JP2000049863A - Network model preparation method and device and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the preparation method of a network model as the simulation of the traffic of a packet switching network. SOLUTION: Going and returning routes between a fixed observation point and plural hosts are detected in a going and returning route detection part 111, the required time of the going and returning routes is detected in a going and returning time measurement part 112 and the correlation of the traffic of respective links and the traffic of the entire network is learned by using a neural network 116 based on the data set of the going and returning routes and the going and returning time. By inputting a test vector for specifying an optional route to the learning completed neural network, the expected passing time of the optical route is estimated.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for creating a network model for simulating traffic in a packet-switched network constituted by a plurality of path control devices, and a recording medium storing the program.

[0002]

2. Description of the Related Art In a small-scale packet-switched network such as an intranet (hereinafter simply referred to as "network"), the number of elements (for example, host computers, routers and lines) constituting the network is small. Thus, in such a small-scale network, these components can be defined in advance. Further, in such a small-scale network, a change in the usage can be estimated to some extent.

Conventionally, many methods have been proposed for simulating such small-scale network traffic. In the simulation,
Usually, a network model is created using software that applies queues and probabilities. Specifically, for example, a packet signal is transmitted according to a specific probability distribution on each host, and a communication buffer on each router is considered as a queue. Then, based on the bandwidth of the line and the processing capacity derived from the specifications of the router, the average processing time for each packet signal is obtained in multiple stages for each hop using queuing theory.

[0004]

In a large-scale open packet switching network such as the Internet, network components such as routers and lines belong to many organizations. For this reason, in a large-scale network, it is difficult to grasp the bandwidth of each router or line and the link state at one place.

[0005] In particular, between independent organizations (for example, between commercial providers), a network managed by each organization is not open to the public. For this reason, it is difficult to grasp information on the usage status of the entire network, such as where in the network, how services are used, and how much traffic is generated.

Further, in a large-scale network,
The usage information changes drastically, as well as the grasp of the information on the usage status. Therefore, it is practically impossible to create a model of a large-scale network using a conventional probabilistic model using a queue or the like, and to simulate traffic.

[0007] For this reason, conventionally, for a large-scale network, it has not been useful for predicting traffic using a network model to control a route. In general, large-scale networks are designed and controlled based on intuition based on experience.

The present invention has been made to solve the above problems, and has as its object to provide a network model creation technique as a simulation of a track of a packet-switched network.

[0009]

Means for Solving the Problems (Network Model Creation Method) In order to achieve this object, according to the network model creation method of the present invention, in a packet switching network constituted by connecting a plurality of links,
A data set of a round trip route and a round trip time of a packet signal between a fixed observation point and a plurality of points is obtained, and based on the data set, a traffic correlation between link traffic and traffic of the entire packet switching network. Is learned by knowledge processing, and using the result of learning,
This is a method of creating a network model by estimating the traffic of an arbitrary route in a packet switching network.

As described above, according to the network model creating method of the present invention, a data set of a round-trip route and a round-trip time is collected at a fixed observation point. Then, learning by knowledge processing is performed based on the data set.
As a result of learning by this knowledge processing, it is possible to obtain information on the correlation between the traffic of each link and the traffic of the entire network.

A packet signal passing through a link on a network is usually transferred from another link adjacent to the link. The adjacent link is
It is adjacent to another link. Therefore, the traffic of each link constituting the network has some correlation with each other. That is, traffic on one link affects traffic on another link. Then, this correlation is obtained as a result of learning by knowledge processing. Then, as a result of the learning, for example, when the traffic of a certain link rises, it is possible to estimate the effect on the traffic of another link.

The data set to be subjected to the knowledge processing is only a part of the information indicating the usage status of the network. That is, according to the present invention, it is possible to estimate the traffic of an arbitrary route using the result of the knowledge processing without grasping all the components of the network and all the usage states. As a result, a network model as a traffic simulation can be created.

Further, in the network model creation method according to the present invention, preferably, the traffic correlation is
It is desirable to make a correlation between the required time required for a packet signal to pass through a link and the round-trip time.

The time required for the link is affected by the traffic volume (communication volume) in the network. Therefore,
The correlation between the required time of a certain link and the round trip time reflects the correlation between the traffic volume of the link and the traffic volume of the round trip route.

In the embodiment of the present invention, preferably, a variable corresponding to the required time of the link on the reciprocating route is provided, the number of times of passage of the link is used as a coefficient, and the sum of the variable multiplied by the coefficient is used as the round-trip time. It is preferable that a set of simultaneous equations is established based on the data set, and a set of solutions of the simultaneous equations is obtained by learning.

Incidentally, the number of data sets is usually smaller than the number of links in the network. Therefore, the number of simultaneous equations corresponding to each data set is smaller than the number of variables corresponding to the required time of each link. Therefore,
Usually, it is not possible to uniquely find the solution of the simultaneous equations.

All solutions (time required for each link)
Has a constraint that it takes a value of 0 or more. Therefore, it is possible to solve this simultaneous equation as a linear programming problem using this constraint. However, for example, when a solution is obtained by the simplex method as a linear programming problem, only the value of the solution that can be the maximum value (the link that can maximize the required time) can be obtained from all the solution sets.

On the other hand, the data set on which the simultaneous equations are based is merely a measurement of the state of the network at a certain point in time. On the other hand, network usage changes drastically. In addition, the data set measurements typically include errors. Therefore, it is not practical to calculate only the solution that can be the maximum value as a linear programming problem.

That is, when only one solution is calculated, it is not possible to obtain another solution having a large required time, which is likely to exist in addition to the calculated solution. In addition, the calculated solution itself may be wrong due to an error.
Therefore, rather than calculating one solution, it is practically desirable to estimate, by a knowledge process, a group of candidates for a solution (a ring that may take a long time) that can take a large value with a certain probability. Therefore, in the present invention,
For a problem that can be solved as a linear programming problem,
Apply knowledge processing.

In implementing the present invention, preferably, in learning by knowledge processing, a neural network having an input layer having a node equal to the number of all links and one output layer is used, and all links are learned. A vector of dimensions equal to the number, with the value of the component corresponding to each link on the round trip route being the value obtained by dividing each coefficient value by the maximum coefficient value (the maximum value of the coefficient values) as input data It is preferable to input the teacher time to the neural network corresponding to the round trip time.

When input data based on a data set that is an actually measured value is input, the neural network compares the output with a value of a teacher signal as a sample value that is an actually measured value to correct the neural network. Learning with.

In the network model creation method according to the present invention, when estimating the traffic of an arbitrary route, preferably, the neural network includes a vector having a dimension equal to the number of all links, Input test data obtained by dividing the value of the component corresponding to the link by the number of times of passage of the link by the maximum number of times of passage (the value of the maximum number of times of passage), and as an output of the neural network, It is desirable to obtain an estimated communication time corresponding to the estimated required time of the route.

(Network Model Creation Apparatus) According to the network model creation apparatus of the present invention, the network model creation apparatus is provided at a fixed observation point connected to a packet switching network formed by connecting a plurality of links. A round-trip path detecting unit that detects a round-trip path of a packet signal between a plurality of points; a round-trip time detecting unit that detects a round-trip time required for the packet signal to be transferred along the round-trip path; A storage unit for storing a data set with round-trip time, and passing a value of a component of a vector having a dimension equal to the number of all links based on the data set and corresponding to each link on the round-trip route through the link; Create a vector with the value obtained by dividing the number of times by the maximum number of passes, and set the value of the component corresponding to other links to zero as input data, and the round trip time An input data / teacher signal generation unit for generating a corresponding teacher signal, and a vector of the same dimension as the input data, the value of a component corresponding to each link on an arbitrary path is set to the maximum number of times of passage of the link. And a test data generation unit that generates a vector with the value of the component corresponding to the other link being zero as test data, and the time required for each link and the round trip time based on the input data and the teacher signal. While learning the correlation,
A neural network that outputs an estimated passage time of an arbitrary route specified by the test data based on the learning result.

As described above, according to the network model creating apparatus of the present invention, at the fixed observation point, the round-trip path detecting section detects the round-trip path, and the round-trip time detecting section detects the round-trip time. Then, the data set of the detected round-trip path and round-trip time is stored in the storage unit.

Then, the input data / teacher signal creation unit
Based on the data set, input data and a teacher signal are created and input to the neural network. As a result of learning, the neural network obtains information on the correlation between the traffic of each link and the traffic of the entire network.

Further, the test data creating section creates test data specifying an arbitrary route and inputs the created test data to the neural network. As a result, the neural network outputs the estimated transit time of the route specified by the test data. Therefore, an expected transit time of an arbitrary route can be obtained, so that a network model as a traffic simulation can be created.

(Recording Medium) According to the recording medium of the present invention, in a packet switching network constituted by connecting a plurality of links, a round trip path of a packet signal between a fixed observation point and a plurality of points. A process of obtaining a data set of the data and the round trip time, a process of learning the traffic correlation between the traffic of the link and the traffic of the entire packet switching network based on the data set by a knowledge process, and using the result of the learning. And a program for causing a computer to execute a process of creating a network model by estimating traffic on an arbitrary route of a packet-switched network.

By reading and executing the program recorded on the recording medium of the present invention by a computer, a data set of a round trip route and a round trip time is collected at a fixed observation point. Then, learning by knowledge processing is performed based on the data set. As a result of learning by this knowledge processing, information on the correlation between the traffic of each link and the traffic of the entire network is obtained. Further, the traffic of an arbitrary route is estimated using the result of the knowledge processing. As a result, a network model as a traffic simulation can be created.

[0029]

Embodiments of the present invention will be described below with reference to the drawings. The present invention is not limited only to the illustrated example. In this embodiment, an example will be described in which a network model is created using the first host computer (first host) 10 in the network 100 as a fixed observation point shown in FIG.

The processing of the network model creating method according to the present embodiment is executed in the first host 10 computer under the control of a program. This program is provided, for example, by a recording medium. As the recording medium, for example, a magnetic disk, a semiconductor memory, or any other computer-readable medium can be used.

The network 100 includes a first router 20,
Second router 22, third router 24 and fourth router 26
It consists of. Then, the first router 20
A first host computer (first host) 10 is connected. Further, a second host computer (second host) 12 is connected to the second router 22. In addition, a third host computer (third host) is connected to the third router 24. A fourth host computer (fourth host) 16 is connected to the fourth router 26.

The first router 20 and the second router 22 are connected by a link A. The second router 22 and the third router 24 are connected by a link B. The third router 24 and the fourth router 26 are connected by a link C. The fourth router 26 and the first router 20 are connected by a link D.
In addition, the fourth router 26 and the second router 22
Connected by

The first host 10 is provided with a network model creation device as a fixed observation point. Here, FIG.
0 shows a block diagram. As shown in FIG. 2, the network model creation device 110 includes a round-trip route detection unit 111
A round-trip time measuring unit 112 as a round-trip time detecting unit;
It comprises a database 113 as a storage unit, an input data / teacher signal creating unit 114, a test vector creating unit 115 as a test data creating unit, and a neural network (neural network) 116.

Then, the round-trip route detecting unit 111 includes a fixed observation point and a plurality of points, for example, the first host 10 and the second host 10.
To the fourth hosts 12, 14, and 16, respectively, to detect the reciprocating path of each packet signal.

Here, the first host 10 and the third host 14
An example of detection of a round-trip route between and will be described. In this embodiment, first, a plurality of search packet signals (traffic data collection packets) are created. The search packet signal is set with the first host 10 as a sender as a destination and with the third host 14 as a waypoint.

Next, a plurality of search packet signals are set by sequentially increasing the lifetime of the search packet signals and transmitted from the first host 10 sequentially. Then, an error packet signal transmitted by a router or the third host 14 on the way when the lifetime of the search packet signal expires is received. Then, a route that passes through the router or the third host 14 in the order in which the error packet signals are transmitted is detected as a round-trip route. Further, other round-trip routes are similarly detected.

The network model creation device 110
The round-trip time measuring unit 112 detects a round-trip time required for a packet signal to be transferred on a round-trip path.

The database 113 stores a data set of the detected round-trip route and round-trip time as traffic data.

The input data / teacher signal creating section 114
Creates input data and a teacher signal based on a data set. This input data is a vector having a dimension equal to the number of all links, and a value obtained by dividing the value of the component corresponding to each link on the reciprocating route by the number of passages of the link by the maximum number of passages. . Therefore, the value of the component corresponding to the link other than the round trip route is zero. The teacher signal has a value corresponding to the round trip time of the round trip route corresponding to the input data.

The test vector creating section 115 creates a test vector as test data. This test vector is a vector of the same dimension as the input data, and is a value obtained by dividing the value of the component corresponding to each link on an arbitrary route by the number of times of passage of the link by the maximum number of times of passage,
This is a vector in which the value of the component corresponding to another link is zero. That is, an arbitrary route of the network can be designated by the test data. Note that information representing the correspondence between each link and each component of the vector is transferred from the database 113 to the test vector creation unit 115.

Further, the neural network 116
The correlation between the required time and the round-trip time of each link is learned based on the input data and the teacher signal. Then, the neural network 116, based on the learning result,
Outputs the estimated transit time of any route specified in the test data.

Next, a specific example of the network model creation method of the present invention, that is, the operation of the network model creation device will be described with reference to FIG. FIG. 3 is a flowchart for explaining the network creation method.

In creating a network model, first, a round trip path of a packet signal between a fixed observation point and a plurality of points is detected (S1 in FIG. 3). Here, first,
A first round trip path between the first host 10 and the second host 12 is detected. The first round-trip route departs from the first host 10 and moves to the first router 20, the second router 22, the second host 1,
2. It is assumed that the path passes through the second router 22 and the fourth router 26 sequentially and returns to the first host 10 again. The first reciprocating route is represented by a link “link A → link B → link D”.

Next, the time required for the packet signal to be transferred on the first round trip path is determined as the first round trip time (S2 in FIG. 3). Here, the first round trip time is "10 seconds"
Assume that

Next, it is determined whether or not necessary data sets of the round-trip route and the round-trip time have been obtained (S in FIG. 3).
3). The reference amount for the determination can be determined in advance.
In this embodiment, it is determined that the necessary data has been obtained based on the fact that the data set of the round-trip route and the round-trip time between the first host 10 and each of the other hosts has been obtained.

Then, the first host 10 and the third host 14
Is detected as “link D → link E → link B → link C → link E → link A”. Also, the second round trip time, which is the required time of the second round trip route, is set to “5
0 seconds "is detected.

The third round trip route between the first host 10 and the fourth host 16 is detected as "link D → link C → link B → link A". Further, the third round trip time, which is the required time of the third round trip route, is detected as "40 seconds". The data set of the round trip route and the round trip time detected in this way is stored in the database 113.

Next, a variable corresponding to the required time of the link on the round trip route is provided. That is, the variable a is given as the required time of the link A. Similarly, a variable b is given to the link B, a variable c is given to the link C, a variable d is given to the link D, and a variable e is given to the link E.

The sum of the time required for each link on the reciprocating route multiplied by the number of times of passage of the link is equal to the reciprocating time, so the simultaneous equations shown in the following equations (1) to (3) hold. .

That is, for the first reciprocating route, a + d + e = 10 (1) holds. For the second reciprocating route, the following holds: a + b + c + d + 2e = 50 (2) However, the coefficient “2” of the variable e is equal to the link E
Represents the number of passes. That is, in the second round trip route, the vehicle passes through the link E twice. For the third reciprocating route, a + b + c + d = 40 (3) holds.

The solution set of the simultaneous equations consisting of the above equations (1) to (3) holds the correlation between the required time required for the packet signal to pass through each link and the round trip time. . This correlation is, in other words, a traffic correlation between the traffic of the link and the traffic of the entire packet switching network. Then, this correlation is obtained as a learning result of the knowledge processing.

Therefore, in order to train the simultaneous equations using a neural network, input data is created and input to the neural network together with the teacher signal (S4 in FIG. 3). The input data is a vector of a dimension equal to the number of all links, and is a value obtained by dividing the value of the component corresponding to each link on the round trip route by each coefficient by the maximum coefficient value (normalized by the maximum coefficient value). Vector). Here, the variables a to e are respectively assigned to the first to first vectors.
Corresponds to the fifth component. Then, the round trip time is used as a teacher signal. The teacher signal is also normalized by dividing by the maximum coefficient value. Specifically, for the first round trip route, input data: (1, 0, 0, 1, 1) Teacher signal: 10 is input.

For the second round-trip path, the input data: (0.5, 0.5, 0.5, 0) is defined as a value obtained by dividing each coefficient value and the teacher signal by the maximum coefficient (maximum number of times of passage) 2. .5,1) Teacher signal: 25 (= 50/2) is input.

For the third round trip path, input data: (1, 1, 1, 1, 0) teacher signal: 40 is input.

Here, a neural network having an input layer having nodes equal in number of all links and one output layer is used as the neural network. In the neural network, each teacher signal is standardized and used, with the maximum value of the teacher signal as 1.

The neural network learns the neural network by comparing the output based on the input data with the teacher signal and correcting the variables (S5 in FIG. 3). As a result of this learning, a neural network corresponding to the network model is obtained. Then, the result of the learning holds the magnitude of the influence of each component of the input data on the output, that is, information on the correlation between each link and the round trip time. Therefore, the neural network obtained as a result of this learning corresponds to the network model.

Next, the traffic of an arbitrary route in the packet switching network is estimated using the result of the learning (S6 in FIG. 3). For this purpose, a test vector is created in the test vector creation unit 115 here.
This test vector is a unit vector. And
The test vector is a vector of a dimension equal to the number of all links, and is a value obtained by dividing the value of a component corresponding to a link on an arbitrary route by the number of times of passage of the link by the maximum number of times of passage, and is used for other links. This is a vector in which the value of the corresponding component is zero.

Then, the test vector is input to the learned neural network. Then, as an output of the neural network, an expected communication time corresponding to a required time estimated to be required for the packet signal to pass through an arbitrary path specified by the test vector is obtained.
As the estimated communication time, the estimated required time itself may be output, or data that can easily obtain the estimated required time, for example, an integer multiple of the estimated required time may be output. good.

In the above-described embodiment, an example has been described in which the present invention is configured under specific conditions. However, the present invention can be variously modified. For example, in the above embodiment, an example in which a neural network is used as knowledge processing has been described.
Knowledge processing is not limited to this. As the knowledge processing, for example, a genetic algorithm may be used.

[0060]

As described above in detail, according to the present invention, a data set of a round-trip route and a round-trip time is collected at a fixed observation point. Then, learning by knowledge processing is performed based on the data set. As a result of learning by this knowledge processing, information on the correlation between the traffic of each link and the traffic of the entire network is obtained. Further, the traffic of an arbitrary route is estimated using the result of the knowledge processing. As a result, a network model as a traffic simulation can be created.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of the Internet for explaining this embodiment.

FIG. 2 is a block diagram for explaining a network model creation device according to the embodiment.

FIG. 3 is a flowchart illustrating a network model creation method according to the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 1st host computer (1st host) 12 2nd host computer (2nd host) 14 3rd host computer (3rd host) 16 4th host computer (4th host) 20 1st router 22 2nd router 24th 3 router 26 fourth router 100 network 110 network model creation device 111 round-trip route detection unit 112 round-trip time measurement unit 113 database 114 input data / teacher signal creation unit 115 test vector creation unit 116 neural network

Claims (7)

[Claims] 1. In a packet-switched network configured by connecting a plurality of links, a data set of a round-trip route and a round-trip time of a packet signal between a fixed observation point and a plurality of points is obtained. And learns the traffic correlation between the traffic of the link and the traffic of the entire packet-switched network by knowledge processing, and estimates the traffic of an arbitrary route of the packet-switched network using the result of the learning. Thereby creating a network model. 2. The network model creating method according to claim 1, wherein the traffic correlation is a correlation between a required time required for the packet signal to pass through the link and the round-trip time. Characteristic network model creation method. 3. The network model creating method according to claim 3, wherein a variable corresponding to a required time of the link on the round trip route is provided, and the number of times the link passes is used as a coefficient, and the variable is multiplied by the coefficient. A network model creation method comprising: establishing a simultaneous equation with the round trip time as the sum of the equations based on the data set; and obtaining a set of solutions of the simultaneous equation by the learning. 4. The network model creating method according to claim 3, wherein the learning by the knowledge processing uses a neural network having an input layer having nodes equal to all the number of links and one output layer. A vector having a dimension equal to the number of all the links, and a value obtained by dividing a value of a component corresponding to each link on the round trip route by each of the coefficients by a maximum coefficient value, as input data; A method for creating a network model, wherein a signal is input to the neural network in accordance with the round trip time. 5. The network model creating method according to claim 4, wherein, when estimating the traffic of the arbitrary route, the neural network includes a vector having a dimension equal to the number of all the links, The test data obtained by dividing the value of the component corresponding to the link on the route by dividing the number of passages of the link by the maximum number of passages is input, and the output of the neural network corresponds to the estimated required time of the arbitrary route. A network model creation method characterized by obtaining an expected communication time to be performed. 6. A fixed observation point connected to a packet switching network formed by connecting a plurality of links and detecting a round trip path of a packet signal between the fixed observation point and a plurality of points. A round-trip path detecting unit; a round-trip time detecting unit that detects a round-trip time required for the packet signal to be transferred along the round-trip path; a storage unit that stores a data set of the round-trip path and the round-trip time; Based on the data set, a vector having a dimension equal to the number of all the links, and a value obtained by dividing the value of the component corresponding to each link on the reciprocating route by the number of passes of the link by the maximum number of passes, Input data and teacher signal creation for creating a vector with the value of the component corresponding to another link being zero as input data and creating a teacher signal corresponding to the round trip time And a vector having the same dimension as the input data and having a value obtained by dividing the value of the component corresponding to each link on an arbitrary route by the number of times of passage of the link by the maximum number of times of passage as test data A test data creating unit that learns the correlation between the required time of each link and the round trip time based on the input data and the teacher signal, and based on the learning result, A neural network that outputs an estimated transit time of an arbitrary route. 7. A process for obtaining a data set of a round trip path and a round trip time of a packet signal between a fixed observation point and a plurality of points in a packet switching network configured by connecting a plurality of links; Based on the data set, a process of learning a traffic correlation between the traffic of the link and the traffic of the entire packet-switched network by a knowledge process, and using a result of the learning, an arbitrary route of the packet-switched network. A recording medium recording a program for causing a computer to execute a process of creating a network model by estimating traffic.