JP6815976B2 - Converter, conversion method and program - Google Patents

Description

The present invention relates to a conversion device, a conversion method and a program.

The application of AI (Artificial Intelligence) technology is being considered in order to enhance the operation of communication networks (NWs). For example, AI (Patent Document 1, Non-Patent Document 1) that calculates an optimum communication path, AI that performs failure detection and failure location identification, and the like have been studied (Non-Patent Document 2). It is expected that AIs for these various uses will be systematized in the future and connected to an operation system (OpS) (Non-Patent Document 3).

The AI system is provided with various input / output interfaces (IFs) depending on the application. When the OpS connects to a plurality of AI systems, it is necessary to individually deal with the IF of each AI system, the development cost is high, and a system capable of flexibly converting the data format is required. As a system that realizes IF conversion, EMS (Element Management System) that absorbs the difference in IF of NW device (NE: Network Element), API gateway that absorbs the difference in IF (API (Application Program Interface)) of application, etc. However, a system that absorbs the difference of the AI system (hereinafter referred to as "AI management system") has not been proposed.

Japanese Unexamined Patent Publication No. 2016-144068

Masahiro Kobayashi, Ryutaro Matsumura, Takumi Kimura, Toshiaki Tsuchiya, Katsutoshi Noritake, "Examination of Resource Allocation Method to Efficient Virtual Network Using Service Capacity", Shingaku Giho, vol. 115, no. 404, NS2015-151, pp. 29-34, January 2016. Ryoichi Kawahara, "Utilization of AI / Machine Learning for Advanced Network Operation and Control Technology", Shinkai Society Conference, BT-2-1, 2017. Kaoruaki Harada, Keishiro Watanabe, Yusuke Nakano, Yasuhiro Ikeda, Yoichi Matsuo, Masahiro Kobayashi, Akihito Suzuki, Ryoichi Kawahara, "Introduction of Initiatives for Realizing Proactive Controlled Networks", Shinkai General Conference, B- 7-32, 2017.

EMS and API gateways are equipped with various functions, but from the viewpoint of IF conversion, basically only data format conversion based on fixed conversion rules is performed.

On the other hand, in the AI management system, in addition to the data format conversion, it is assumed that a conversion function when it is desired to utilize the AI developed for a specific purpose for other similar purposes is also required. For example, if you want to use the AI system that calculates the route that minimizes the usage of bandwidth resources as the AI system that calculates the route that avoids damaged equipment and equipment that is under security attack, the latter AI system is used for calculation. It is necessary to convert the input to the AI system for the former so that the result can be obtained, but such a conversion cannot be realized by the conventional system.

In the above conversion function, the optimum conversion rule is not always known. Therefore, it is difficult to apply a rule that is difficult for humans to notice with the conventional IF conversion technique that gives a fixed conversion rule in advance, and there is a problem from the viewpoint of calculation speed and calculation accuracy. Such a problem is considered to be common not only in the AI system but also in a general information processing system having input and output.

The present invention has been made in view of the above points, and an object of the present invention is to improve the flexibility of conversion rules regarding input / output of a system.

Therefore, in order to solve the above-mentioned problem, the conversion device applies the input data related to the second process to the first system in accordance with the conversion rule for the execution request of the second process to the first system that executes the first process. The first conversion unit, which is converted so as to correspond to the processing and inputs the converted input data to the first system, and the output data from the first system are subjected to the second conversion rule according to the conversion rule. The second conversion unit that converts to correspond to the output data related to the processing, learns the history of the input data before conversion and the output data after conversion, and updates the conversion rule based on the learning result. It has an update unit.

It is possible to improve the flexibility of conversion rules regarding system input / output.

It is a figure which shows the system configuration example in embodiment of this invention. It is a figure which shows the hardware configuration example of the AI management system 10 in embodiment of this invention. It is a sequence diagram for demonstrating an example of the processing procedure executed in this Embodiment. It is a figure for demonstrating the 1st specific example of a problem conversion process. It is a figure for demonstrating the 2nd specific example of a problem conversion process. It is a figure for demonstrating the 3rd specific example of a problem conversion process. It is a figure for demonstrating the 1st specific example of a rule learning process. It is a figure for demonstrating the 2nd specific example of a rule learning process.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing an example of a system configuration according to an embodiment of the present invention. In FIG. 1, NE40 is a group of NW devices (NE: Network Element) such as routers and switches, and OpS30 is a device (OpS: Operation System) that controls the entire NW.

Each AI system 20 of the AI systems 20-1 to N (hereinafter referred to as "AI system 20" when they are not distinguished) is a process for enhancing some functions of NW operation (for solving a problem). It is a system that executes calculation). In this embodiment, a plurality of AI systems 20 are prepared for each application. However, only one AI system 20 may be used.

The AI management system 10 interprets the contents of the problem (input) to be solved by the desired AI, and the problem can be solved by the existing AI system 20 (similar AI system 20) having an application similar to the desired AI (similar AI system 20). By converting to (input) (hereinafter referred to as "problem conversion") and having the similar AI system 20 calculate the converted problem, a solution equivalent to the desired solution (output) can be obtained from the similar AI system 20. One or more computers that enable it.

In FIG. 1, the AI management system 10 is an IF conversion unit 11 and a conversion rule learning unit for each AI system 20 of the IF conversion units 11-1 to N (hereinafter, when they are not distinguished, they are referred to as “IF conversion unit 11”). 12 and conversion rule DB 13 (Data Base), etc. are included.

In the conversion rule DB 13, for each input / output data related to each AI system 20, conversion rules for realizing the above problem conversion are stored in each IF conversion unit 11. Each conversion rule is periodically delivered to the IF conversion unit 11 corresponding to the conversion rule. The conversion rule is described in a program format as described later. The initial value (initial rule) of the conversion rule is given by the user.

Each IF conversion unit 11 performs problem conversion so as to correspond to the AI system 20 regarding the input / output of the AI system 20 corresponding to the IF conversion unit 11 according to the distribution conversion rule. The problem conversion specifically refers to input / output data conversion related to the AI system 20. However, the "data" of the "data conversion" referred to here includes not only the data that is the target of the problem (processing) but also the "data" that defines the problem (processing). One conversion rule includes a set of a conversion rule for input data (hereinafter, referred to as "input conversion rule") and a conversion rule for output data (hereinafter, referred to as "output conversion rule").

The conversion rule learning unit 12 learns the input / output tendency of the IF conversion unit 11 in order to enable problem conversion and more efficient calculation by the AI system 20 than when an existing conversion rule such as an initial rule is applied. However, by discovering conversion rules that are difficult for humans to notice and updating existing conversion rules, AI calculation will be speeded up. Trend learning can be achieved using a framework such as supervised machine learning. The input / output of the IF conversion unit 11 means an input to the IF conversion unit 11 and an output from the IF conversion unit 11.

FIG. 2 is a diagram showing a hardware configuration example of the AI management system 10 according to the embodiment of the present invention. The AI management system 10 of FIG. 2 has a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an interface device 105, and the like, which are connected to each other by a bus B, respectively.

The program that realizes the processing in the AI management system 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed in the auxiliary storage device 102 from the recording medium 101 via the drive device 100. However, the program does not necessarily have to be installed from the recording medium 101, and may be downloaded from another computer via the network. The auxiliary storage device 102 stores the installed program and also stores necessary files, data, and the like.

The memory device 103 reads and stores the program from the auxiliary storage device 102 when the program is instructed to start. The CPU 104 executes the function related to the AI management system 10 according to the program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

The IF conversion unit 11 and the conversion rule learning unit 12 in FIG. 1 are realized by a process of causing the CPU 104 to execute one or more programs installed in the AI management system 10. The conversion rule DB 13 in FIG. 1 can be realized by using, for example, an auxiliary storage device 102, a storage device that can be connected to the AI management system 10 via a network, or the like.

Hereinafter, the processing procedure executed in the present embodiment will be described. FIG. 3 is a sequence diagram for explaining an example of the processing procedure executed in the present embodiment. It is assumed that the AI system 20 in FIG. 3 is a specific AI system 20. Further, it is assumed that the IF conversion unit 11 is an IF conversion unit 11 corresponding to the AI system 20. However, the same processing procedure is executed for the other AI system 20 and the IF conversion unit 11.

As a periodic process, when the AI management system 10 is started up, the conversion rule is changed, or the like, the conversion rule DB 13 distributes the conversion rule corresponding to the IF conversion unit 11 to the IF conversion unit 11 (S101). The IF conversion unit 11 reads the distributed conversion rule (S102).

When the request message (problem calculation request) from OpS30 to the AI system 20 arrives at the IF conversion unit 11 (S103), the IF conversion unit 11 performs problem conversion according to the conversion rule (input conversion rule) (S104), and the AI The converted request message is transmitted to the system 20 (S105).

The AI system 20 executes processing according to the converted request message (S106), and transmits a response message including the processing result (calculation result) to the IF conversion unit 11 (S107). Similar to the request message, the IF conversion unit 11 converts the processing result according to the conversion rule (output conversion rule) (S108), and sends the converted response message to OpS30 (S109).

OpS30 interprets the processing result included in the received response message (S110) and reflects the setting (for example, route setting, etc.) for NE40 (S111).

On the other hand, the IF conversion unit 11 that has performed the conversion transmits the conversion history related to the conversion to the conversion rule learning unit 12 (S112). The conversion history is data including a history of input to the IF conversion unit 11 and a history of output from the IF conversion unit 11. The conversion rule learning unit 12 learns the conversion rule based on the conversion history (S113), and based on the learning result, the conversion registered in the conversion rule DB 13 (for the IF conversion unit 11) regarding the conversion. Update the rule (S114). The updated conversion rule is transmitted to the IF conversion unit 11 at the timing of the next periodic update, for example. However, the updated conversion rule may be immediately transmitted to the IF conversion unit 11.

Next, specific examples of the three problem conversion processes will be described.

[Problem conversion process specific example 1]
FIG. 4 is a diagram for explaining a first specific example of the problem conversion process. FIG. 4 shows an example in which the AI system 20 (Patent Document 1 and Non-Patent Document 1) that calculates the route that maximizes the resource utilization efficiency of the NW (network) is used as the AI system that calculates the route that avoids the fault link. It is shown. That is, in the problem conversion process specific example 1, the existing AI system 20 is utilized for a problem to which a constraint condition such as avoiding a failure link is added (the constraint condition is changed).

When calculating the route so as not to use the failed link, the route calculation may be performed for the NW topology excluding the link that may have a failure.

In order to carry out such NW topology change processing, the conversion rule DB 13 has a link as an input conversion rule (OpS → AI system conversion rule) in which the failure rate of the input data may be greater than 0%. Is registered from the physical NW topology information. Further, as an output conversion rule (AI system → OpS conversion rule), an output conversion rule for restoring a link deleted at the time of input data conversion is registered in the conversion rule DB 13.

In FIG. 4, the request message to the IF conversion unit 11 includes a calculation request for a route from node A to node B in a certain NW topology and NW topology information. In this case, the IF conversion unit 11 removes the failure link (link marked with x) from the NW topology information in accordance with the OpS → AI system conversion rule for the request message, and sends the request message including the NW topology information. It is transmitted to the AI system 20. Therefore, from the AI system 20, a calculation result (broken line path) avoiding the fault link can be obtained. The IF conversion unit 11 restores the failure link (deleted link) to the NW topology information according to the AI system → OpS conversion rule for the calculation result. The final response to OpS30 is the optimal route around the faulty link.

[Problem conversion process specific example 2]
FIG. 5 is a diagram for explaining a second specific example of the problem conversion process. In FIG. 5, the AI system 20 (Patent Document 1 and Non-Patent Document 1) that calculates the optimum route from a single start point to a single end point is shown in a cloud data center or CDN (Contents Delivery) from a single start point. An example is shown in which the optimum route is calculated with one server in the group of servers scattered in the network as the end point, such as Network). That is, in the problem conversion process specific example 2, it is utilized for a problem in which the constraint condition is changed, such that the candidate of the end point is not limited to one.

Here, it is assumed that a single node is specified as the start point and a plurality of node candidates are specified as the end point as the requirement condition of the route. Further, it is desired to find a route having the highest NW resource utilization rate among the routes from the start point to each end point candidate node.

In graph theory, such problems are often dealt with by adding virtual nodes (virtual nodes) and links (virtual links) to the physical NW topology. Connect virtual nodes to all potential end point nodes with virtual links. The bandwidth of the virtual link is set to ∞ and the delay is set to 0. Then, the optimum route to the virtual node can be obtained by calculating the route with the existing AI system 20 with the virtual node as the end point. After that, the desired result can be obtained by deleting the virtual node and the virtual link from the obtained route.

In order to carry out such a change process of the NW topology and the route request condition, the conversion rule DB 13 has an input conversion rule (OpS → AI system conversion rule), and when the number of end nodes is more than 1, all the end nodes are included. A conversion rule is registered that connects the virtual node S with a virtual link and sets the end point to the virtual node S. Further, as an output conversion rule (AI system → OpS conversion rule), a conversion rule for deleting the virtual link and virtual node added by OpS30 → AI conversion from the physical NW topology information and the route result is registered in the conversion rule DB 13.

In FIG. 5, the request message to the IF conversion unit 11 includes a calculation request for a route in which node A is a start point node and nodes B and C are end point nodes in a certain NW topology, and NW topology information. Is done. In this case, the IF conversion unit 11 connects the end node B and C and the virtual node S with a virtual link in the NW topology information according to the OpS → AI system conversion rule, and sets the end node of the route in the request message to B. And rewrite from C to S. Therefore, the optimum route (broken line route) to the virtual node S can be obtained from the AI system 20. The IF conversion unit 11 deletes the virtual node and the virtual link from the NW topology information and the obtained route according to the AI system → OpS conversion rule. The result finally responded to OpS30 is the route with the highest resource utilization efficiency with node C as the end point among the end point candidate nodes.

[Problem conversion process specific example 3]
FIG. 6 is a diagram for explaining a third specific example of the problem conversion process. FIG. 6 shows an intermediate point designated between the start point and the end point using the AI system 20 (Patent Document 1 and Non-Patent Document 1) that calculates the optimum route from a single start point to a single end point. An example of finding the optimum route of the virtual NW via the above is shown. That is, in the problem conversion process specific example 3, it is utilized for the problem to which the constraint condition such as passing through the intermediate point is added (the constraint condition is changed).

This problem divides the original request into two, a route calculation request from the start point to the intermediate point and a route calculation request from the intermediate point to the end point, and causes the AI system 20 to calculate the route for the two route calculation requests. The resulting pathways may be combined.

In order to carry out such a problem conversion process, if an intermediate point exists (number of intermediate points = 1) as an input conversion rule (OpS → AI system conversion rule) in the conversion rule DB 13, the original route calculation request r is set. , A conversion rule for dividing the route calculation request r1 from the start point to the intermediate point and the route calculation request r2 from the intermediate point to the end point is registered. The physical network conditions of r1 and r2 are the same as r. Further, the required bandwidths of the virtual NWs of r1 and r2 are the same as r. On the other hand, the start point, end point, and request delay of r1 and r2 are changed from r. The request delays of r1 and r2 are set as 1/2 of r because the route is divided into two. Further, as the AI system-> OpS conversion rule, a conversion rule that combines the route obtained for r1 and the route obtained for r2 is registered in the conversion rule DB 13.

In FIG. 6, the request message to the IF conversion unit 11 includes a calculation request r (optimal route calculation of A⇒B⇒C) of a route from node A to node C via node B in a certain NW topology. Request) is included. In this case, the IF conversion unit 11 divides the optimum route calculation request of A⇒B⇒C into the calculation request r1 of A⇒B and the calculation request r2 of B⇒C according to the OpS → AI system conversion rule. Therefore, from the AI system 20, an optimum route (broken line route) can be obtained for each of the calculation requests r1 and r2. The IF conversion unit 11 outputs a desired result by combining each route (broken line route) according to the AI system → OpS conversion rule.

Next, specific examples of the two rule learning processes will be described.

[Specific example 1 of rule learning process]
FIG. 7 is a diagram for explaining a first specific example of the rule learning process. FIG. 7 shows an example in which the conversion rule given in [Specific example 1 of problem conversion processing] is updated by learning.

The IF conversion history, which is the history information of the problem conversion process, is given as the input to the conversion rule learning unit 12. In FIG. 7, it is assumed that the physical network is composed of three sub-networks A, B, and C, and the problem conversion process is performed in the case of a link failure between the sub-networks A and C (a failure of a link assigned with x). The input / output history of is shown.

From the input / output history, in the case of a link failure between the sub-networks A and C, there is a possibility that the virtual NW whose starting point is the node in the sub-network A and the ending point is the node in the sub-network B will pass through the sub-network C. It turns out that there are very few. The conversion rule learning unit 12 learns this tendency with a machine learning algorithm and generates a new conversion rule. The new conversion rule removes the sub-network C from the physical network topology when the failed link and the start and end points of the virtual network meet the above conditions. By doing so, in the AI system 20, when calculating the route, it is possible to avoid searching for a link that is unlikely to be assigned, and it is expected that the calculation speed will be improved.

[Specific example 2 of rule learning process]
FIG. 8 is a diagram for explaining a second specific example of the rule learning process. FIG. 8 shows an example in which the conversion rule given in [Problem conversion process Example 3] is updated by learning.

As an input to the conversion rule learning unit 12, an IF conversion history is given as in [Rule learning process Example 1]. In FIG. 8, the input / output history of the route calculation for the virtual NW of the start point A → the intermediate point B → the end point C is shown. It is assumed that nodes A and B are adjacent to each other on the physical NW, and node C is located at a point farther than nodes A and B.

From the input / output history, it can be seen that the route between B and C is longer than that between A and B. The conversion rule learning unit 12 learns this tendency with a machine learning algorithm and generates a new conversion rule. Specifically, in the conventional conversion rule, when dividing a request, it is divided by [request delay from start point to intermediate point]: [request delay from intermediate point to end point] = 1: 1. The division ratio has been changed to 1: 2. By doing so, it is possible to perform optimal request division compared to the rules initially given by humans, and it is expected that the accuracy of route calculation will be improved.

As described above, according to the present embodiment, the conversion rule is automatically learned and updated. That is, the flexibility of conversion rules regarding system input / output can be improved. Therefore, for example, it is possible to support the efficiency and speed of processing of the AI system 20. Further, by converting the problem, the AI system 20 for a specific purpose can be used for a similar purpose, and the function expansion of the NW operation system in the future can be realized at low cost.

In the present embodiment, the AI management system 10 is an example of a conversion device. The IF conversion unit 11 is an example of a first conversion unit and a second conversion unit. The conversion rule learning unit 12 is an example of an update unit.

Although the embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various aspects are within the scope of the gist of the present invention described in the claims. It can be transformed and changed.

10 AI management system 11 IF conversion unit 12 Conversion rule learning unit 13 Conversion rule DB
20 AI system 30 OpS
40 NE
100 Drive device 101 Recording medium 102 Auxiliary storage device 103 Memory device 104 CPU
105 Interface device B bus

Claims (7)

Regarding the execution request of the second process to the first system that executes the first process, the input data related to the second process is converted so as to correspond to the first process according to the conversion rule, and after the conversion. A first conversion unit that inputs input data to the first system, and
A second conversion unit that converts the output data from the first system so as to correspond to the output data related to the second process according to the conversion rule.
An update unit that learns the history of the input data before conversion and the output data after conversion and updates the conversion rule based on the learning result.
A converter characterized by having. The second process is a process in which the constraint conditions are changed with respect to the first process.
The conversion device according to claim 1, wherein the conversion device is characterized by the above. The first process and the second process are processes for searching for a route in a network.
The conversion device according to claim 1 or 2, wherein the conversion device is characterized by the above. Regarding the execution request of the second process to the first system that executes the first process, the input data related to the second process is converted so as to correspond to the first process according to the conversion rule, and after the conversion. The first conversion procedure for inputting the input data to the first system and
A second conversion procedure for converting the output data from the first system so as to correspond to the output data related to the second process according to the conversion rule.
An update procedure for learning the history of the input data before conversion and the output data after conversion and updating the conversion rule based on the learning result.
A conversion method characterized by a computer performing. The second process is a process in which the constraint conditions are changed with respect to the first process.
The conversion method according to claim 4, wherein the conversion method is characterized by the above. The first process and the second process are processes for searching for a route in a network.
The conversion method according to claim 4 or 5, characterized in that. A program characterized in that a computer functions as each part according to any one of claims 1 to 3.

Images