Abnormal flow monitoring system of DPU

By deploying the DPU abnormal traffic monitoring system in Istio, using the traffic detection model to identify and deal with abnormal traffic, the problems of misreport and false alarms in the existing technology are solved, and the accurate identification and flexible adaptation of abnormal traffic is achieved, and network security is improved. processing capability.

CN119094170APending Publication Date: 2024-12-06YUSUR TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411113087.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing Istio security mechanism does not perform well in dealing with new threats and unknown attacks, and is prone to missed and false alarms, and cannot accurately monitor abnormal traffic.

Method used

A DPU's abnormal traffic monitoring system is designed. By deploying a data processor DPU in each host and deploying a traffic detection model on it, it captures business traffic, identifies abnormal traffic, and sends it to the monitoring platform, according to exceptions Traffic data is restricted or fuseed, and the training and aggregation technology of the flow detection model is used to improve detection accuracy and flexibility.

Benefits of technology

It realizes accurate identification of abnormal traffic, reduces the rate of missed and false alarms, adapts to more application scenarios, and adapts to different abnormal scenarios through traffic control, improving network security processing capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094170A_ABST
    Figure CN119094170A_ABST
Patent Text Reader

Abstract

The invention discloses a DPU (Data Processing Unit) abnormal traffic monitoring system, which comprises a plurality of hosts, a monitoring platform and at least one security control device, and is characterized in that a data processor DPU is deployed in each host, and a traffic detection model is deployed in each data processor DPU; the traffic detection model is used for capturing the service traffic passing through the traffic interface of the data processor DPU, outputting a traffic detection result, and sending the input service traffic as abnormal traffic data to the monitoring platform when the traffic detection result indicates that the traffic is abnormal; and the safety control device is used for acquiring at least part of abnormal traffic data from the monitoring platform, determining a target data processor DPU according to the abnormal traffic data, and carrying out traffic limitation on the target data processor DPU. According to the method and the device, the traffic is captured and analyzed when the service traffic is generated by the DPU deployed in each host through the traffic detection model, so that the abnormal traffic data can be accurately identified, and the abnormal traffic data is not easy to generate missing report and false report.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and in particular to an abnormal traffic monitoring system for a DPU. Background Art

[0002] Istio is an open source service mesh that transparently layers onto existing distributed applications. Istio's powerful features provide a unified and more efficient way to secure, connect, and monitor services. Istio is the path to load balancing, service-to-service authentication, and monitoring.

[0003] Currently, mTLS (two-way TLS) and JWT (JSON Web Token) are mainly used in Istio's security mechanism. mTLS relies on public key infrastructure (PKI) to verify the identities of both communicating parties through certificate signing. The certificate itself is signed by a certification authority (CA), which guarantees the legitimacy of the certificate. JWT contains a signature part that verifies the authenticity and integrity of the token. The signature is generated by the issuer who holds the key, and the recipient can verify the signature using the public key.

[0004] However, the above security mechanisms usually rely on technologies such as rule engines and signature-based detection, and their data filtering and identification capabilities are relatively rigid. They perform poorly in dealing with new threats and unknown attack types, are prone to missed reports and false reports, and cannot accurately monitor abnormal traffic. Summary of the invention

[0005] In view of this, the present application provides a DPU abnormal traffic monitoring system to solve the problem that traditional solutions perform poorly in abnormal traffic monitoring.

[0006] In a first aspect, the present application provides an abnormal traffic monitoring system for a DPU, comprising: multiple hosts, a monitoring platform and at least one security control device, wherein a data processor DPU is deployed in each of the hosts, and a traffic detection model is deployed in each of the data processor DPU; the traffic detection model is used to capture the business traffic passing through the traffic interface of the data processor DPU and output the traffic detection result, and when the traffic detection result indicates an abnormality, the input business traffic is sent to the monitoring platform as abnormal traffic data; the security control device is used to obtain at least part of the abnormal traffic data from the monitoring platform, determine the target data processor DPU based on the abnormal traffic data, and limit the traffic of the target data processor DPU.

[0007] Optionally, the system also includes: at least one control platform; the safety control device is used to generate a flow limitation configuration or a flow fuse configuration based on the abnormal flow data, and send the flow limitation configuration or the flow fuse configuration to the control platform corresponding to the target data processor DPU; the control platform is used to limit or fuse the flow of the target data processor according to the flow limitation configuration or the flow fuse configuration.

[0008] Optionally, a data interaction component is deployed in each of the data processors DPU; the control platform is used to configure the data interaction component in the target data processor DPU according to the traffic limitation configuration or the traffic fuse configuration, so that the traffic of the data interaction component in the target data processor DPU is restricted or fuse-cut.

[0009] Optionally, the control platform is used to limit the business traffic of the target data processor DPU to output or reception through the data interaction component included therein according to the traffic fusing configuration.

[0010] Optionally, the traffic detection model is trained in the following manner: collecting a data set about business traffic; performing feature extraction on the data set to obtain a training set; selecting at least part of the training set from the training set, inputting a basic network model, and obtaining multiple pre-trained models; aggregating at least two pre-trained models through an aggregation module to obtain multiple aggregated models; aggregating the multiple aggregated models through a model generation module to obtain the traffic detection model.

[0011] Optionally, during the training process of the traffic detection model, feature extraction is performed on the data set to obtain a training set, including: extracting the source IP address, destination IP address, source port number, destination port number, protocol type and data packet size of the business traffic in the data set as feature values, and digitizing and normalizing the feature values ​​to obtain the training set.

[0012] Optionally, during the training process of the traffic detection model, the training process of the traffic detection model further includes: before inputting at least part of the training set into the basic network model, defining the training model category as supervised learning or unsupervised learning.

[0013] Optionally, during the training process of the traffic detection model, at least two pre-trained models are aggregated through an aggregation module to obtain multiple aggregated models, including: assigning weights to the multiple pre-trained models based on at least one of the training iteration number, model accuracy and training start time of the pre-trained models received by the aggregation module, wherein at least one of the training iteration number, model accuracy and training start time is positively correlated with the weight; and aggregating the multiple pre-trained models according to the weight to obtain multiple aggregated models.

[0014] Optionally, during the training process of the traffic detection model, the multiple pre-trained models are aggregated according to the weights to obtain multiple aggregate models, including: when at least two of the pre-trained models arrive at the receiving queue of the aggregation module, the arriving pre-trained models are aggregated by the aggregation module according to the weights of the arriving pre-trained models to obtain a first aggregate model, and the first aggregate model obtained after aggregation is sent to the model generation module through the sending queue of the aggregation module; after obtaining the Nth sub-aggregation model, if at least one of the pre-trained models arrives at the receiving queue of the aggregation module, the arriving pre-trained model and the Nth aggregate model are aggregated by the aggregation module according to the weights of the arriving pre-trained models to obtain the N+1th aggregate model, and the N+1th aggregate model obtained after aggregation is sent to the model generation module through the sending queue of the aggregation module.

[0015] Optionally, during the training process of the traffic detection model, the multiple aggregation models are aggregated through the model generation module to obtain the traffic detection model, including: when at least two of the aggregation models reach the model generation module, the aggregation models are aggregated through the model generation module to obtain a first traffic detection model; after obtaining the Mth traffic detection model, if at least one of the aggregation models reaches the model generation module, the received aggregation model and the Mth traffic detection model are aggregated through the model generation module to obtain the M+1th traffic detection model; if there is no aggregation model that has not arrived, the Mth traffic detection model is determined as the traffic detection model, where M is a positive integer.

[0016] The present application uses a traffic detection model to capture and analyze the traffic when the data processor DPU deployed in each host generates business traffic. It can accurately identify abnormal traffic data. The traffic detection model has high flexibility and can also be updated. Therefore, it is not easy to miss or misreport abnormal traffic data, and can adapt to more application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is a schematic diagram of an abnormal flow monitoring system of a DPU according to an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of an abnormal flow monitoring system of a DPU according to another embodiment of the present application;

[0020] Figure 3 is a schematic diagram of an abnormal flow monitoring system of a DPU according to another embodiment of the present application;

[0021] Figure 4 This is a flow chart of the steps of a flow detection model training method according to an embodiment of the present application;

[0022] Figure 5 It is an exemplary system suitable for the traffic detection model training method of the embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.

[0024] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0025] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0026] The technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application. In the absence of conflict, the following embodiments and their technical features can be combined with each other.

[0027] Istio is an open source service mesh. In Istio's security mechanism, mTLS (two-way TLS) and JWT (JSON Web Token) are mainly used. mTLS relies on public key infrastructure (PKI) to verify the identities of both communicating parties through certificate signing. The certificate itself is signed by a certification authority (CA), which guarantees the legitimacy of the certificate. JWT contains a signature part that verifies the authenticity and integrity of the token. The signature is generated by the issuer who holds the key, and the recipient can verify the signature using the public key.

[0028] However, traditional methods perform poorly in dealing with new threats and unknown attack types, and are prone to missed reports and false positives. It is impossible to cope with complex and changing threat environments, and the rules and signatures of traditional methods are usually relatively solidified, making it difficult to cope with complex and changing threat environments. With the continuous evolution and changes in network attack technology, the rules and signature libraries of traditional methods need to be constantly updated and maintained, and the management cost is high. And it is vulnerable to attack and bypass. The rules and signatures of traditional methods are usually public, and attackers can bypass detection by analyzing rules and features. Malicious attackers can take various measures to circumvent traditional detection methods, such as modifying attack payloads, using encrypted communications, etc. Moreover, the big data processing capacity is insufficient, and traditional methods usually have poor performance when processing large-scale data, and it is difficult to cope with the processing requirements of high-speed networks and massive data packets. For network security scenarios that require large-scale data analysis and mining, the processing capacity of traditional methods is limited. Therefore, the present application proposes an abnormal traffic monitoring system for DPU to at least partially solve the above problems.

[0029] Figure 1 FIG. 1 is a schematic diagram of an abnormal flow monitoring system of a DPU according to an embodiment of the present application. Figure 1As shown, the abnormal traffic monitoring system 100 of DPU includes: multiple hosts 101, a monitoring platform 102 and at least one security control device 103, a data processor DPU 104 is deployed in each host 101, and a traffic detection model 105 is deployed in each data processor DPU 104.

[0030] The traffic detection model 105 is used to capture the service traffic passing through the traffic interface 1041 of the data processor DPU 104 and output the traffic detection result. When the traffic detection result indicates an abnormality, the input service traffic is sent as abnormal traffic data to the monitoring platform 102. The security control device 103 is used to obtain at least part of the abnormal traffic data from the monitoring platform 102, determine the target data processor DPU according to the abnormal traffic data, and limit the traffic of the target data processor DPU.

[0031] When service traffic is generated in the communication network composed of multiple data processors DPU104, for example, host A sends a data request to host B, a request service will be generated in host A, and then the request service will reach the data processor DPU A deployed in host A, and the request service will be sent to the traffic interface B of the data processor DPU B deployed in host B through the traffic interface A of the data processor DPU A. In this process, the traffic detection model A deployed in the data processor DPU A will capture the data traffic passing through the traffic interface A of the data processor DPU A and analyze it. If the traffic detection result output by the traffic detection model A indicates an abnormality, the captured data traffic will be sent to the monitoring platform 102 as abnormal traffic data. After the traffic interface B receives the request service, the traffic detection model B deployed in the data processor DPU B will capture the data traffic passing through the traffic interface B of the data processor DPUB and analyze it. If the traffic detection result output by the traffic detection model B indicates an abnormality, the captured data traffic will be sent to the monitoring platform 102 as abnormal traffic data. After that, the data processor DPU B will send the request service to the host B.

[0032] The monitoring platform 102 will store abnormal flow data, and the monitoring platform 102 can send the abnormal flow data to the alarm center 200, and the alarm center 200 executes the internally stored alarm strategy according to the received abnormal flow data. In addition, since at least one safety control device 103 can be set in the abnormal flow monitoring system 100 of the DPU, at least part of the abnormal flow data will be obtained from the monitoring platform 102. For example, if only one safety control device 103 is set in the abnormal flow monitoring system 100 of the DPU, the safety control device 103 will obtain all abnormal flow data in the monitoring platform 102, but multiple safety control devices 103 can also be set in the abnormal flow monitoring system 100 of the DPU, and each safety control device 103 is responsible for flow control of part of the data processor DPU104. In this scenario, each safety control device 103 will only obtain abnormal flow data related to the data processor DPU104 for which it is responsible from the monitoring platform 102. After obtaining the abnormal flow data, the safety control device 103 will determine the target data processor DPU that generates the abnormal flow data according to the abnormal flow data, and then perform flow control on the target data processor DPU.

[0033] In an embodiment of the present application, the traffic detection model 105 is used to capture and analyze the data processor DPU104 deployed in each host 101 when it generates business traffic, so that abnormal traffic data can be accurately identified. In addition, the traffic detection model 105 has high flexibility and can also be updated. Therefore, it is not easy to generate underreporting and false alarms for abnormal traffic data, and can adapt to more application scenarios.

[0034] In one possible implementation, the abnormal traffic monitoring system 100 of the DPU includes: at least one control platform 106 .

[0035] The safety control device 103 is used to generate a flow restriction configuration or a flow fuse configuration according to the abnormal flow data, and send the flow restriction configuration or the flow fuse configuration to the control platform 106 corresponding to the target data processor DPU. The control platform 106 is used to restrict or fuse the flow of the target data processor DPU according to the flow restriction configuration or the flow fuse configuration.

[0036] The abnormal flow monitoring system 100 of the DPU may also include at least one control platform 106, such as Istiod. The control platform 106 communicates with at least some of the data processors DPU104 in the abnormal flow monitoring system 100 of the DPU. If only one control platform 106 is set in the abnormal flow monitoring system 100 of the DPU, the control platform 106 will communicate with all the data processors DPU104 in the abnormal flow monitoring system 100 of the DPU. However, multiple control platforms 106 can also be set in the abnormal flow monitoring system 100 of the DPU, and each control platform 106 is responsible for flow control of some data processors DPU104. In this scenario, each control platform 106 will only communicate with the data processor DPU104 for which it is responsible.

[0037] The safety control device 103 generates a traffic limitation configuration or a traffic fuse configuration based on the acquired abnormal traffic data, and sends it to the control platform 106 that communicates with the target data processor DPU corresponding to the abnormal traffic data. The control platform 106 will limit the traffic of the target data processor according to the traffic limitation configuration, for example, limiting the traffic of the target data processor to less than 50kb / s, or fuse the traffic of the target data processor according to the traffic fuse configuration, so that the target data processor cannot send or receive business traffic.

[0038] It should be noted that if Figure 2 As shown, the security control device 103 and the control platform 106 can not only be set outside the host 101, but also be set in a certain host 101 respectively, or the security control device 103 and the control platform 106 can be set in the same host 101.

[0039] In an embodiment of the present application, the control platform 106 limits or fuses the flow of the target data processor DPU according to the flow limitation configuration or the flow fuse configuration, and can control the flow of the target data processor DPU according to the abnormal scenarios indicated by different abnormal flow data to adapt to different abnormal scenarios.

[0040] Figure 3 is a schematic diagram of an abnormal flow monitoring system of a DPU according to another embodiment of the present application. Figure 3 As shown, a data interaction component 1042 is deployed in each data processor DPU104.

[0041] The control platform 106 is used to configure the data interaction component 1042 in the target data processor DPU according to the flow restriction configuration or the flow fusing configuration, so that the flow of the data interaction component 1042 in the target data processor DPU is restricted or fusing.

[0042] A data interaction component 1042, such as Envoy, is deployed in each data processor DPU104 to regulate all inbound and outbound traffic of most services in the data processor DPU104. The control platform 106 can configure the data interaction component 1042 in the target data processor DPU according to the traffic restriction configuration or the traffic fuse configuration, so that the traffic of the data interaction component 1042 in the target data processor DPU is restricted or fused.

[0043] It should be noted that not all business traffic in the data processor DPU104 is transmitted to the traffic interface 1041 via the data interaction component 1042, and some business traffic will directly reach the traffic interface 1041. Therefore, the control platform 106 can restrict the business traffic in the target data processor DPU according to the traffic fusing configuration, forcing all business traffic to be output or received via the data interaction component 1042, and then fusing the data interaction component 1042, thereby ensuring that the traffic of the target data processor DPU is fusing.

[0044] In the embodiment of the present application, the control platform 106 configures the data interaction component 1042 deployed in the data processor DPU 104 to improve the efficiency of flow control.

[0045] Figure 4 is a flow chart of the steps of the flow detection model training method according to an embodiment of the present application. Figure 4 The traffic detection model training method includes the following steps:

[0046] Step 401: Collect data sets about business traffic.

[0047] Step 402: extract features from the data set to obtain a training set.

[0048] Step 403: Select at least part of the training set from the training set, input the basic network model, and obtain multiple pre-trained models.

[0049] Step 404: Aggregate at least two pre-trained models through an aggregation module to obtain multiple aggregated models.

[0050] Step 405: Aggregate multiple aggregation models through a model generation module to obtain a traffic detection model.

[0051] In order to train the traffic detection model 105, firstly, a data set about the business traffic is collected, and then the data set can be cleaned. After filtering out dirty data, features are extracted from the data set to obtain a training set for training the traffic detection model 105. In the process of model training, a distributed architecture client-leader-server can be used for training, and at least part of the training set is selected from the training set to input multiple basic network models of the client layer, thereby obtaining multiple pre-trained models. After that, after receiving the multiple pre-trained models sent by the client layer, the leader layer aggregates at least two pre-trained models through the aggregation module to obtain multiple aggregated models. Finally, after receiving the multiple aggregated models sent by the leader layer, the server layer aggregates the multiple aggregated models through the model generation module to obtain a traffic detection model.

[0052] The tree information status can be transmitted between the client layer, leader layer and server layer. If an unexpected situation occurs during training, such as network interruption or computing resource failure, the model status can be restored by loading the tree information to avoid data loss during training and restart training.

[0053] During the model training process, part of the training set that is not selected for training the model can be used as a validation set to verify the training results and perform appropriate optimization.

[0054] Specifically, step 402 may include: extracting the source IP address, destination IP address, source port number, destination port number, protocol type and data packet size of the business traffic in the data set as feature values, and digitizing and normalizing the feature values ​​to obtain a training set.

[0055] In addition, before inputting at least part of the training set into the basic network model, the training model category can be defined as supervised learning or unsupervised learning, thereby limiting the category of model training.

[0056] It should be noted that an exemplary system suitable for the flow detection model training method can be as follows: Figure 5 shown.

[0057] Different at least partial training sets are respectively input into the basic network models in multiple client layers. After the training is completed, the client layer can send the pre-trained model to the receiving queue of the leader layer. When there are at least two pre-trained models in the receiving queue of the leader layer, the pre-trained model is aggregated by the aggregation module, and the aggregated model obtained after aggregation is sent to the receiving queue of the server layer through the sending queue. When there are at least two aggregated models in the receiving queue of the server layer, the pre-trained model is aggregated by the model generation module until the leader layer has sent all the aggregated models, and then the traffic detection model 105 is generated.

[0058] In the embodiment of the present application, the traffic detection model 105 is obtained through distributed architecture training, which can make the performance of the traffic detection model 105 more accurate.

[0059] In one possible implementation, at least two pre-trained models are aggregated through an aggregation module to obtain multiple aggregated models, and the process may also include: assigning weights to the multiple pre-trained models based on at least one of the training iteration number, model accuracy and training start time of the pre-trained models received by the aggregation module, and aggregating the multiple pre-trained models according to the weights to obtain multiple aggregated models.

[0060] The aggregation algorithm used in the leader layer can adopt an outdated function to mitigate errors caused by outdated models, i.e., weight them by staleness, and can refer to the number of model iterations and the accuracy of the model when training the model in the client layer to weight them, and add a regularization term to penalize unbalanced data and unstable device problems, that is, at least one of the training iteration number, model accuracy and training start time of the pre-trained model is positively correlated with the weight.

[0061] The model accuracy of the pre-trained model can be obtained by verifying the pre-trained model, and the training start time of the pre-trained model can be represented by a timestamp.

[0062] For example, the weight can be calculated as follows:

[0063] wi=a*P+b*Q+c*R

[0064] wi is used to represent the weight, P is used to represent the training iteration number of the pre-trained model, Q is used to represent the model accuracy of the pre-trained model, R is used to represent the training start time, and a, b, and c are positive numbers.

[0065] Specifically, the process of aggregating multiple pre-trained models according to weights to obtain multiple aggregated models may also include:

[0066] When at least two pre-trained models arrive at the receiving queue of the aggregation module, the arriving pre-trained models are aggregated by the aggregation module according to the weights of the arriving pre-trained models to obtain a first aggregated model, and the first aggregated model obtained after aggregation is sent to the model generation module through the sending queue of the aggregation module. After obtaining the Nth sub-aggregate model, if at least one pre-trained model arrives at the receiving queue of the aggregation module, the arriving pre-trained model and the Nth aggregated model are aggregated by the aggregation module according to the weights of the arriving pre-trained models to obtain the N+1th aggregated model, and the N+1th aggregated model obtained after aggregation is sent to the model generation module through the sending queue of the aggregation module.

[0067] When the leader layer aggregation module aggregates at least two pre-trained models according to the weights during the first aggregation, obtains the first aggregated model, and sends it to the model generation module. In the subsequent aggregation process, as long as a pre-trained model arrives at the receiving queue of the aggregation module, the aggregation module will aggregate the Nth aggregated model obtained in the last aggregation process with the received pre-trained model, and send it to the model generation module again, until the client layer no longer sends the pre-trained layer.

[0068] It should be noted that the receiving queue and sending queue of the aggregation module are asynchronous queues, and their threads are not mutually exclusive.

[0069] In an embodiment of the present application, the model arrival time of the pre-trained model received by the aggregation module is used to assign a weight to each pre-trained model, which can improve the stability and performance of the global model.

[0070] In a possible implementation, a process of aggregating multiple aggregation models through a model generation module to obtain a traffic detection model may also include:

[0071] When at least two aggregation models arrive at the model generation module, the aggregation models are aggregated through the model generation module to obtain the first flow detection model. After obtaining the Mth flow detection model, if at least one aggregation model arrives at the model generation module, the received aggregation model and the Mth flow detection model are aggregated through the model generation module to obtain the M+1th flow detection model. If there is no aggregation model that has not arrived, the Mth flow detection model is determined as the flow detection model, where M is a positive integer.

[0072] The recv thread of the server layer collects the aggregation models sent by each leader layer, and then aggregates the aggregation models through the model generation module. In the first aggregation process, at least two aggregation models that have arrived will be aggregated to obtain the first traffic detection model. In the subsequent aggregation process, the model generation module will aggregate the Mth aggregation model obtained in the last aggregation process with the received aggregation model again until the leader layer has no aggregation model that has not reached the server layer, and the result of the last aggregation is determined as the traffic detection model.

[0073] When the aggregation model is aggregated at the server layer, similar to the aggregation algorithm used by the leader layer, weights are added based on at least one of the training iteration number, model accuracy, and training start time of the aggregation model.

[0074] In an embodiment of the present application, the model generation module starts aggregation when at least two aggregation models are received, and aggregates the previous aggregation result again according to the subsequently arrived aggregation model. This can improve the model generation efficiency while ensuring the accuracy of the aggregated traffic detection model.

[0075] The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as software or computer code that can be stored in a recording medium (such as CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as a computer code originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware (such as ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (for example, RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a special-purpose computer for executing the method shown here. Although the present application has been shown and described with respect to one or more implementations, those skilled in the art will think of equivalent deformations and modifications based on the reading and understanding of this specification and the accompanying drawings. The present application includes all such modifications and variations, and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-described components, the terms used to describe such components are intended to correspond to any component that performs the specified function of the component (e.g., which is functionally equivalent) (unless otherwise indicated), even if not structurally equivalent to the disclosed structure that performs the function in the exemplary implementation of the present specification shown herein.

[0076] That is, the above description is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the specification and drawings of the present application, such as the mutual combination of technical features between the embodiments, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

[0077] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0078] In order to enable any technician in the field to implement and use the application, the application provides the above description. In the above description, various details are listed for the purpose of explanation. It should be understood that those of ordinary skill in the art can recognize that the application can also be implemented without using these specific details. In other embodiments, the well-known process will not be elaborated in detail to avoid unnecessary details that make the description of the application obscure. Therefore, the application is not intended to be limited to the embodiments shown, but is consistent with the widest range of principles and features disclosed in the application.

[0079] It should be noted that, under the premise of no conflict, the various embodiments and / or the technical features in the various embodiments described in this application can be arbitrarily combined with each other, and the technical solution obtained after the combination should also fall within the protection scope of this application.

[0080] It should be understood that the specific examples in the embodiments of the present application are only to help those skilled in the art to better understand the embodiments of the present application, rather than to limit the scope of the embodiments of the present application. Those skilled in the art can make various improvements and modifications based on the above embodiments, and these improvements or modifications all fall within the scope of protection of the present application.

[0081] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A DPU abnormal traffic monitoring system, comprising: Multiple hosts, a monitoring platform and at least one security control device, a data processor DPU is deployed in each of the hosts, and a traffic detection model is deployed in each of the data processors DPU; The traffic detection model is used to capture the business traffic passing through the traffic interface of the data processor DPU and output the traffic detection result. When the traffic detection result indicates an abnormality, the input business traffic is sent to the monitoring platform as abnormal traffic data; The safety control device is used to obtain at least part of the abnormal traffic data from the monitoring platform, determine the target data processor DPU according to the abnormal traffic data, and limit the traffic of the target data processor DPU.

2. The system according to claim 1, characterized in that The system further comprises: at least one control platform; The safety control device is used to generate a flow restriction configuration or a flow fuse configuration according to the abnormal flow data, and send the flow restriction configuration or the flow fuse configuration to a control platform corresponding to a target data processor DPU; The control platform is used to limit or fuse the flow of the target data processor DPU according to the flow limitation configuration or the flow fusing configuration.

3. The system according to claim 2, characterized in that A data interaction component is deployed in each of the data processors DPU; The control platform is used to configure the data interaction component in the target data processor DPU according to the flow restriction configuration or the flow fusing configuration, so that the flow of the data interaction component in the target data processor DPU is restricted or fused.

4. The system according to claim 3, characterized in that The control platform is used to limit the business traffic of the target data processor DPU to output or receive through the data interaction component included therein according to the traffic fusing configuration.

5. The system according to claim 1, characterized in that The traffic detection model is trained in the following way: Collect data sets about business traffic; Performing feature extraction on the data set to obtain a training set; Select at least part of the training set from the training set, input the basic network model, and obtain multiple pre-training models; Aggregating at least two pre-trained models through an aggregation module to obtain multiple aggregated models; The multiple aggregation models are aggregated through a model generation module to obtain the traffic detection model.

6. The system according to claim 5, characterized in that In the training process of the traffic detection model, extracting features from the data set to obtain a training set includes: The source IP address, destination IP address, source port number, destination port number, protocol type and data packet size of the business traffic in the data set are extracted as feature values, and the feature values ​​are digitized and normalized to obtain the training set.

7. The system according to claim 5, characterized in that During the training process of the flow detection model, the training process of the flow detection model further includes: Before inputting at least part of the training set into the basic network model, the training model category is defined as supervised learning or unsupervised learning.

8. The system according to claim 5, characterized in that In the training process of the traffic detection model, the at least two pre-trained models are aggregated by the aggregation module to obtain multiple aggregated models, including: Assigning weights to the multiple pre-trained models according to at least one of the number of training iterations, the model accuracy, and the training start time of the pre-trained models received by the aggregation module, wherein the number of training iterations, the model accuracy, and the training start time are positively correlated with the weights; The multiple pre-trained models are aggregated according to the weights to obtain multiple aggregated models.

9. The system according to claim 8, characterized in that In the training process of the traffic detection model, the multiple pre-trained models are aggregated according to the weights to obtain multiple aggregated models, including: When at least two of the pre-trained models arrive at the receiving queue of the aggregation module, the aggregation module aggregates the arrived pre-trained models according to the weights of the arrived pre-trained models to obtain a first aggregated model, and sends the first aggregated model obtained after aggregation to the model generation module through the sending queue of the aggregation module; After obtaining the Nth sub-aggregation model, if at least one of the pre-trained models arrives at the receiving queue of the aggregation module, the aggregation module aggregates the arriving pre-trained model and the Nth aggregation model according to the weight of the arrived pre-trained model to obtain the N+1th aggregation model, and sends the N+1th aggregation model obtained after aggregation to the model generation module through the sending queue of the aggregation module.

10. The system according to claim 5, characterized in that In the training process of the traffic detection model, the multiple aggregation models are aggregated by the model generation module to obtain the traffic detection model, including: When at least two of the aggregated models arrive at the model generation module, the aggregated models are aggregated by the model generation module to obtain a first traffic detection model; After obtaining the Mth flow detection model, if at least one of the aggregation models reaches the model generation module, the received aggregation model and the Mth flow detection model are aggregated by the model generation module to obtain the M+1th flow detection model. If there is no aggregation model that has not arrived, the Mth flow detection model is determined as the flow detection model, where M is a positive integer.