KR102649741B1 - Optimized route allocation system using artificial intelligence-based multimedia multipath - Google Patents

Abstract

The optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention includes a gateway server that collects path quality information for a plurality of network paths; MMMP manager server that collects user profiles and service profiles; And receiving path quality information from the gateway server, receiving a user profile and a service profile from the MMMP manager server, predicting a saturated path through an artificial intelligence algorithm, and then generating policy information assigning a transmission path for the service profile. MMMP policy control server that transmits the relevant policy information to the gateway server; prevents saturation of a specific network through load distribution and quickly secures an alternative path in the event of a network failure, thereby providing the maximum transmission possible in the entire network It has the effect of achieving efficiency.

Description

Optimized route allocation system using artificial intelligence-based multimedia multipath {OPTIMIZED ROUTE ALLOCATION SYSTEM USING ARTIFICIAL INTELLIGENCE-BASED MULTIMEDIA MULTIPATH}

The present invention relates to an optimized path allocation system using artificial intelligence-based multimedia multipath, and more specifically, quality information of network paths (wired Internet network, Wi-Fi, 5G, LTE, microwave network, satellite network, etc.) An artificial intelligence-based multimedia system that collects and analyzes the network optimization path using machine learning and distributes the service profile to each of the different network paths by considering load distribution to ensure smooth data transmission. This relates to an optimized path allocation system using multiple paths.

The information and communication technologies of the 4th Industrial Revolution provide a variety of network services such as high-speed, hyper-connectivity, and intelligence. The conventional technology is a method of communicating using a single medium (Active-Backup). Commercial networks provide individual communication services by selecting a route for each terminal, and in the case of disaster networks, only fixed routes are used for individual response by each organization. Information sharing is insufficient and the defense network utilizes a single medium. These prior technologies have limitations in providing various network services such as high-speed and hyper-connectivity.

In particular, communication network services that are responsible for ultra-high speed, hyper-connectivity, and intelligence require high survivability and efficiency for connecting multiple tasks and mission elements and for smooth distribution of information.

To realize these capabilities, it is necessary to build a network capable of improving operation efficiency based on a hyper-connected network and supporting multi-layer integrated communication such as defense mobile, satellite, M/W (Microwave), and wired/wireless.

However, major communication infrastructures that form the basis of society, such as public and national defense, build spare networks in preparation for failures in communication lines in use, but there is a problem in that these spare networks are idle and wasted in normal times.

There is also a lack of communication network technology that can connect and utilize available wired and wireless communication facilities to respond quickly when a failure occurs in the country's major communication facilities.

Republic of Korea Patent Publication No. 10-2292444 (2021.08.17)

In consideration of the problems and circumstances described above, the present invention is an artificial intelligence-based transmission technology using the available communication wired and wireless infrastructure multimedia simultaneously to distribute the load in case of traffic congestion and the RNN (Recurrent Neural Networks) algorithm, and network layer analysis. The purpose is to provide an optimized route allocation system using multi-media multi-path based on artificial intelligence applying fault detection technology.

In addition, in consideration of the problems and circumstances described above, the present invention can link and utilize available wired and wireless communication media to quickly respond when a failure occurs in the country's major basic communication facilities such as the national commercial network, disaster network, and military communication network. By adding a new UI (User Interface) function and a fault detection function to the Multi-Media Multi-Path (MMMP) communication network technology that enables real-time situation response in a dynamic, mission-oriented network environment. The purpose is to provide an optimized route allocation system using artificial intelligence-based multimedia multipath.

In order to achieve the above-described object, an optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention includes a gateway server that collects path quality information for a plurality of network paths; MMMP manager server that collects user profiles and service profiles; And receiving path quality information from the gateway server, receiving a user profile and a service profile from the MMMP manager server, predicting a saturated path through an artificial intelligence algorithm, and then generating policy information assigning a transmission path for the service profile. and an MMMP policy control server that transmits the corresponding policy information to the gateway server.

In addition, in order to achieve the above-mentioned purpose, the MMMP policy control server of the artificial intelligence-based multi-media multi-path optimized path allocation system according to the present invention saturates the user profile, service profile information, and path quality information by learning them with a learning algorithm. It is characterized by predicting the path.

In addition, in order to achieve the above-mentioned purpose, the MMMP policy control server of the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the present invention uses a recurrent neural network (RNN) to saturate the above through time series information. It is characterized by predicting the path.

In order to achieve the above-mentioned purpose, the MMMP policy control server of the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention communicates with the MMMP manager server to receive user profile, service profile, and MMMP gateway device information. A first interface for: a second interface for communicating with the gateway server to collect network path quality information; a distributed database for storing and managing the network path quality information; Relational database for storing and managing user profile information and service profile information; a control framework for linking the distributed database, relational database, first interface, and second interface; And a learning module for predicting network saturation through learning with a preset algorithm.

In order to achieve the above-mentioned purpose, the learning module of the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the present invention is characterized by predicting the expected RTT using the TensorFlow library when gateway information and RTT information are read. Do it as

In order to achieve the above-mentioned purpose, the learning module of the optimized route allocation system using artificial intelligence-based multimedia multipath according to the present invention uses the number of recent N RTTs as input values to determine the RTT (Round Trip Time: packet round trip time) It is characterized by predicting. An optimized route allocation system using multi-media multi-path based on artificial intelligence.

In order to achieve the above-mentioned purpose, the learning module of the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention uses the most recent N RTT sequences for each path as input when determining the network path, and uses RNN to predict future network paths. It is characterized by predicting RTT.

In order to achieve the above-mentioned purpose, the MMMP manager server of the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention links with the Protector service for authentication processing (token issuance) for access control. It is characterized by

The optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention is an optimal network that takes into account the bandwidth, delay, cost, and security characteristics of individual networks as a whole in a structure where heterogeneous networks with various characteristics of transmission nodes coexist. Connection priority management technology that helps connections allows for improved communication speeds and network load distribution by utilizing multiple communication media, and has the effect of improving network survivability through communication with minimum common specifications.

In addition, the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention prevents saturation of a specific network through load distribution and quickly secures an alternative path in a network failure situation, thereby providing the maximum possible path allocation for the entire network. There is an effect of achieving transmission efficiency.

Figure 1 is a diagram illustrating the configuration of an optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention.
Figure 2 is a block diagram of the MMMP policy control server of the optimized path allocation system using artificial intelligence-based multimedia multipath according to the invention.
Figure 3 is a diagram illustrating the learning module of the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the invention.
Figure 4 is a user/user profile creation and policy path allocation relationship table used in the artificial intelligence-based, multi-media, multi-path optimized path allocation system according to the invention.
Figure 5 is a diagram showing the configuration of the KOREN interworking test network implementing the optimized path allocation system using artificial intelligence-based multimedia multipath according to the invention.
Figure 6 is a screenshot of the policy information UI screen of the artificial intelligence-based optimized route allocation system using multi-media multi-path according to the invention.
Figure 7 is a network path status monitoring graph and RTT numerical screen of the optimized path allocation system using artificial intelligence-based multimedia multipath according to the invention.
Figure 8 is a screen capture diagram before and after applying a load to path 1 of the artificial intelligence-based multi-media multi-path optimized path allocation system according to the invention.
Figure 9 is a diagram of a monitoring UI screen capture when load saturation of LAMPAD equipment of the optimized path allocation system using artificial intelligence-based multi-media multi-path according to the invention.
Figure 10 is a log file diagram showing the average and predicted values of RTT in the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the invention.
Figure 11 is a diagram capturing the path allocation distribution ratio before and after applying a load to the optimized path allocation system using artificial intelligence-based multi-media multi-path according to the invention.

Terms or words used in this specification and claims should not be construed as limited to their common or dictionary meanings, and the inventor may appropriately define the concept of terms in order to explain his or her invention in the best way. Based on the principle that there is, it must be interpreted with a meaning and concept consistent with the technical idea of the present invention.

Therefore, the embodiments described in this specification and the configurations shown in the drawings are only one of the most preferred embodiments of the present invention and do not represent the entire technical idea of the present invention, so they can be replaced at the time of filing the present application. It should be understood that various equivalents and variations may exist.

below. Referring to the attached drawings, the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention will be described in detail.

Network environments that require absolute connectivity and survivability, such as defense networks or disaster networks, require networking technology that supports high reliability and high survival, enabling real-time situational response by utilizing all idle communication infrastructure resources.

To this end, the optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention recognizes the bandwidth situation by utilizing heterogeneous communication media, increases performance and efficiency through distributed transmission of data, and improves performance and efficiency through distributed transmission of data. The goal is to secure user network technology that transmits data by selecting the underlying media.

Figure 1 is a configuration diagram of an optimized path allocation system using artificial intelligence-based multimedia multipath according to the present invention.

As shown in Figure 1, the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the present invention includes a physically separated gateway server 100, an MMMP manager server 200, and an MMMP policy control server 300. Includes.

The gateway server 100 integrates and transmits networks.

The MMMP manager server 200 manages user and service information and gateway setting information.

In addition, the MMMP manager server 200 is linked with the MMMP policy control server 300 and policy decision information, and provides information on equipment information, policy information, etc. to the corresponding gateway server 100 at the request of the gateway server 100. ) is provided.

Additionally, the MMMP manager server 200 connects with the Protector service for authentication processing (token issuance) for access control.

The MMMP policy control server 300 performs network optimization through an artificial intelligence algorithm.

The description will focus on the MMMP policy control server 300 in the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the present invention.

The MMMP policy control server (Policer: 300) is a component that determines the transmission policy in the MMMP system.

The MMMP policy control server 300 receives the user and service profile input from the MMMP manager server 200, and allocates an efficient path to the corresponding service profile based on the received user and service profile.

The path allocation relationship table based on user profile and service profile creation and policy information is shown in FIG. 4.

Policy information on which the MMMP policy control server 300 allocates an efficient path to the service profile is transmitted to the MMMP manager server 200.

When the MMMP manager server 200 transmits the policy information to the gateway server 100, the gateway server 100 allows different network information to be used according to the policy information for each service profile.

That is, the gateway server 100 performs different path selection according to policy information for each service profile.

In order to determine the transmission policy of the service profile described above, the MMMP policy control server 300 collects and analyzes quality information of each network path (Wi-Fi, 5G, LTE, etc.) from the gateway server 100, The quality of the network path is predicted using machine learning, and the service profile is distributed to each of the different network paths by considering load distribution.

The detailed configuration of the MMMP policy control server 300 is as shown in FIG. 2.

The MMMP policy control server 300 collects user and service profile information from the MMMP manager server 200 and path quality information from the gateway server 100.

Thereafter, the MMMP policy control server 300 predicts a saturated path by learning the collected user and service profile information and path quality information using a learning algorithm.

At this time, the MMMP policy control server 300 preferably predicts saturation from time series information using a recurrent neural network (RNN) as an algorithm for predicting path saturation.

Also, at this time, it is desirable that the recommendation algorithm that ultimately recommends the path used by the MMMP policy control server 300 is recommended based on rules by considering both the path preference obtained from the recurrent neural network and the operator requirements. .

That is, since various rules in defense communication network operation may be applied, such as that some service information must be distributed only through a specific network, the MMMP policy control server 300 refers to an artificial intelligence-based algorithm that selects a network path through learning. Network path recommendation results are derived using a rule-based algorithm.

The MMMP policy control server 300 derives a selection result from a path different from the path predicted to be saturated, stores it in a policy profile, and then transmits the policy profile to the MMMP manager server 200.

The MMMP policy control server 300 includes a first interface 310 for communicating with the MMMP manager server 200 to receive user and service profiles and MMMP gateway device information, and a gateway server (310) to collect network path quality information. It includes a second interface 320 for communicating with 100).

In addition, the MMMP policy control server 300 further includes a distributed database 330 for storing the network path quality information and a relational database 340 for storing user profile and service profile information.

The relational database 340 stores and manages user and service profiles, route information, and gateway information.

In addition, the MMMP policy control server 300 further includes a control framework 250 for linking the distributed database 330, the relational database 340, the first interface 310, and the second interface 320. do.

The MMMP policy control server 300 further includes a learning module 360 storing an algorithm for predicting network saturation through learning.

The algorithm stored in the learning module 360 may be written in tensorflow language.

As shown in FIG. 2B, the distributed database 330 and the relational database 340 use hbase and MariaDB, respectively, and are linked by the spring framework.

The tensorflow library for the above learning runs on Python, so it uses a cross-language compiler (thrift) to retrieve network quality data.

When the learning module 360 reads GW (gateway) information and RTT (Round Trip Time) information, it predicts the RTT using the TensorFlow library.

At this time, it is preferable that the learning module 360 predicts the RTT using the number of recent N RTTs as an input value.

That is, as shown in FIG. 3, when determining a network path (Wi-Fi, 5G, LTE, etc.), the learning module 360 uses the most recent N RTT sequences for each path as input and uses RNN to determine future network paths. Predict RTT.

The learning module 360 inputs N+1 RNN sequences used as input values into a batch queue, which is a queue for each job, and determines the size of the batch queue. When is above a certain level, a random sequence is selected from the queue and RNN learning is performed.

In addition, the MMMP policy control server 300 may further include a Web GUI service platform 370 for selecting a probing path and visualizing probing results.

An experiment was conducted by implementing an optimized path allocation system environment using artificial intelligence-based multimedia multipath according to the present invention having the configuration described above.

Using the hyper-connected intelligent research and development network (KOREN) of AI_Network_Lab operated by the Korea Internet & Security Agency (NIA), the entire KOREN test network was configured as shown in Figure 5 and interconnection and verification tests were performed.

To configure the interconnection test, one optical line is provided and installed from AI_Network_Lab, and network quality information is collected for an optimized path allocation test through network policy control, which is the purpose of the present invention. Based on this, the network status is learned to determine the path format.

By additionally installing the gateway server 100 at both ends of Seoul and Pangyo (Path 2), which predicts saturation and recommends routes with less saturation, it was possible to overcome line limitations.

In addition to the gateway server 100, the MMMP policy control server (Policer Server: 300) and the MMMP manager server (Manager Server: 200) are installed at the host agency in Seoul, and the gateway server (100) and the network at both ends of the network are installed at the joint research agency. The test was performed by installing LAMPAD, a packet collection device that detects failures.

Table 1 below shows the test items and target values.

Test Items target value One Control support through policy profile Confirmed or not 2 Network quality measurement cycle >30sec 3 AI learning result prediction rate MAE<10 4 Optimal network path allocation distribution ratio 0.01~0.99

In order to conduct an experiment to implement an optimized path allocation system environment using artificial intelligence-based multimedia multipath according to the present invention, interconnection and verification tests were performed using the hyper-connected intelligent research and development network (KOREN) of the AI_Network_Lab.

To check the operation of the MMMP policy control server 300, the code operation itself using a combination of command line-based and web-based UI (User Interface) is mainly checked through the command line, and the visualization operation is checked through the web. did.

Compared to generally supporting control with a single service profile, the artificial intelligence-based optimized path allocation system using multi-media multi-path according to the present invention confirms the network path identification support function by supporting policy profile control according to the relationship between user profile and service profile. A test was conducted.

First, the request items for registering a service profile in Table 2 below and the request items for registering a user profile in Table 3 are entered into the UI screen, and as shown in FIG. 6, the user enters the service profile ( By referring to the SERVICE_NAME value of (Profile_service) and finding the path_id (policy identification id) within the policy definition of the policy profile, we confirmed that straight path (Path 1) and detour path (Path 2) control support is possible.

item Service
Name Protocol
Num MPTCP
Ability ALT
ability LB
Ability Secure
Req User
Rank One Path1 101 DISABLE ENABLE ENABLE ENABLE D 2 Path2 101 ENABLE ENABLE ENABLE ENABLE D

<Request items for registering service profile>

item User Name User IP Device User Rank Mission Service One Agent1 45.248.74.102 Agent1 D test Path1/Path2 2 Agent2 45.248.74.114 Agent2 D test Path1/Path2

<Request items for user profile registration>

As a result of checking the policy by looking for "path id" in "Policy Define" by referring to the "SERVICE_NAME" value of the Service Profile above, by comparing Path 1 and 2, the relatively congested path is Red, and the smooth path is Blue. It is displayed as, and the Policy info UI screen according to the policies of Path1 (straight path) and Path2 (detour path) could be checked.

The standard for measuring network quality in the MMMP system is the network parameter RTT, and the RTT collection status is monitored through RTT information collection (within a cycle of 30 seconds) when performing a session between gateways (Path1 Gateway1-Gateway2, Path2 Gateway3-gateway4).

Figure 7 shows the results of checking the Path 1 status monitoring result graph, the exact RTT values (RTT response speed and number) and period (10 seconds) of the graph indicators, and the Probe Statistic Monitoring UI screen.

The Probe Statistic Monitoring UI screen shows the path prediction values of Path1 and Path2, with a congested path displayed in red and a smooth path displayed in blue.

The MMMP policy control server 300 uses machine learning to predict the next RTT based on the latest RTT results.

And the predictions were tested and evaluated through artificial intelligence learning.

As shown in Figure 8, the screen is shown before and after loading path 1. For this test, after connecting to the MMMP policy control server (300), the source client (iperf -c) executes a traffic generation command process toward the destination server (iperf -s) to artificially increase the RTT value by putting a load on Path1. The command was executed and learning started as shown in [Table 4] below.

policer@policer:~$ cd ml/gw1
policer@policer:~/ml/gw1$ /start
gateway1@gateway1:~$ iperf -c 4524874107 -t 70

Performance results Figure 9 shows a UI screen that monitors real-time network traffic status changes. The network path status monitoring UI graph screen of Path 1 is shown at the bottom of FIG. 9.

It was confirmed that traffic occurred in the hat-shaped section for 70 seconds during which the command was given. The predicted values at the top of Figure 8 were 7,78828 μs for Path 1 and 6,72928 μs for Path 2. The path of Path 1 was congested and the path of Path 2 was relatively smooth.

After returning to the pre-load state after the load state for 70 seconds, the predicted path at the bottom of Figure 8 was 1,76188 μs for Path1 and 3,36335 μs for Path2, showing a smooth return to Path 1.

In addition, as shown in Figure 9, the failure detection equipment LAMPAD (packet collection equipment) accurately monitors and displays the time when packet traffic appears under load.

Next, we calculated the prediction performance using the average actual and predicted values of the 20 recently collected RTTs. We checked whether RTT (Round Trip Time) prediction performance satisfied the specified prediction performance metric criteria (MAE < 10). The standard here is the MAE (Mean Absolute Error) value, where error refers to the difference between the value predicted by the algorithm and the actual value.

The MAE formula for checking the RTT prediction performance is as shown in [Equation 1] below.

The procedure converts the difference between the actual value and the predicted value into an absolute value and then adds them to obtain the average. To calculate the next MAE, it is calculated based on the 20 recently collected actual and predicted values. Table 5 shows the average MAE value for the difference between the actual and predicted values of the 20 RTTs of Path 1 and Path 2.

Path 1 is 1.29 and Path 2 is 5.19, showing that the performance target value to be measured (MAE<10) is less than 10. The smaller the MAE, the better the performance of the algorithm.

Path 1 Path 2 actual value predicted value Difference (absolute value actual value predicted value Difference (absolute value 1026.45 1024.19 2.26 1644.05 1646.64 2.59 1026.35 1023.76 2.25 1642.50 1641.52 0.98 1026.45 1024.22 2.23 1643.95 1645.34 1.39 1026.50 1023.88 2.26 1643.45 1633.79 9.66 1026.60 1028.05 1.45 1652.00 1652.39 0.36 1026.55 1028.28 1.73 1651.95 1649.94 2.01 1027.35 1028.05 0.70 1651.25 1650.60 0.65 1027.40 1029.19 1.79 1651.30 1651.47 0.17 1029.45 1028.67 0.78 1649.50 1654.72 5.22 1027.50 1028.90 1.40 1647.85 1656.00 8.15 1027.75 1028.45 0.70 1645.25 1650.36 5.11 1027.90 1029.21 1.31 1645.30 1646.48 1.18 1027.95 1028.60 0.65 1647.45 1649.11 1.66 1027.75 1028.76 1.01 1649.15 1659.25 10.10 1027.75 1028.85 1.10 1648.35 1653.40 5.05 1027.65 1028.96 1.31 1648.65 1648.85 0.20 1027.45 1027.11 0.34 1651.20 1651.89 0.69 1027.45 1027.91 0.46 1653.05 1640.62 12.43 1027.60 1027.36 0.24 1652.25 1631.12 21.13 1027.65 1028.84 1.19 1653.60 1638.61 14.99 MAE average value 1.29 MAE average value 5.19

As a final test, calculate the distribution ratio for assigning the optimal network path to the two network paths according to the path allocation distribution formula and check whether the corresponding figure is displayed accurately on the user interface screen.

The distribution ratio is calculated in the range of 0.01 to 0.99, and the sum of the distribution ratios calculated for each two network paths is 1. For this test, AI learning is performed for each network traffic collection path for two network paths (Path 1, Path 2).

Figure 10 is a log file showing the average and predicted values of the most recent 20 RTTs of Path 1. This log file shows the average actual and predicted values of the collected RTT.

Next, calculate the allocation distribution ratio per path by substituting the predicted values for each network path for Path 1 and Path 2 into the path allocation distribution formula.

Figure 11 shows the path allocation distribution ratio before and after loading.

The predicted path allocation distribution formula is as follows. In order to optimize the route, the distribution vector value is determined as the inverse ratio of RTT _i (weighted average value), and the distribution formula considering network congestion is as shown in [Equation 2] below.

In [Equation 2] above, W _i is the number of services to be allocated, N _p is the total number of services, and COST _i is the reciprocal of the RTT prediction value of path i.

If only the path allocation ratio is calculated from the path allocation ratio of the path 1 and path 2 RTT predictions in the state before the load at the top of FIG. 11, it is as shown in [Equation 3] below.

The allocation ratios of Path 1 and Path 2 according to [Equation 3] above are as [Equation 4] and [Equation 5] below, respectively.

As can be seen from the calculation results of [Equation 4] and [Equation 5] above, the optimal path distribution (allocation) ratio is Path 1 = 0.62, Path 2 = 0.36, resulting in a distribution ratio of about 6:4. .

If the total number of services N _p = 10, it can be expected that 6 services will be allocated to path 1 and 4 to path 2.

As shown in the lower part of FIG. 11, applying the path allocation distribution ratio formula to the path 1 and path 2 RTT estimates in the state after load generation is calculated as 1 (Path 1) = 0.42 and Path 2 (Path 2) = 0.58. Load balancing service is performed with a distribution ratio of approximately 4:6.

Select "Probe" → "Policy Define" from the left menu of the POL-SP tab screen of the MMMP policy control server 300, and select path 1 and path in the "distribution ratio" column of the "Distribution Ratio for the path allocation" list. You can check the allocation distribution ratio per route in 2.

The number displayed on the user interface (UI) is the calculated allocation distribution ratio per route, rounded to three decimal places.

As described above, the ability to allocate an optimized path was confirmed through the AI learning model of the artificial intelligence-based optimized path allocation system using multimedia multipath.

In the above, the technical idea of the present invention has been described along with the accompanying drawings, but this is an exemplary description of a preferred embodiment of the present invention and does not limit the present invention. In addition, it is clear that anyone skilled in the art of the present invention can make various modifications and imitations without departing from the scope of the technical idea of the present invention.

100: gateway server
200: MMMP manager server
300: MMMP policy control server
310: first interface
320: second interface
330: Distributed database
340: Relational database
250: Control framework
260: Learning module
270: Web GUI service platform

Claims (8)

a gateway server that collects path quality information for a plurality of network paths;
MMMP manager server that collects user profiles and service profiles; and
Receives path quality information from the gateway server, receives a user profile and a service profile from the MMMP manager server, predicts a saturated path through an artificial intelligence algorithm, considers load distribution, and assigns a transmission path for the service profile. It includes an MMMP policy control server that generates policy information and transmits the policy information to the gateway server,
The MMMP policy control server is
a first interface for communicating with the MMMP manager server to receive user profile, service profile and MMMP gateway device information;
a second interface for communicating with the gateway server to collect network path quality information;
a distributed database for storing and managing the network path quality information;
Relational database for storing and managing user profile information and service profile information;
a control framework for linking the distributed database, relational database, first interface, and second interface; and
Includes a learning module to predict network saturation through learning with a stored algorithm,
An artificial intelligence-based, multi-media, multi-path optimized route allocation system that ultimately recommends a route based on rules by considering both route preference and operator requirements.
According to clause 1,
The MMMP policy control server is
An optimized path allocation system using artificial intelligence-based multimedia multipath, characterized in that it predicts a saturated path by learning the user profile, service profile information, and network path quality information using a learning algorithm.
In paragraph 2
The MMMP policy control server is
An artificial intelligence-based, multi-media, multi-path optimized path allocation system characterized by predicting the saturated path through time series information using a recurrent neural network (RNN).
delete According to clause 1,
The learning module is
An artificial intelligence-based, multi-media, multi-path optimized path allocation system that predicts the expected RTT using the TensorFlow library when gateway information and RTT information are read.
According to clause 5,
The learning module is
An artificial intelligence-based, multi-media, multi-path optimized path allocation system characterized by predicting the RTT using the number of recent N RTTs as input.
According to clause 5,
The learning module is
An artificial intelligence-based, multi-media, multi-path optimized path allocation system that predicts the RTT for future network paths using RNN, using the most recent N RTT sequences for each path as input when determining a network path.
According to clause 1,
The MMMP manager server is
An optimized path allocation system using artificial intelligence-based multimedia multipath, characterized by linking with the Protector service for authentication processing (token issuance) for access control.