Classification of unknown network traffic

A network device using DPI and ML classification techniques addresses the challenge of encrypted traffic by identifying unknown applications with specific types and identities, enhancing security policy management and reducing network vulnerabilities.

CN119094240APending Publication Date: 2024-12-06HEWLETT PACKARD ENTERPRISE DEV LP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411486023.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2019-05-17
Filing Date
2019-09-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing network security systems struggle with encrypted network traffic, as deep packet inspection (DPI) methods classify a significant portion of traffic as unknown applications, leading to ineffective security policies that either block legitimate traffic or expose networks to threats, causing interruptions and vulnerabilities.

Method used

Implementing a network device that combines deep packet inspection (DPI) and machine learning (ML) classification techniques to identify unknown applications with specific types and identities, allowing for fine-grained management of security policies and updating policies when applications are identified.

Benefits of technology

Enhances network security by enabling precise application classification, reducing computing resources, and improving network reliability by applying tailored security policies to unknown applications, thus minimizing interruptions and vulnerabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094240A_ABST
    Figure CN119094240A_ABST
Patent Text Reader

Abstract

The invention relates to classification of unknown network traffic. A network device may receive network traffic for an application. The network device may determine a first classification for network traffic according to a first classification technique. The first classification may identify network traffic as being related to a particular application or an unknown application. The network device may determine a second classification for the network traffic according to a second classification technique. The second classification may identify network traffic as relating to unknown applications of a particular type and identity. The network device may process the network traffic according to a first security policy associated with the particular application or a second security policy associated with the unknown application of the particular type and identity based on whether a first classification identifies the network traffic as relating to the particular application or the unknown application.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] This application is a divisional application of an invention patent application with an application date of September 27, 2019, application number 201910924508.3, and invention name “Classification of unknown network traffic”. Background Art

[0002] Application identification can support network security and services such as application-aware firewalls, application-aware traffic visibility and control, application-aware quality of service (QoS), application-aware rate limiting, application-based routing, etc. Applications can be identified based on deep packet inspection (DPI) of network traffic. DPI can use application signatures that have been generated (e.g., by a DPI vendor) to identify known applications. Summary of the invention

[0003] According to some implementations, a method may include: receiving, by a network device, network traffic for an application. The method may include: determining, by a network device, a first classification for the network traffic according to a first classification technique, wherein the first classification identifies the network traffic as being associated with a specific application or an unknown application. The method may include: determining, by a network device, a second classification for the network traffic according to a second classification technique, wherein the second classification identifies the network traffic as being associated with an unknown application of a specific type and identity. The method may include: processing the network traffic, by a network device, and based on whether the first classification identifies the network traffic as being associated with a specific application or an unknown application, according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific type and identity.

[0004] According to some implementations, a network device may include one or more memories and one or more processors, the one or more processors being used to receive a first network traffic for an application. The one or more processors may determine a first classification for the first network traffic based on a first classification technique, wherein the first classification identifies the first network traffic as being associated with an unknown application of a specific type and identity. The one or more processors may process the first network traffic based on a security policy associated with an unknown application of a specific type and identity, based on the first classification. The one or more processors may receive a second network traffic for an application after processing the first network traffic. The one or more processors may determine a second classification for the second network traffic based on a second classification technique, wherein the second classification identifies the second network traffic as being associated with a specific application. The one or more processors may associate a security policy associated with an unknown application of a specific type and identity with a specific application based on the second classification. The one or more processors may process the second network traffic based on the security policy.

[0005] According to some implementations, a non-transitory computer-readable medium may store instructions, the instructions including one or more instructions, which when executed by one or more processors of a network device, cause the one or more processors to receive network traffic for an application. The one or more instructions may cause the one or more processors to determine a first classification for the network traffic based on one or more first features of the network traffic, wherein the first classification identifies the network traffic as being associated with a specific application or an unknown application. The one or more instructions may cause the one or more processors to determine a second classification for the network traffic based on one or more second features of the network traffic, wherein the second classification identifies the network traffic as being associated with an unknown application of a specific identity. The one or more instructions may cause the one or more processors to accept the following to obtain the accepted classification: a first classification when the first classification identifies the network traffic as being associated with a specific application, or a second classification when the first classification identifies the network traffic as being associated with an unknown application. The one or more instructions may cause the one or more processors to process the network traffic based on the accepted classification according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific identity. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figures 1A to 1C is a diagram of one or more example implementations described herein.

[0007] Figure 2 is a diagram of an example implementation described herein.

[0008] Figure 3 is a diagram of an example environment in which the systems and / or methods described herein may be implemented.

[0009] Figure 4 yes Figure 3 A diagram of example components of one or more devices.

[0010] Figures 5 to 7 is a flow chart of an example process for classifying unknown network traffic. DETAILED DESCRIPTION

[0011] The following detailed description of example implementations refers to the accompanying drawings.The same reference numbers in different drawings may identify the same or similar elements.

[0012] An application firewall may identify traffic as associated with an application and implement security policies for the application. For example, an application firewall may identify applications through classification techniques such as DPI. DPI helps identify applications based on application signatures associated with the applications. For example, an application signature may include a string and / or binary pattern unique to the application (e.g., within a protocol or protocol context unique to the application). An application firewall may employ an application signature library to permit the DPI component of the application firewall to identify applications based on application signatures.

[0013] Sometimes, an application's traffic cannot be identified based on the application signature known to the application firewall. This may happen, for example, if the traffic is encrypted. Encrypted traffic (such as most traffic to and from the Internet) has become commonplace, so DPI may classify a large portion of the traffic as being associated with unknown applications. In this case, the application firewall can apply security policies specified for unknown applications (such as allowing or blocking packets associated with unknown applications) to the traffic. However, applying such broad security policies to a large portion of the traffic reduces the effectiveness of the application firewall.

[0014] For example, a broad security policy for an unknown application may block legitimate traffic, rendering certain devices of the network unreachable or unable to communicate outside the network. As a result, communications associated with such devices may experience interruptions and become less reliable. As another example, a broad security policy for an unknown application (e.g., a policy that allows traffic from such an application) may allow malicious traffic, rendering the network more vulnerable to security threats. This may result in diverting computing resources and network congestion (e.g., when a device is infected with software that causes the device to send spam or participate in a distributed denial of service (DoS) attack) and in the hacking of sensitive data (e.g., private video sources, personal identification information, transaction card identifiers, etc.).

[0015] Some implementations described herein provide a network device that performs DPI classification techniques and machine learning (ML) classification techniques on network traffic. Traffic classified as an unknown application by the DPI classification technique can be classified according to the classification determined by the ML classification technique. For example, the ML classification technique can classify traffic as being associated with an unknown application of a specific type and identity. The network device can process traffic according to a security policy associated with an unknown application of a specific type and identity. In addition, if the unknown application is subsequently identified as a specific application by the DPI classification technique (e.g., based on subsequent network traffic of the unknown application), the security policy and other data of the network device associated with the unknown application of a specific type and identity can be updated and associated with the specific application.

[0016] In this way, network devices facilitate fine-grained management of security policies for unknown applications. For example, ML classification techniques can classify unknown applications based on the specific types of traffic associated with the unknown applications (e.g., streaming video, file uploads, email, etc.), rather than simply classifying unknown applications as unknown (a broad category that can include many applications with various characteristics). In addition, ML classification techniques can further distinguish multiple unknown applications of the same traffic type by associating unknown applications with specific identities (e.g., "Unknown Video 25" or "Unknown Video 103"). In this way, network devices can implement security policies that are specific to or otherwise intended for specific unknown applications, thereby improving the effectiveness of network devices.

[0017] In addition, the network device can maintain security policies and / or records associated with a specific unknown application. This permits the network device to effectively reconcile security policies and / or records when the unknown application subsequently becomes a known specific application. For example, the network device can associate security policies and records for a specific unknown application with the specific application, rather than generating a new security policy for the specific application or starting with no records of the behavior of the specific application. Accordingly, the network device facilitates efficient processing and review of newly identified network traffic, thereby saving computing resources (e.g., processing resources, memory resources, etc.).

[0018] Figures 1A to 1C is a diagram of one or more example implementations 100 described herein. Figures 1A to 1C As shown, (multiple) example implementations 100 may include a network device and a training platform. The network device may be a firewall, a gateway, a router, etc. The network device may communicate with a data network such as the Internet.

[0019] The network device may perform packet filtering on inbound and / or outbound communications with a data network based on DPI classification techniques and ML classification techniques. In some implementations, the network device may include a DPI classifier component to perform DPI classification techniques, and an ML classifier component to perform ML classification techniques. The ML classifier component may include an application classifier, a traffic type classifier, and an application identity classifier. In some implementations, the ML classifier component may include one or more additional classifiers. The outputs of the DPI classifier component and the ML classifier component may be provided to a classification result evaluator component of the network device. The classification result evaluator component may determine the security policy to be implemented with respect to the communication based on the output.

[0020] In some implementations, the network device can be associated with a training platform that uses data from the network device to train ML models for application classifiers, traffic type classifiers, and application identity classifiers. For example, a data exporter component of the network device can provide data to the training platform to permit the training platform to train one or more classification models.

[0021] As in Figure 1A As shown by reference numeral 105 in FIG. 1 , a network device may receive network traffic for an application. For example, the traffic may consist of one or more packets. The packet may include a header and a payload. The header may identify source and destination network addresses, error detection codes, and sequencing information. The payload may provide the content of the packet.

[0022] As in Figure 1B , and as shown by reference numeral 110, the network device may determine a first classification for the traffic (e.g., using a first classification technique). The first classification technique may determine the first classification for the traffic based on one or more characteristics of the traffic. The first classification may classify the traffic as being associated with a particular application (e.g., "email application ABC") or an unknown application.

[0023] In some implementations, the first classification technique may be a DPI classification technique, and the network device may perform the DPI classification technique using a DPI classifier component. The DPI classification technique may identify known applications based on application signatures that have been generated for known applications (e.g., by a DPI vendor). For example, a network device (e.g., using a DPI classifier component) may determine whether a string and / or binary pattern of traffic corresponds to a string and / or binary pattern defined in an application signature. If a correspondence is determined, the network device may classify the traffic as being associated with a specific application (e.g., a specific application associated with an application signature). Otherwise, the network device may classify the traffic as being associated with an unknown application.

[0024] As shown by reference numeral 115, the network device may determine a second classification for the traffic (e.g., using a second classification technique). The second classification technique may determine the second classification for the traffic based on one or more features of the traffic. In some implementations, one or more first features used by the first classification technique may be different from one or more second features used by the second classification technique (e.g., at least one feature is not shared by one or more first features and one or more second features). Accordingly, the first classification technique and the second classification technique may be different. The second classification may identify the application as a specific application (e.g., "email application ABC") or an unknown application of a specific type and identity (referred to herein as a "specific unknown application," e.g., "unknown email 53").

[0025] In some implementations, the second classification technique can be an ML classification technique, and the network device can use an ML classifier component to perform the ML classification technique. The ML classifier component can employ one or more ML models to identify the traffic as being associated with a specific application or a specific unknown application. In some implementations, the network device can determine one or more features associated with the traffic (e.g., associated with the traffic flow and / or encryption of the traffic), and can use one or more ML models to process the one or more features to determine a classification for the traffic.

[0026] In some implementations, features associated with the flow may be associated with portions of one or more packets of the flow (e.g., the first 50 bytes, the first 500 bytes, the first 1000 bytes, etc. of the payload of the packet). Additionally, features associated with the flow may relate to a flow of the flow. In some implementations, features associated with the flow of the flow may relate to a sequence of packet lengths (e.g., in bytes) and / or arrival times (e.g., arrival times of packets relative to arrival times of previous packets) for a particular number of packets of the flow. Additionally or alternatively, features associated with the flow of the flow may relate to transitions in packet size. For example, a transition from a smaller packet (e.g., a packet that does not meet a threshold size) to a larger packet (e.g., a packet that meets a threshold size), a degree of transition (e.g., in bytes or percentiles), a number of transitions, etc. Additionally or alternatively, features associated with the flow of the flow may relate to transitions in packet arrival times—e.g., pauses between packet arrivals (e.g., pauses of a duration that meets a threshold), a duration of a pause, a number of pauses, etc.

[0027] In addition, the features related to the traffic may relate to encryption of the traffic. In some implementations, the features related to encryption of the traffic may relate to the type of encryption (e.g., encryption based on the Secure Sockets Layer (SSL) protocol, the Transport Layer Security (TLS) protocol, etc., which may be identified by a byte pattern), the encryption suite specified by the client during the handshake, the encryption suite specified by the server during the handshake, the encryption extension (e.g., the type of extension and / or the length of the extension), the length of the public key, the server certificate, the handshake type, the encryption record length and / or type (e.g., client to server or server to client), the encryption record time (e.g., client to server or server to client), etc.

[0028] In some implementations, the ML classification technique can determine whether the traffic is associated with a particular application. For example, an application classifier of an ML classifier component of a network device can classify the traffic as being associated with a particular application. This can help confirm the classification determined by the DPI classification technique. For example, when the traffic impersonates an application signature associated with a known application, the DPI classification technique may incorrectly classify the traffic. Therefore, the application classifier of the ML classifier component can provide a secondary classification of the traffic based on features related to the encryption of the traffic flow and / or the traffic, which cannot be impersonated.

[0029] The application classifier of the ML classifier component may include an application classifier model. In some implementations, the application classifier model may be a supervised or semi-supervised artificial neural network, such as a multilayer perceptron or a convolutional neural network. The application classifier model may receive features related to the traffic (e.g., traffic flow, encryption of the traffic, etc.) as input and provide a specific application (e.g., application identifier, application name, etc.) identified from the input as output. If the specific application cannot be identified by the application classifier model, the application classifier model may provide an output indicating that the input is associated with an unknown application.

[0030] When the application classifier determines that the traffic is associated with a specific application, the traffic may not need to be further processed according to the ML classification technique. However, when the application classifier determines that the traffic is associated with an unknown application, the ML classifier component of the network device can use a traffic type classifier and an application identity classifier to process the traffic. The traffic type classifier can identify the unknown application as being associated with a specific type of traffic (e.g., streaming video, file uploads, email, chat, peer-to-peer networks, etc.). The application identity classifier can identify the unknown application (e.g., an unknown application classified as a specific type) as being associated with a specific identity (e.g., an identity based on characteristics of the traffic that are characteristic of the unknown application).

[0031] For example, assume that a network device is receiving traffic including three traffic flows that are classified by the DPI classifier component and the application classifier of the ML classifier component as being associated with an unknown application. Based on features associated with the traffic (e.g., encryption of the traffic flow and / or the traffic), the traffic type classifier of the ML classifier component may classify one of the traffic flows as being associated with email traffic (e.g., “unknown email”) and two of the traffic flows as being associated with streaming video traffic (e.g., “unknown video”). Additionally, based on features associated with the traffic (e.g., one or more features or one or more different features based on which the traffic type classifier classified), the application identity classifier of the ML classifier component may assign the traffic flows to a cluster (e.g., a cluster associated with a unique identifier). For example, with respect to two traffic flows classified as streaming videos, the application identity classifier may identify the first traffic flow as being associated with a first specific identity (e.g., “unknown video 25”) and identify the second traffic flow as being associated with a second specific identity (e.g., “unknown video 103”). In this manner, the ML classification technique may determine that the unknown application is a specific unknown application.

[0032] The network device may use the output of the ML classifier component (e.g., “Unknown Video 25”) as an identifier of a particular unknown application (e.g., “Unknown Video 25” may identify a first particular unknown application, and “Unknown Video 103” may identify a second particular unknown application). In some implementations, the identifier may also include additional information about the traffic, such as encryption or a protocol (e.g., a protocol associated with layers 4 to 7 of the Open Systems Interconnection (OSI) model) associated with the traffic (e.g., “Unknown TLS Video 25”). The identifier may be determined based on a pattern for marking a particular unknown application (e.g., “Unknown” - <protocol> - <encrypted or unencrypted> - <traffic type>). The identifier may be used to record characteristics of traffic associated with a particular unknown application, associate security policies with a particular unknown application, enforce security policies for traffic associated with a particular unknown application, and the like. In this manner, the network device may process and / or record traffic for a particular unknown application in a manner similar to traffic associated with a known application.

[0033] The traffic type classifier of the ML classifier component may include a traffic type classifier model. In some implementations, the traffic type classifier model may employ a supervised or semi-supervised artificial neural network, such as a multi-layer perceptron or a convolutional neural network. The traffic type classifier model may receive features related to traffic (e.g., traffic flow, encryption of traffic, etc.) as input and provide a specific type of traffic (e.g., streaming video, file upload, email, etc.) identified from the input as output. As the traffic type classifier model learns to identify new traffic types, the specific types of traffic that the traffic type classifier model is able to identify may change over time.

[0034] The application identity classifier of the ML classifier component may include an application identity classifier model. In some implementations, the application identity classifier model may employ an unsupervised two-stage machine learning technique. The first stage of the machine learning technique may employ a multilayer perceptron, an autoencoder based on a convolutional neural network, and the second stage of the machine learning technique may employ a k-means clustering algorithm, etc. For example, according to the first stage, the application identity classifier model may receive features related to traffic (e.g., traffic flow, encryption of traffic, etc.) and the output of the traffic type classifier model (e.g., an output identifying a specific type of traffic) as input, and may provide a vector as output that identifies a set of features learned from the input and associated with the identification of an unknown application. Continuing with the previous example, according to the second stage, the application identity classifier model may receive the output of the first stage (e.g., a vector including a set of learned features) as input, and may provide a cluster to which the unknown application belongs (e.g., "Video 25") as output (e.g., an existing cluster or a new cluster created for the unknown application).

[0035] As shown by reference numeral 120, a network device may receive a DPI classification (eg, determined by a DPI classification technique) or an ML classification (eg, determined by an ML classification technique).

[0036] When the DPI classification identifies the traffic as a specific application and the ML classification identifies the traffic as an unknown application (e.g., the application classifier of the ML classifier component identifies the traffic as an unknown application), the network device can accept the DPI classification of the traffic. Likewise, when the DPI classification identifies the traffic as associated with an unknown application and the ML classification identifies the traffic as associated with a specific application (e.g., the application classifier of the ML classifier component identifies the traffic as associated with a specific application), the network device can accept the ML classification of the traffic.

[0037] In some implementations, both the DPI classification and the ML classification (e.g., the ML classification determined by the application classifier of the ML classifier component) can classify the traffic as being associated with a specific application (e.g., the same specific application). In this case, the network device can accept the DPI classification and the ML classification of the traffic as being associated with the specific application.

[0038] In some implementations, DPI classification may classify traffic as being associated with a first specific application, and ML classification (e.g., ML classification determined by an application classifier of an ML classifier component) may classify traffic as being associated with a second specific application. In some implementations, when traffic is not encrypted, DPI classification techniques may be more reliable than ML classification techniques, and the network device may accept DPI classification (e.g., according to default behavior); when traffic is encrypted, ML classification techniques may be more reliable than DPI classification techniques, and the network device may accept ML classification (e.g., according to default behavior). In some implementations, the network device may employ another criterion (e.g., using a machine learning model) to determine whether to accept DPI classification or ML classification. This may result in the network device accepting DPI classification when traffic is encrypted and / or accepting ML classification when traffic is not encrypted.

[0039] In some implementations, the DPI classification may classify the traffic as being associated with an unknown application, and the ML classification (e.g., the ML classification determined by the traffic type classifier and the application identity classifier of the ML classifier component) may classify the traffic as being associated with a specific unknown application (e.g., an unknown application of a specific type and identity). In this case, the network device may accept the ML classification of the traffic as being associated with the specific unknown application.

[0040] Accordingly, the accepted classification may correspond to a specific application or a specific unknown application (e.g., an unknown application of a specific type and identity). In some implementations, the network device may use a classification result evaluator component to determine whether to accept a DPI classification or an ML classification. The classification result evaluator component of the network device may output an accepted classification of the DPI classification and the ML classification to permit the network device to process the traffic according to the accepted classification. In the case where the DPI classification is the accepted classification, the network device may provide data related to the DPI classification and the characteristics of the traffic to the training platform to permit training of the ML classifier component, as described below.

[0041] As shown by reference numeral 125, the network device may process the traffic according to the security policy associated with the accepted classification. For example, when the accepted classification is a specific application (e.g., "email application ABC"), the network device may process the traffic according to the security policy associated with the specific application (e.g., the security policy associated with "email application ABC"). As another example, when the accepted classification is a specific unknown application (e.g., an unknown application of a specific type and identity), the network device may process the traffic according to the security policy associated with the specific unknown application (e.g., the security policy associated with "unknown video 25").

[0042] In some implementations, when the accepted classification is a specific unknown application, the network device can process the traffic according to a security policy associated with a specific type (e.g., a traffic type) of the specific unknown application. For example, the network device can process the traffic of a specific unknown application as an unknown video according to a specific security policy configured for unknown videos. In some implementations, an administrator of the network device can determine the security policy for a specific traffic type and provide the security policy to the network device (e.g., via a command line interface). For example, an administrator of the network device can configure a security policy for a specific traffic type (e.g., a specific traffic type that was not previously identified by the network device) based on data related to the specific traffic type (e.g., data related to characteristics of the traffic flow and / or encryption of the traffic, etc.). In this way, a default security policy can be configured for a new traffic type learned by a traffic type classifier model.

[0043] In some implementations, an administrator of the network device may determine a security policy for a specific application or a specific unknown application and provide the security policy to the network device (e.g., via a command line interface). For example, after identifying a specific unknown application, the network device may provide (e.g., export) data related to the specific unknown application (e.g., data related to characteristics of the traffic flow and / or encryption of the traffic, the type of traffic, an identifier of the specific unknown application, etc.) to allow the administrator of the network device to configure a security policy for the specific unknown application.

[0044] Alternatively, the network device may generate (e.g., automatically generate) a security policy based on one or more characteristics of the specific application or the specific unknown application (e.g., the source of the application, the type of the application, the traffic pattern associated with the application, etc.). For example, the network device may generate a security policy for the specific application or the specific unknown application based on a security policy of a related application having characteristics similar to one or more characteristics of the specific application or the specific unknown application.

[0045] In some implementations, a network device may generate (e.g., automatically generate) a security policy for a specific application or a specific unknown application based on a set of standards. The set of standards may relate to the type of traffic, the source of the traffic, the destination of the traffic, the user associated with the traffic, the traffic flow, the encryption of the traffic, the frequency of the traffic, and the like. The set of standards may be associated with an action for generating a security policy. For example, the action may indicate that the network device will generate a security policy that denies traffic, allows traffic, isolates traffic, and the like. In some implementations, a set of standards may be specified for a specific type of traffic and / or a specific type of encryption (e.g., a set of standards may be specified for unknown encrypted upload traffic). The network device may determine whether traffic (e.g., traffic associated with a specific unknown application) corresponds to a set of standards, and generate a security policy according to an action associated with the set of standards (e.g., based on determining that the traffic corresponds to the set of standards).

[0046] In some implementations, the network device may record information related to the accepted classified specific application or specific unknown application. For example, the network device may record information related to traffic associated with the specific application or the specific unknown application (e.g., the source address of the traffic, the destination address of the traffic, the time of the traffic, the way the network device processes the traffic, etc.). The network device may record the information in a manner that associates the information with the specific application or the specific unknown application. For example, the network device may record the information in association with an identifier of the specific application or the specific unknown application.

[0047] As in Figure 1C , and as shown by reference numeral 130, the network device may reconcile an ML classification (e.g., an ML classification that identifies a particular unknown application) with a DPI classification (e.g., a DPI classification that identifies a particular application). For example, after classifying traffic as being associated with a particular unknown application (e.g., according to an ML classification technique), the network device may receive subsequent traffic for the application and use the DPI classification technique to classify the subsequent traffic as being associated with the particular application. This scenario may occur when a DPI classifier component of the network device is unaware of an application signature for the application upon receiving initial traffic for the application (e.g., resulting in classification of the traffic as the particular unknown application), but subsequently obtains the application signature prior to receiving subsequent traffic for the application.

[0048] In this case, the DPI classification technique may classify the subsequent traffic as a specific application (e.g., based on an application signature), while the ML classification technique may classify the subsequent traffic as a specific unknown application (e.g., because the ML classifier component has not been trained to recognize that the subsequent traffic is associated with the specific application). Based on these classifications, the network device may identify the output of the ML classifier component for the specific unknown application (e.g., "Unknown Video 25") as an alias for the specific application (e.g., "Email Application ABC"). Accordingly, the network device may reconcile the ML classification with the DPI classification.

[0049] For example, the network device can update the security policy associated with a specific unknown application to be associated with a specific application. In this way, the network device can implement the security policy for the specific application instead of generating a new security policy for the specific application, thereby saving computing resources.

[0050] As another example, the network device may update an application signature associated with a particular application to indicate that the output of the ML classifier component for the particular unknown application (e.g., "unknown video 25") is an alias for the particular application. In this case, the network device may include an entry (e.g., in an alias list) for an application signature for the particular unknown application (e.g., an application signature determined by the ML classifier component) that identifies the association with the particular application, and / or may include an entry (e.g., in an alias list) for an application signature for the particular application that identifies the association with the particular unknown application. As yet another example, the network device may generate a record (e.g., a system log or IPFIX record) that maps the output of the ML classifier component for the particular unknown application (e.g., "unknown video 25") to the particular application. In this way, historical logs or other records associated with the particular unknown application may be associated with the particular application. This may be useful for forensic purposes. For example, if the particular application is malware, the mapping of the particular unknown application to the particular application may be used to identify historical traffic information associated with the particular application.

[0051] As shown by reference numeral 135, the network device can export data to the training platform to enable training of the ML classifier component. For example, the network device can use a data exporter component of the network device to export data to the training platform. The data exported to the training platform may include data related to the traffic (e.g., the flow of the traffic and / or encryption of the traffic), data related to classifying the traffic according to the DPI classification technique, data related to the association between a specific application and a specific unknown application, etc. In some implementations, multiple network devices can export such data to the training platform to enable training of the ML classifier component.

[0052] The training platform may use the data to train (e.g., adjust) an application classifier model, a traffic type classifier model, and / or an application identity classifier model. For example, the training platform may use information related to multiple traffic flows, multiple DPI classifications associated with multiple traffic flows, and / or multiple ML classifications associated with the DPI classifications to train the model. As an example, the training platform may determine that a feature of a past traffic flow is associated with a threshold probability, which is associated with a specific application or a specific unknown application. In some implementations, the training platform may use a scoring system (e.g., with a relatively high score and / or a relatively low score) to identify and / or classify a traffic flow as being associated with a specific application or a specific unknown application. In this case, the training platform may determine to assign a relatively high score (e.g., likely to be identified) to such a traffic flow, which is determined to be the same or similar to a previously identified traffic flow of a specific application or a specific unknown application. Conversely, the training platform may determine to assign a relatively low score (e.g., unlikely to be identified) to such a traffic flow, which is determined to be different from a past identified traffic flow of a specific application or a specific unknown application.

[0053] In some implementations, the training platform may perform training operations when generating a model. For example, the training platform may divide the traffic flow into a training set (e.g., a data set for training the model), a validation set (e.g., a data set for evaluating the fitness of the model and / or fine-tuning the model), a test set (e.g., a data set for evaluating the final fitness of the model), etc. In some implementations, the training platform may pre-process and / or perform dimensionality reduction to reduce the features of the traffic flow to a minimum feature set. In some implementations, the training platform may train the model on the minimum feature set, thereby reducing the processing used to train the model, and classification techniques may be applied to the minimum feature set.

[0054] In some implementations, the training platform may use classification techniques, such as logistic regression classification techniques, random forest classification techniques, gradient boosting machine learning (GBM) techniques, etc., to determine classification results (e.g., whether the features of the traffic flow correspond to or do not correspond to a specific application or a specific unknown application). Additionally or alternatively, the training platform may use naive Bayes classifier technology. In this case, the training platform may perform binary recursive partitioning to divide the data of the minimum feature set into partitions and / or branches, and use the partitions and / or branches to perform predictions (e.g., whether the features of the traffic flow correspond to or do not correspond to a specific application or a specific unknown application). Based on the use of recursive partitioning, the training platform can reduce the utilization of computing resources relative to manual linear sorting and analysis of data points, thereby using thousands, millions, or billions of data points to train the model, which may produce a more accurate model than using fewer data points.

[0055] Additionally or alternatively, the training platform may use support vector machine (SVM) classifier technology to generate nonlinear boundaries between data points in the training set. In this case, the nonlinear boundary is used to classify test data (e.g., data related to traffic flows) into specific categories (e.g., a category indicating that the traffic flow corresponds to a specific application or a specific unknown application, a category indicating that the traffic flow does not correspond to a specific application or a specific unknown application, etc.).

[0056] Additionally or alternatively, the training platform may train the model (e.g., the application classifier model and / or the traffic type classifier model) using a supervised training procedure that includes receiving input to the model from a subject matter expert, which may reduce the amount of time, amount of processing resources, etc. to train the model relative to an unsupervised training procedure. In some implementations, the training platform may use one or more other model training techniques, such as neural network techniques, latent semantic indexing techniques, etc. For example, the training platform may perform artificial neural network processing techniques (e.g., using a two-layer feedforward neural network architecture, a three-layer feedforward neural network architecture, etc.) to perform pattern recognition regarding whether different traffic flows are differently associated with a particular application or a particular unknown application. In this case, the use of artificial neural network processing techniques may improve the accuracy of the model generated by the training platform by being more robust to noisy, imprecise, or incomplete data and enabling the training platform to use less complex techniques to detect patterns and / or trends that human analysts or systems cannot detect.

[0057] As an example, the training platform may use supervised multi-label classification techniques to train models (e.g., application classifier models and / or traffic type classifier models). For example, as a first step, the training platform may map a traffic flow to a specific application or a specific unknown application. In this case, the traffic flow may be characterized as being associated with or not associated with a specific application or a specific unknown application based on features of the traffic flow (e.g., whether the features of the traffic flow are similar to or associated with features of a specific application or a specific unknown application) and analysis of the traffic flow (e.g., analysis by a technician, thereby reducing the processing required to analyze each activity relative to the training platform). As a second step, the training platform may determine a classifier chain through which labels of the target variable may be associated (e.g., in this example, the labels may be features of the traffic flow, and the correlation may refer to being associated with a common specific application or a specific unknown application). In this case, the training platform can use the output of the first labeling as input for the second labeling (as well as one or more input features, which can be other data related to the specific application or the specific unknown application), and can determine the likelihood that a specific traffic flow including a set of features (some features are associated with the specific application or the specific unknown application, and some features are not associated with the specific application or the specific unknown application) is associated with the specific application or the specific unknown application based on similarity to other traffic flows including similar features. In this way, the training platform converts the classification from a multi-label classification problem to multiple single-classification problems, thereby reducing processing utilization. As a third step, the training platform can determine a Hamming loss metric related to the accuracy of the labeling when performing classification by using a validation set of data. For example, weighted accuracy is applied to each feature and whether each feature is associated with a specific application or a specific unknown application, so that a correct prediction is made as to whether the traffic flow corresponds to the specific application or the specific unknown application, thereby illustrating that the relevance of any one feature has different degrees of influence on the traffic flow classified as associated with the specific application or the specific unknown application. As a fourth step, the training platform can finalize the model based on the markings satisfying a threshold accuracy associated with the Hamming loss metric, and the model can be used to make subsequent predictions as to whether the characteristics of the traffic flow will cause the traffic flow to be classified as associated with a particular application or a particular unknown application.

[0058] As another example, the training platform may use linear regression techniques to determine that a threshold percentage of features in a feature set are not associated with a specific application or a specific unknown application, and may determine that these features will receive a relatively low association score. Conversely, the training platform may determine that another threshold percentage of features are associated with a specific application or a specific unknown application, and may assign a relatively high association score to these features. Based on the features being associated (or not associated) with a specific application or a specific unknown application, the training platform may generate a model, and may use the model to analyze new features identified by the training platform.

[0059] After performing a training operation for an application classifier model, a traffic type classifier model, and / or an application identity classifier model, or after another interval (e.g., daily, weekly, monthly, etc.), the training platform can generate a package that includes the application classifier model, the traffic type classifier model, the application identity classifier model, and / or one or more updates thereto. The network device can obtain the package from the training platform and use the package to update the ML classifier component of the network device. In this way, the classification performed by the ML classifier component of the network device can be improved over time.

[0060] As indicated above, Figures 1A to 1C are provided only as one or more examples. Other examples may differ from those described herein. Figures 1A to 1C as described.

[0061] Figure 2 is a diagram of an example implementation 200 described herein. Figure 2 As shown, the network device 205 may include a configuration component 210 , a feature extractor component 215 , a DPI classifier component 220 , an ML classifier component 225 , a classification result evaluator component 230 , and a data exporter component 235 .

[0062] The network device 205 can store information identifying features to be extracted from traffic received by the network device 205 using the configuration component 210. The configuration component 210 can include a data structure (e.g., a database, a linked list, a table, etc.) used to store information. The network device 205 can extract features (e.g., based on features identified by the configuration component 210) from the traffic received by the network device 205 using the feature extractor component 215. These features can relate to a source address (e.g., an Internet Protocol (IP) address), a source port number, a destination address (e.g., an IP address), a destination port number, a routing instance associated with the traffic, a transport layer protocol (e.g., a transmission control protocol (TCP), a user datagram protocol (UDP), etc.), an application layer protocol (e.g., a file transfer protocol (FTP), a hypertext transfer protocol (HTTP), HTTP security (HTTPS) (e.g., HTTP over SSL, HTTP over TLS, etc.), etc.), a portion of a packet payload (e.g., the first 50 bytes, the first 500 bytes, the first 1000 bytes, etc.), a sequence of packet lengths and / or arrival times, a conversion of packet sizes, a conversion of packet arrival times, etc.

[0063] The network device 205, using the DPI classifier component 220, can determine a DPI classification for traffic received by the network device 205 in a manner similar to that described elsewhere herein (e.g., based on a plurality of application signatures included in the DPI classifier component 220). The DPI classifier component 220 can provide information determined from the DPI classification (e.g., protocol information, such as Server Name Indication (SNI) for TLS traffic) for use in assisting the feature extractor component 215. The network device 205, using the ML classifier component 225, can determine an ML classification for traffic received by the network device 205 in a manner similar to that described elsewhere herein (e.g., based on one or more features extracted by the feature extractor component 215).

[0064] The network device 205 using the classification result evaluator component 230 can determine to accept the DPI classification of the DPI classifier component 220 or the ML classification of the ML classifier component 225 in a manner similar to that described elsewhere herein. The network device 205 using the data exporter component 235 can export data to the training platform 240.

[0065] like Figure 2 As shown, the training platform 240 may include: a data aggregator component 245 , a data storage component 250 , a training component 255 , and a packaging component 260 .

[0066] The training platform 240 can collect data from multiple network devices (e.g., multiple network devices including the network device 205), organize the data, modify the data, perform pre-processing on the data, etc. using the data aggregator component 245. The training platform 240 can store the data aggregated by the data aggregator component 245 using the data storage component 250. The data storage component 250 can include a data structure (e.g., a database, a linked list, a table, etc.) used to store data.

[0067] The training platform 240, using the training component 255, can train one or more ML models to be used by the ML classifier component 225 of the network device 205. For example, the training component 255 can train an application classifier model, a traffic type classifier model, and / or an application identity classifier model similar to the models described elsewhere herein. The training component 255 can train one or more ML models using data aggregated from multiple network devices (e.g., data stored by the data storage component 250) in a manner similar to that described elsewhere herein.

[0068] In some implementations, the application classifier model, the traffic type classifier model, and / or the application identity classifier model can learn to identify applications, traffic types, and / or application identities that were previously identified as unidentified, respectively. In this case, the training platform 240 can export (e.g., using the encapsulation component 260) information related to the previously unidentified applications, traffic types, and / or application identities to allow administrators to mark previously unidentified applications, traffic types, and / or application identities, configure security policies for previously unidentified applications, traffic types, and / or application identities, etc. For example, the traffic type classifier model can learn to identify previously unidentified traffic types. Continuing with the previous example, the training platform 240 can export (e.g., using the encapsulation component 260) information related to previously unidentified traffic types to the network device 205 to allow administrators of the network device 205 to configure default security policies for previously unidentified traffic types. The network device 205 can automatically implement a default security policy in conjunction with unknown applications of previously unidentified traffic types.

[0069] The training platform 240, using the packaging component 260, can generate a package (e.g., a downloadable package) that includes one or more trained ML models (e.g., ML models trained by the training component 255) and / or one or more updates to the ML models. The packaging component 260 can make the package available to the network device 205 (e.g., in a repository of the packaging component 260 that the network device 205 can check regularly or irregularly to download new packages). Additionally or alternatively, the packaging component 260 can push the package to the network device 205 regularly (e.g., daily, weekly, monthly, etc.) or irregularly (e.g., after a training operation is performed by the training component 255).

[0070] In some implementations, one or more components of training platform 240 can be included in network device 205 (eg, network device 205 can include training component 225 ).

[0071] As indicated above, Figure 2 It is provided as an example only. Other examples may differ from those regarding Figure 2 as described. Figure 2 The number and arrangement of components shown are provided as one or more examples. Figure 2 There may be additional components, fewer components, different components, or components that are arranged differently than those shown. Figure 2 Two or more of the components shown may be implemented in a single device, or Figure 2 The single device shown may be implemented as multiple distributed devices. Additionally or alternatively, a device or set of components (eg, one or more devices) of the example implementation 200 may perform one or more functions described as being performed by another device or set of components of the example implementation 200.

[0072] Figure 3 is a diagram of an example environment 300 in which the systems and / or methods described herein may be implemented. Figure 3 As shown, environment 300 may include one or more network devices 310-1 to 310-N (N≥1) (hereinafter collectively referred to as "network device 310" and individually referred to as "network device 310"), server device 320, network 330, training platform 340, computing resources 345, and cloud computing environment 350. The devices of environment 300 may be interconnected via wired connections, wireless connections, or a combination of wired and wireless connections.

[0073] The network device 310 includes one or more devices (e.g., one or more traffic transmission devices) capable of processing and / or transferring traffic between endpoint devices. For example, the network device 310 may include a firewall, a router, a gateway, a switch device, a hub, a bridge, a reverse proxy, a server (e.g., a proxy server), a top-of-rack (ToR) switch, a load balancer, a switch interface board, a controller, a switch component, a packet processing component, or the like. In some implementations, the network device 310 may use DPI classification techniques and / or ML classification techniques to perform classification of network traffic, process network traffic according to security policies, update security policies, etc.

[0074] In some implementations, the network devices 310 may form a distributed architecture such that a first network device 310 performs classification using a DPI classification technique and a second network device 310 performs classification using an ML classification technique. In some implementations, the network device 310 may be a physical device implemented within a housing such as a chassis. In some implementations, the network device 310 may be a virtual device implemented by one or more computer devices of a cloud computing environment or a data center.

[0075] The server device 320 includes one or more devices capable of receiving, generating, storing, processing and / or providing information (such as the information described herein). For example, the server device 320 may include a computing device such as a server, a desktop computer, a laptop computer, a tablet computer, a handheld computer, or a similar device.

[0076] The network 330 includes one or more wired and / or wireless networks. For example, the network 330 may include a cellular network, a public land mobile network ("PLMN"), a local area network ("LAN"), a wide area network ("WAN"), a metropolitan area network ("MAN"), a telephone network (e.g., a public switched telephone network ("PSTN")), an ad hoc network, an intranet, the Internet, a fiber-based network, and / or a combination of these or other types of networks.

[0077] The training platform 340 includes one or more computing resources associated with a training model (e.g., a machine learning model). For example, the training platform 340 can be a platform implemented by a cloud computing environment 350 that can train and / or update one or more models for classifying network traffic.

[0078] The training platform 340 may include one or more server devices. In some implementations, the training platform 340 is implemented by computing resources 345 of a cloud computing environment 350. It should be noted that although the implementations described herein describe the training platform 340 as being hosted in a cloud computing environment 350, in some implementations, the training platform 340 may not be cloud-based or may be partially cloud-based.

[0079] The cloud computing environment 350 includes an environment that delivers computing as a service, through which shared resources, services, etc. can be provided to the network device 310, the server device 320, etc. The cloud computing environment 350 can provide computing, software, data access, storage, and / or other services that do not require the end user to have knowledge of the physical location and configuration of the system and / or device that delivers the service. As shown, the cloud computing environment 350 can include a training platform 340 and computing resources 345.

[0080] The computing resources 345 include one or more personal computers, workstation computers, server devices, or another type of computing and / or communication devices. In some implementations, the computing resources 345 can host the training platform 340. Cloud resources can include computing instances executed in the computing resources 345, storage devices provided in the computing resources 345, data transmission devices provided by the computing resources 345, etc. In some implementations, the computing resources 345 can communicate with other computing resources 345 via wired connections, wireless connections, or a combination of wired and wireless connections.

[0081] like Figure 3 As further shown, the computing resources 345 may include a set of cloud resources, such as one or more applications ("APP") 345-1, one or more virtual machines ("VM") 345-2, virtualized storage devices ("VS") 345-3, one or more hypervisors ("HYP") 345-4, etc.

[0082] Applications 345-1 include one or more software applications that may be provided to or accessed by network device 310, server device 320, etc. Applications 345-1 may eliminate the need to install and execute software applications on network device 310, server device 320, etc. For example, applications 345-1 may include software associated with training platform 340 and / or any other software that can be provided via cloud computing environment 350. In some implementations, one application 345-1 may send / receive information to / from one or more other applications 345-1 via virtual machine 345-2.

[0083] Virtual machine 345-2 includes a software implementation of a machine (e.g., a computer) that executes programs like a physical machine. Virtual machine 345-2 can be a system virtual machine or a process virtual machine, depending on the use and correspondence of virtual machine 345-2 to any real machine. System virtual machines can provide a complete system platform that supports the execution of a complete operating system ("OS"). Process virtual machines can execute a single program and can support a single process. In some implementations, virtual machine 345-2 can execute on behalf of a user and can manage the infrastructure of cloud computing environment 350, such as data management, synchronization, or long-term data transfer.

[0084] The virtualized storage device 345-3 includes one or more storage systems and / or one or more devices using virtualization technology within the storage system or device of the computing resource 345. In some implementations, within the context of the storage system, the types of virtualization may include block virtualization and file virtualization. Block virtualization may refer to the abstraction (or separation) of logical storage from physical storage, so that the storage system can be accessed without considering the physical storage or heterogeneous structure. Separation may allow the storage system to flexibly manage how administrators manage the storage of end users. File virtualization may eliminate the dependency between data accessed at the file level and the location of the physical storage file. This may enable optimization of storage usage, server consolidation, and / or non-disruptive file migration performance.

[0085] Hypervisor 345-4 provides hardware virtualization technology that allows multiple operating systems (e.g., "guest operating systems") to execute simultaneously on a host computer (such as computing resource 345). Hypervisor 345-4 can present a virtual operating platform to the guest operating systems and can manage the execution of the guest operating systems. Multiple instances of various operating systems can share virtualized hardware resources.

[0086] Figure 3 The number and arrangement of devices and networks shown are provided as one or more examples. Figure 3 There may be additional devices and / or networks, fewer devices and / or networks, different devices and / or networks, or differently arranged devices and / or networks than those shown. Figure 3 Two or more of the devices shown may be implemented in a single device, or Figure 3 The single device shown may be implemented as multiple distributed devices. Additionally or alternatively, a set of devices (eg, one or more devices) of environment 300 may perform one or more functions described as being performed by another set of devices of environment 300.

[0087] Figure 4 4 is a diagram of example components of device 400. Device 400 may correspond to network device 310, server device 320, training platform 340, and / or computing resource 345. In some implementations, network device 310, server device 320, training platform 340, and / or computing resource 345 may include one or more devices 400 and / or one or more components of device 400. Figure 4 As shown, device 400 may include a set of input components 405, a switch component 410, a set of output components 415, and a controller 420. The components of device 400 may be interconnected via wired connections, wireless connections, or a combination of wired and wireless connections.

[0088] The input component 405 can be an attachment point for a physical link connected to the device 400, and can be an entry point for incoming traffic (e.g., packets) received by the device 400. The input component 405 can process the incoming traffic, such as by performing data link layer encapsulation or decapsulation. In some implementations, the input component 405 can send and / or receive packets. In some implementations, the input component 405 can include an input line card, which includes one or more packet processing components (e.g., in the form of an integrated circuit), such as, one or more interface cards (IFCs), packet forwarding components, line card controller components, input ports, processors, memory, and / or input queues.

[0089] The switch component 410 can interconnect the input component 405 with the output component 415. In some implementations, the switch component 410 can be implemented via one or more crossbar switches, via one or more buses, and / or using shared memory. The shared memory can act as a temporary buffer to store packets from the input component 405 before the packets are finally scheduled for delivery to the output component 415. In some implementations, the switch component 410 can support input component 405, output component 415, and / or controller 420 communications.

[0090] The output component 415 can be an attachment point for a physical link connected to the device 400, and can be an exit point for outgoing traffic (e.g., packets) transmitted by the device 400. The output component 415 can store packets and / or can schedule packets for transmission on an output physical link. The output component 415 can support data link layer encapsulation or decapsulation and / or various high-level protocols. In some implementations, the output component 415 can send packets and / or receive packets. In some implementations, the output component 415 may include an output line card, which includes one or more packet processing components (e.g., in the form of an integrated circuit), such as one or more IFCs, packet forwarding components, line card controller components, output ports, processors, memories, and / or output queues. In some implementations, the input component 405 and the output component 415 can be implemented by the same set of components (e.g., the input / output component can be a combination of the input component 405 and the output component 415).

[0091] The controller 420 includes a processor in the form of, for example, a central processing unit (CPU), a microprocessor, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), and / or another type of processor that can interpret and / or execute instructions. The processor is implemented in hardware, firmware, or a combination of hardware and software. In some implementations, the controller 420 may include one or more processors that can be programmed to perform a function.

[0092] In some implementations, the controller 420 may include random access memory (RAM), read-only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, optical memory, etc.) that stores information and / or instructions for use by the controller 420.

[0093] In some implementations, the controller 420 can communicate with other devices, networks, and / or systems connected to the device 400 to exchange information about the network topology. The controller 420 can create a routing table based on the network topology information, can create a forwarding table based on the routing table, and can forward the forwarding table to the input component 405 and / or the output component 415. The input component 405 and / or the output component 415 can use the forwarding table to perform a route lookup for incoming and / or outgoing packets.

[0094] The controller 420 may perform one or more processes described herein. The controller 420 may perform these processes in response to executing software instructions stored by a non-transient computer readable medium. A computer readable medium is defined herein as a non-transient memory device. A memory device includes storage space within a single physical storage device or storage space distributed on multiple physical storage devices.

[0095] The software instructions may be read from another computer-readable medium or from another device into a memory and / or storage component associated with the controller 420 via a communication interface. When executed, the software instructions stored in the memory and / or storage component associated with the controller 420 may cause the controller 420 to perform one or more processes described herein. Additionally or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more processes described herein. Thus, the implementations described herein are not limited to any specific combination of hardware circuitry and software.

[0096] Figure 4 The number and arrangement of components shown are provided as examples. Figure 4 The device 400 may include additional components, fewer components, different components, or differently arranged components compared to the components shown. Additionally or alternatively, a set of components (e.g., one or more components) of the device 400 may perform one or more functions described as being performed by another set of components of the device 400.

[0097] Figure 5 is a flow chart of an example process 500 for classifying unknown network traffic. In some implementations, Figure 5 One or more process blocks of may be performed by a network device (e.g., network device 310). In some implementations, Figure 5One or more process blocks of may be performed by another device or device group that is separate from or includes the network device, such as a server device (eg, server device 320), a training platform (eg, training platform 340), and the like.

[0098] like Figure 5 As shown, process 500 may include receiving network traffic for an application (block 510). For example, a network device (eg, using input component 405, switch component 410, controller 420, etc.) may receive network traffic for an application, as described above.

[0099] like Figure 5 As further shown, process 500 may include determining a first classification for network traffic according to a first classification technique, wherein the first classification identifies the network traffic as being associated with a particular application or an unknown application (block 520). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine a first classification for network traffic according to a first classification technique, as described above. In some implementations, the first classification identifies the network traffic as being associated with a particular application or an unknown application.

[0100] like Figure 5 As further shown, process 500 may include determining a second classification for the network traffic according to a second classification technique, wherein the second classification identifies the network traffic as being associated with an unknown application of a particular type and identity (block 530). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine a second classification for the network traffic according to a second classification technique, as described above. In some implementations, the second classification identifies the network traffic as being associated with an unknown application of a particular type and identity.

[0101] like Figure 5 As further shown, process 500 may include processing network traffic according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific type and identity, based on whether the first classification identifies the network traffic as being associated with the specific application or the unknown application (block 540). For example, a network device (e.g., using switch component 410, output component 415, controller 420, etc.) may process network traffic according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific type and identity, based on whether the first classification identifies the network traffic as being associated with the specific application or the unknown application, as described above.

[0102] Process 500 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in conjunction with one or more other processes described elsewhere herein.

[0103] In some implementations, the first classification technique may be a deep packet inspection classification technique, and the second classification technique may be a machine learning classification technique. In some implementations, the second classification technique may include: using a first machine learning model to classify network traffic into a particular type of traffic, and based on the network traffic, using a second machine learning model to determine a particular identity for the application.

[0104] In some implementations, the unknown application of a specific type and identity is included in a set of unknown applications of a specific type, and the second security policy is associated with the set of unknown applications of the specific type. In some implementations, the network device may accept the following to obtain an accepted classification: a first classification when the first classification identifies the network traffic as being associated with a specific application or a second classification when the first classification identifies the network traffic as being associated with an unknown application. In this case, the network device may process the network traffic according to the first security policy or the second security policy based on the accepted classification.

[0105] In some implementations, the specific application may be a first specific application, and the second classification may identify the application as an unknown application of a specific type and identity or a second specific application. In this case, when accepting to obtain the accepted classification, the network device may accept the first classification of the network traffic as being related to the first specific application when the first specific application corresponds to the second specific application, accept the first classification of the network traffic as being related to the first specific application when the network traffic is not encrypted, or accept the second classification of the network traffic as being related to the second specific application when the network traffic is encrypted. When processing the network traffic, the network device may process the network traffic according to a first security policy associated with the first specific application, a second security policy associated with an unknown application of a specific type and identity, or a third security policy associated with the second specific application.

[0106] In some implementations, the network traffic may be first network traffic, and processing the network traffic may be based on a second security policy associated with an unknown application of a specific type and identity. In this case, the network device may receive second network traffic for an application after processing the first network traffic, and determine a third classification for the second network traffic based on the first classification technique, wherein the third classification identifies the second network traffic as being associated with a specific application. In addition, the network device may associate a second security policy (associated with an unknown application of a specific type and identity) with the specific application based on the third classification, and process the second network traffic based on the second security policy.

[0107] although Figure 5 Example blocks of process 500 are shown, but in some implementations, Figure 5Process 500 may include additional blocks, fewer blocks, different blocks, or blocks that are arranged differently than the blocks depicted. Additionally or alternatively, two or more blocks of process 500 may be performed in parallel.

[0108] Figure 6 is a flow chart of an example process 600 for classifying unknown network traffic. In some implementations, Figure 6 One or more process blocks of may be performed by a network device (e.g., network device 310). In some implementations, Figure 6 One or more process blocks of may be performed by another device or device group that is separate from or includes the network device, such as a server device (eg, server device 320), a training platform (eg, training platform 340), and the like.

[0109] like Figure 6 As shown, process 600 may include receiving first network traffic for an application (block 610). For example, a network device (eg, using input component 405, switch component 410, controller 420, etc.) may receive first network traffic for an application, as described above.

[0110] like Figure 6 As further shown, process 600 may include determining a first classification for the first network traffic according to the first classification technique, wherein the first classification identifies the first network traffic as being associated with an unknown application of a particular type and identity (block 620). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine the first classification for the first network traffic according to the first classification technique, as described above. In some implementations, the first classification identifies the first network traffic as being associated with an unknown application of a particular type and identity.

[0111] like Figure 6 As further shown, process 600 may include processing the first network traffic according to a security policy associated with an unknown application of a particular type and identity based on the first classification (block 630). For example, a network device (e.g., using switch component 410, output component 415, controller 420, etc.) may process the first network traffic according to a security policy associated with an unknown application of a particular type and identity based on the first classification, as described above.

[0112] like Figure 6 As further shown, process 600 may include receiving second network traffic for the application after processing the first network traffic (block 640). For example, the network device (e.g., using input component 405, switch component 410, controller 420, etc.) may receive the second network traffic for the application after processing the first network traffic, as described above.

[0113] like Figure 6 As further shown, process 600 may include determining a second classification for the second network traffic according to a second classification technique, wherein the second classification identifies the second network traffic as being associated with a particular application (block 650). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine a second classification for the second network traffic according to the second classification technique, as described above. In some implementations, the second classification identifies the second network traffic as being associated with a particular application.

[0114] like Figure 6 As further shown, process 600 may include associating a security policy associated with an unknown application of a particular type and identity with the particular application based on the second classification (block 660). For example, a network device (e.g., using switch component 410, controller 420, etc.) may associate a security policy associated with an unknown application of a particular type and identity with the particular application based on the second classification, as described above.

[0115] like Figure 6 As further shown, process 600 may include processing the second network traffic according to the security policy (block 670). For example, the network device (eg, using input component 410, output component 415, controller 420, etc.) may process the second network traffic according to the security policy, as described above.

[0116] Process 600 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in conjunction with one or more other processes described elsewhere herein.

[0117] In some implementations, the first classification technique can be based on multiple machine learning models. In some implementations, the first classification technique can be based on machine learning and the second classification technique can be based on deep packet detection.

[0118] In some implementations, the first classification technique may include: using a first machine learning model to classify the first network traffic as a specific type of traffic, and using a second machine learning model to determine a specific identity for the application based on the first network traffic. In some implementations, the specific identity for the application may be based on one or more features of the first network traffic.

[0119] In some implementations, the first classification technique may be based on one or more first features of the first network traffic and the second classification technique may be based on one or more second features of the second network traffic, and at least one feature may not be shared by the one or more first features and the one or more second features.

[0120] In some implementations, the network device may generate a record that associates an unknown application of a specific type and identity with a specific application, wherein the record allows the unknown application of the specific type and identity to be identified as the specific application.

[0121] although Figure 6 Example blocks of process 600 are shown, but in some implementations, Figure 6 Process 600 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks than those depicted. Additionally or alternatively, two or more blocks of process 600 may be performed in parallel.

[0122] Figure 7 is a flow chart of an example process 700 for classifying unknown network traffic. In some implementations, Figure 7 One or more process blocks of may be performed by a network device (e.g., network device 310). In some implementations, Figure 7 One or more process blocks of may be performed by another device or device group that is separate from or includes the network device, such as a server device (eg, server device 320), a training platform (eg, training platform 340), and the like.

[0123] like Figure 7 As shown, process 700 may include receiving network traffic for an application (block 710). For example, a network device (eg, using input component 405, switch component 410, controller 420, etc.) may receive network traffic for an application, as described above.

[0124] like Figure 7 As further shown, process 700 may include determining a first classification for the network traffic based on one or more first characteristics of the network traffic, wherein the first classification identifies the network traffic as being associated with a specific application or an unknown application (block 720). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine the first classification for the network traffic based on one or more first characteristics of the network traffic, as described above. In some implementations, the first classification identifies the network traffic as being associated with a specific application or an unknown application.

[0125] like Figure 7 As further shown, process 700 may include determining a second classification for the network traffic based on one or more second characteristics of the network traffic, wherein the second classification identifies the network traffic as being associated with an unknown application of a specific identity (block 730). For example, a network device (e.g., using switch component 410, controller 420, etc.) may determine the second classification of the network traffic based on one or more second characteristics of the network traffic, as described above. In some implementations, the second classification identifies the network traffic as being associated with an unknown application of a specific identity.

[0126] like Figure 7 As further shown, process 700 may include accepting the following to obtain the accepted classification: a first classification when the first classification identifies the network traffic as being associated with a specific application or a second classification when the first classification identifies the network traffic as being associated with an unknown application (block 740). For example, a network device (e.g., using switch component 410, controller 420, etc.) may accept the following to obtain the accepted classification: a first classification when the first classification identifies the network traffic as being associated with a specific application or a second classification when the first classification identifies the network traffic as being associated with an unknown application, as described above.

[0127] like Figure 7 As further shown, process 700 may include processing network traffic based on the accepted classification according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific identity (block 750). For example, a network device (e.g., using switch component 410, output component 415, controller 420, etc.) may process network traffic based on the accepted classification according to a first security policy associated with a specific application or a second security policy associated with an unknown application of a specific identity, as described above.

[0128] Process 700 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in conjunction with one or more other processes described elsewhere herein.

[0129] In some implementations, the first classification may be based on a first classification technique using one or more first features, and the second classification may be based on a second classification technique using one or more second features. In some implementations, at least one feature may not be shared by one or more first features and one or more second features. In some implementations, an unknown application of a specific identity may be defined by one or more second features of network traffic. In some implementations, the second classification may identify network traffic as being associated with an unknown application of a specific identity and associated with a specific type of traffic.

[0130] In some implementations, the network traffic may be first network traffic, and the accepted classification may be a second classification for the first network traffic as an unknown application of a specific identity. In this case, the network device may receive second network traffic for an application after processing the first network traffic, and determine a third classification for the second network traffic based on one or more first features of the second network traffic, wherein the third classification identifies the second network traffic as being associated with a specific application. In addition, the network device may associate a second security policy (associated with an unknown application of a specific identity) with the specific application based on the third classification, and process the second network traffic based on the second security policy.

[0131] although Figure 7 Example blocks of process 700 are shown, but in some implementations, Figure 7 Process 700 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks than those depicted. Additionally or alternatively, two or more blocks of process 700 may be performed in parallel.

[0132] The foregoing disclosure provides illustration and description, but is not intended to be exhaustive or to limit the implementation to the precise form disclosed. Modifications and variations may be made in light of the above disclosure or may be acquired from practice of the implementation.

[0133] As used herein, the term "component" is intended to be broadly interpreted as hardware, firmware, and / or a combination of hardware and software.

[0134] As used herein, the term traffic or content may include a collection of packets. A packet may refer to a communication structure for transmitting information, such as a protocol data unit (PDU), a network packet, a datagram, a segment, a message, a block, a cell, a frame, a subframe, a time slot, a symbol, a portion of any of the above, and / or another type of formatted or unformatted data unit capable of being transmitted via a network.

[0135] As used herein, satisfying a threshold may refer to a value being greater than a threshold, more than a threshold, above a threshold, greater than or equal to a threshold, less than a threshold, less than a threshold, below a threshold, less than or equal to a threshold, equal to a threshold, etc., depending on the context.

[0136] It will be apparent that the systems and / or methods described herein may be implemented in various forms of hardware, firmware, or a combination of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods does not limit the implementation. Thus, the operation and behavior of the systems and / or methods are described herein without reference to specific software code—it is understood that software and hardware can be designed to implement the systems and / or methods based on the description herein.

[0137] Even though particular combinations of features are recited in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of the various implementations. In fact, many of these features may be combined in ways not specifically recited in the claims and / or disclosed in the specification. Although each dependent claim listed below may directly depend on only one claim, the disclosure of the various implementations includes each dependent claim in combination with every other claim in the claim set.

[0138] Unless explicitly described in this way, the elements, actions or instructions used herein should not be interpreted as key or necessary. Moreover, as used herein, the articles "one" and "an" are intended to include one or more projects, and can be used interchangeably with "one or more". In addition, as used herein, the term "set" is intended to include one or more projects (e.g., related projects, unrelated projects, combinations of related projects and unrelated projects, etc.), and can be used interchangeably with "one or more". In the case of only one project, the phrase "only one" or similar language is used. Moreover, as used herein, the terms "has", "have", "having" etc. are intended to be open terms. Further, unless otherwise expressly provided, the phrase "based on" is intended to mean "based at least in part on".

Claims

1. A network device, comprising: one or more memories; as well as One or more processors, the one or more processors configured to: Receive network traffic; Determine the classification of the network traffic according to the classification technology, wherein the classification technique is a machine learning technique that identifies the network traffic as being associated with a particular type of unknown application, and wherein the specific type corresponds to a type of traffic associated with a plurality of unknown applications, wherein the plurality of unknown applications include the unknown application; and The network traffic is processed according to a policy associated with the unknown application of the particular type.

2. The network device of claim 1, wherein the unknown application of the specific type occurs when a signature of the network traffic does not correspond to an application signature known to the network device.

3. The network device of claim 1 , wherein the one or more processors are further configured to: generating a record associated with the unknown application of the particular type, The record is used to permit identification of the unknown application of the specific type.

4. The network device according to claim 1, wherein the unknown application of the specific type comprises one or more of the following: a streaming video associated with said unknown application of said particular type, a file upload associated with said unknown application of said particular type, or An email associated with the unknown application of the particular type.

5. The network device of claim 1, wherein the network device comprises a machine learning classifier component to perform the classification technique, Wherein the machine learning classifier component comprises one or more of the following: Apply the classifier, Traffic type classifier, or Apply the identity classifier.

6. The network device of claim 1, wherein the network device comprises a machine learning classifier component and a classification result evaluator component, and The one or more processors are further configured to: Providing one or more outputs to the classification result evaluator component via the machine learning classifier component; and A policy to be implemented with respect to the communication is determined based on the one or more outputs.

7. The network device of claim 1, wherein the one or more processors are further configured to: determining an identifier based on a pattern for marking the unknown application of the particular type; and The identifier is associated with the policy associated with the unknown application of the particular type.

8. A method comprising: Network traffic is received by the network device; The network device determines the classification of the network traffic according to the classification technology, wherein the classification technology is a machine learning technology that identifies the network traffic as being associated with an unknown application of a specific type and a specific identity, wherein the specific type corresponds to a type of traffic associated with a plurality of unknown applications, and wherein the specific identity distinguishes between multiple unknown applications of the same type, wherein the plurality of unknown applications include the unknown application; as well as The network traffic is processed by the network device according to a policy associated with the unknown application.

9. The method of claim 8, wherein the specific identity of the unknown application is defined by one or more characteristics of the network traffic.

10. The method of claim 8, wherein the policy is a security policy that allows traffic associated with the unknown application.

11. The method of claim 8, wherein the unknown application occurs when a signature of the network traffic does not correspond to an application signature known to the network device.

12. The method of claim 8, wherein the network device includes a machine learning classifier component to perform the classification technique, Wherein the machine learning classifier component comprises one or more of the following: Apply the classifier, Traffic type classifier, or Apply the identity classifier.

13. The method according to claim 8, further comprising: Identify one or more characteristics related to one or more of the following: the flow of the network traffic, or encryption of said network traffic; as well as The one or more features are processed using the machine learning technique to determine the classification for the network traffic.

14. The method according to claim 8, wherein the unknown application is included in a group of unknown applications of the specific type and the specific identity, and The policy is associated with the group of unknown applications of the specific type and the specific identity.

15. A non-transitory computer-readable medium storing instructions, the instructions comprising: One or more instructions that, when executed by one or more processors, cause the one or more processors to: Receive network traffic; determining a category for the network traffic according to one or more characteristics of the network traffic, wherein the classification is determined using machine learning techniques that identify the network traffic as being associated with an unknown application of a particular type and a particular identity, wherein the specific type corresponds to a type of traffic associated with a plurality of unknown applications, wherein the plurality of unknown applications include the unknown application, and wherein the specific identity distinguishes between multiple unknown applications of the same type; and The network traffic is processed according to a policy associated with the unknown application of the particular type and the particular identity.

16. The non-transitory computer-readable medium of claim 15, wherein the unknown application of the specific type and the specific identity is defined by one or more characteristics of the network traffic.

17. The non-transitory computer-readable medium of claim 15, wherein the policy blocks traffic associated with the unknown application of the particular type and the particular identity.

18. The non-transitory computer-readable medium of claim 15, wherein the unknown application occurs when a signature of the network traffic does not correspond to an application signature known to the network device.

19. The non-transitory computer readable medium of claim 15, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to: The one or more characteristics are determined to be associated with one or more of the following: the flow of said network traffic, or encryption of said network traffic; and The one or more features are processed using a machine learning model to determine a classification for the network traffic.

20. The non-transitory computer readable medium of claim 15, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to: determining an identifier based on a pattern for marking the unknown application of the specific type and the specific identity; and The identifier is associated with a policy associated with the unknown application of the particular type and the particular identity.