the experimental environment are arranged, including changing the port state, changing the port ip address and subnet mask, and collecting abnormal data generated in the network topology when the communication network interface fault occurs. Consult the related papers of BP neural network in fault diagnosis, build a BP neural network fault model based on the common faults and related fault documents of network communication equipment interface, and judge the effectiveness of the fault model. Gradually realize all the functions of the system. Finally, the system is tested, and the conclusion is drawn and applied to practice.
1.5 organizational structure of this paper
This paper mainly consists of six chapters, and the main contents of each chapter are as follows:
The first chapter is the introduction. Firstly, this chapter briefly introduces the current development and application status of network management system, and further analyzes the significance of establishing intelligent network management system. This paper expounds the research status of network management system at home and abroad. Finally, the purpose and organizational structure of this paper are discussed.
Chapter II Related Concepts and Related Technologies. This chapter introduces the related technologies of SNMP in detail, including SNMP organization model, SNMP management model, SNMP information model and SNMP communication model. Then the front-end framework Vue and drawing plug-in Echarts technology are introduced, and then the common fault analysis technology, expert system, neural network and so on are introduced. Finally, the basic concept and classification of neural network are briefly introduced.
The third chapter is the fault reasoning model based on BP neural network. This paper introduces the basic concept, network structure, design steps and training process of BP neural network, and introduces in detail the construction process of BP neural network fault model with interface fault as an example.
The fourth chapter is the analysis and system design of intelligent network management system. Firstly, the demand analysis is carried out. Secondly, the architecture design and overall module structure design of the system are expounded. The analysis and design of each functional module of the system are explained in detail with activity diagram. Finally, the design of the database is briefly explained.
The fifth chapter is the realization of intelligent network management system. This paper expounds the overall development process, describes the implementation process of user management module, configuration management module, equipment monitoring module and fault diagnosis module, and shows the implementation results.
Chapter VI System Test and Conclusion. Some functions and performance of the system are tested and analyzed.
The seventh chapter is the summary and prospect. Summarize the research results and existing problems of this paper, and put forward the idea of perfecting the system in the next step and the prospect of the future.
2 related concepts and related technologies
2. 1 Overview of network management
Network management is to ensure that the communication network can run stably and efficiently according to the design objectives through appropriate means and methods. It is necessary not only to accurately locate network faults, but also to predict faults in advance by analyzing data, and to reduce the incidence of faults by optimizing settings.
The five basic functions of the network management system are: configuration management, performance management, fault management, billing management and security management;
1) Configuration management: Configuration management is the most important and basic part. Relevant parameters of network communication equipment can be set to manage the managed equipment, obtain equipment information and running status regularly or in real time according to requirements, view and maintain equipment status list, generate data tables, and provide references and interfaces for administrators to change equipment configuration.
2) Performance management: Performance management is to evaluate the running state and stability of the system network. Its main work includes obtaining the data related to network performance from the managed object, making statistics and analysis on these data, establishing a model to predict the change trend and evaluating the failure risk, and modifying the network parameters through the configuration management module to ensure the optimal utilization of network resources and ensure the smooth operation of the communication network.
3) Fault management: The main function of fault management is to identify the faults in the network in time, find out the causes of the faults, and analyze and deal with the faults. Fault management is generally divided into four parts: (1) fault detection. The fault source can be found by the managed device actively sending fault information to the management station or the management station actively polling the managed device. (2) Call the police. After the management station finds the fault information, it will prompt the administrator by SMS, signal light and other means. (3) Troubleshooting. Analyze the fault information, make clear the cause and type of the fault, and find the corresponding solution. (4) Save historical fault data. Maintaining and backing up the historical fault data can provide some basis for future faults and make it more efficient to deal with network faults.
4) Billing management: The main function of billing management is to provide customers with a reasonable basis for charging, and charge customers by counting the use of customer network resources, such as calculating the cost of customer consumption flow.
5) Security management: The purpose is to ensure the smooth and safe operation of the network, avoid or resist malicious intrusion from outside, and prevent the disclosure of important data, such as the disclosure of users' personal privacy.
According to the architecture of the network management system and the basic functions defined by ISO, the basic model of the network management system based on Web is shown in the figure. The whole model includes six components: Web browser, Web server, management service set, management information base, network management protocol and managed resources.
2.2 SNMP protocol
Simple Network Management Protocol (SNMP) can be used as a protocol or a set of standards. In fact, SNMP has become an industrial standard in the field of network management. Up to now, there are eight versions of * * *, and three versions are widely used in practice, namely SNMPv 1, SNMPv2c and SNMPv3 (Tang Mingbing 20 17). The original SNMPv 1 was mainly designed to meet the needs of network management based on TCP/IP. However, with the rapid development of network management industry, the first version of SNMP protocol has not adapted to the development of network industry, and the exposure of authentication and batch data transmission makes it difficult for SNMPv 1 to support the growing network equipment. The second edition has evolved into a network management protocol running on a variety of network protocols, which has made great progress compared with the first edition. It not only provides more operation types and supports more data types, but also provides more error codes, which can distinguish errors more carefully. In addition, the supported distributed management greatly reduces the pressure on the server to some extent. However, SNMPv2c is still a plaintext transmission key, and its security needs to be improved. SNMPv3 was not officially launched until 1998. The progress of SNMPv3 is mainly reflected in its security performance. It introduces USM and VACM technology. USM adds the concepts of user name and group, and can set authentication and encryption functions to encrypt messages transmitted between NMS and agents, so as to improve their security and prevent eavesdropping. VACM determines whether and how users are allowed to access MIB objects.
2.2. 1 SNMP management model and information model
SNMP system consists of four parts: network management system (NMS), agent process agent, managed object management object and management information base (MIB). The management model diagram is as follows:
1)NMS is called network management system. As the core of the network management process, NMS sends messages to network devices through SNMP protocol, and agents receive the management messages sent by NMS to control the devices uniformly. NMS can actively send management requests to managed objects or passively accept trap messages sent by managed objects.
2)Agent is equivalent to middleware in the process of network management, and it is a kind of software, which is used to process the running data of managed devices and respond to requests from NMS, and return the results to NMS. After receiving the NMS request, the agent completes the corresponding operation by querying the MIB library and returns the data results to NMS. Agent can also be used as a middleware in the process of network management, which can not only make information respond from NMS to specific hardware devices, but also make managed devices actively send events to NMS when devices fail by configuring Trap, so that NMS can find faults in time.
3) Management object refers to the managed object. A device can be one of many managed objects, and a set of parameters configured on hardware and software in the device can be managed objects.
4)MIB is a conceptual database, which can be understood as a management object database maintained by Agent, and stores the relevant variable information of managed devices. The MIB library defines a series of attributes of the managed device: the name of the object, the state of the object, the access rights of the object and the data type of the object. By reading the value of MIB variable, the agent can query the current running status and hardware information of the managed device, so as to achieve the purpose of monitoring network devices. The agent can complete the device configuration by modifying the variable values in the MIB of the corresponding device and setting the state parameters of the managed device.
The management information base of SNMP is a tree structure, similar to DNS, with root nodes and no names. In the MIB function, each device is managed as the end of a branch of the oid tree. Each oid (object identifier) corresponds to a management object in the OID tree and is unique. Using the characteristics of tree structure, the management information stored in MIB can be read efficiently and quickly, and the nodes in the tree can be traversed, and the reading order is from top to bottom. At present, the most widely used management information base is MIB-ⅱ, which is an extension and improvement on the basis of MIB-ⅰ. Schematic diagram of MIB-II structure is shown in Figure 2.3:
(1) system group: as a basic group in MIB, it can be used to obtain basic equipment information and equipment system information.
(2) Interface group: information that defines interfaces, such as interface status and error packets, is often used for fault management and performance management.
(3) Address translation group: used for address mapping.
(4)ip group: contains information about ip, such as network number and the number of ip packets.
(5)icmp group: contains information related to icmp protocol, such as the total number of icmp messages and the number of icmp error messages input and output.
(6)tcp group: Relevant information contained in the tcp protocol, such as the number of tcp messages, retransmission time, congestion setting, etc. Which is applied to network congestion and flow control.
(7)udp group: related to udp protocol, the number of udp messages can be found, and the ip address of udp users is also saved.
(8)egp group: contains relevant information of egp protocol, such as neighbor table information and the number of autonomous systems under EGP protocol.
(9)cmot group: reserved for cmot Protocol.
(10) transmission group: reserved for transmitting information.
(1 1)snmp group: It stores information about snmp operation and implementation, such as the data amount of sending and receiving SNMP messages.
2.2.2 SNMP communication model
SNMP defines five basic data units PDU, which are used for the exchange between management process and agent process.
(1)get-request operation: the management process requests data.
(2)get-next-request operation: read the value of the next parameter from the agent process according to the MIB variable of the current operation.
(3) Setting request operation: used to set up network equipment.
(4)get-response operation: After the above three operations are successfully returned, the data will be returned to the management process. This operation is returned by the agent process to the management process.
(5)trap operation: SNMP agent actively sends trap packets to SNMP management station in asynchronous mode. Generally used for fault alarm and specific events.
SNMP message contains two parts: SNMP header and protocol data unit PDU. According to TCP/IP model, SNMP is an application layer protocol based on UDP, and UDP is based on IP protocol. Therefore, the complete schematic diagram of SNMP message is as follows:
The (1) version number indicates the SNMP version, where the size of the version field is the version number minus 1, and if it is SNMPv2, the displayed field value is 1.
(2) The community name is essentially a string. As a plaintext key, it is used to encrypt messages transmitted between the management process and the agent process. The general default setting is "public".
(3) The request ID is used for message identification. When the management process sends a message, it carries an integer value, and when the agent process returns a message, it carries an identifier. The management process can identify which proxy process returned the data through the identifier, thus finding the message corresponding to the request.
(4) Error status means that when an error occurs, a certain number of error status symbols 0-5 are filled in when the agent process returns, and this number corresponds to relevant error information. The error status descriptor is as follows:
(5) error index means that when an error in Table 2.2 above occurs during communication, the agent process sets an integer when answering the request, and the size of the integer corresponds to the offset of the error variable in the variable list.
(6) The variable name-value pair stores the variable name and the corresponding value in the form of key-value.
(7)trap message is a message sent by the agent process to the management process on its own initiative, without waiting for the next polling of the management process. Compared with SNMPv 1, the trap message format of SNMPv2 is closer to the common SNMP response message and more unified. Taking SNMPv2 as an example, the format of the trap message is as follows:
There are seven specific traps * * * defined by trap types, and the latter is customized by the supplier. The following table shows the trap types:
2.2.3 SNMP organization model
SNMP agent organization can be divided into two modes: decentralized and centralized. In the distributed model, each server corresponds to an SNMP agent, which can be understood as a one-to-one relationship, and the management station communicates with the agents on each managed server respectively.
In the centralized model, only one SNMP agent is created on the management server. The management station only communicates with the SNMP agent on the management server, and the SNMP agent receives all the data in the fixed area. As shown in Figure 2.6:
2.3 Vue
In order to realize the concept of separate development of front-end and back-end, Vue came into being. As a user interface framework, Vue.js is simple and easy to use, which makes the front-end developers do not need to search for related nodes through this complicated DOM operation, which greatly improves the development efficiency. Through the MVVM framework, view synchronization data can be automatically updated. After declaring the instance new Vue(data:data), the data in the data will be bound to its corresponding view. Once the data in the data changes, the corresponding data in the view will also change. Vue.js realizes the consistency of view and data based on MVVM framework, which is divided into three parts: model, view model and view. MVVM framework mode:
The concept of Vue.js is "everything is a component". It can be said that components are the most powerful functions of vue.js. Components can extend HTML elements and encapsulate HTML, CSS and JavaScript into reusable code components, which can be applied to different scenarios and greatly improve efficiency. Compared with traditional JavaScript, it uses virtual DOM to present pages. When the data changes, the virtual DOM structure is compared with the actual page structure, and the different parts are re-rendered to further improve the page performance.
2.4 Electronic Chart
The pure JavaScript sea chart Echarts(Enterprise Charts) developed by Baidu can run smoothly on PC and mobile devices. ECharts is compatible with the current mainstream browsers, and the bottom layer relies on the lightweight canvas library ZRender. Echarts provides intuitive, vivid, interactive and highly customized visual icons. ECharts includes the following functions:
1) rich visualization types: there are conventional charts such as histogram, line chart and pie chart, and there are also heat charts and line charts that can be used for geographic data visualization, as well as parallel coordinates for multidimensional data visualization.
2) Support for multiple data formats * * * Storage: The built-in dataset attribute in version 4.0+ supports direct transmission of two-dimensional tables.
3) Support multidimensional data: multidimensional data can be passed in.
4) Optimization of the mobile terminal: The visualization of the mobile terminal has been optimized to a certain extent, and you can zoom and translate in the coordinate system with your finger.
5) Dynamic type switching: it supports switching between different types of graphics at will, and can display unified data in column chart or line chart or from different angles.
6) Timeline: While visualizing data, it can be displayed regularly or regularly, and all data can be displayed by using the Timeline.