JP2018170675A - Failure recovery procedure optimization system and failure recovery procedure optimization method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To grasp a change in the state of a network and to instantly provide a recovery procedure according to the situation by using machine learning.SOLUTION: A failure recovery procedure optimization includes a feature quantity calculation unit that obtains each network configuration information, digitizes and standardizes the acquired network configuration information, and calculates a feature quantity, a procedure learning and creating unit that acquires combination information of the calculated feature quantity and the fault alarm type generated in a network having and the feature quantity and creates or updates a restoration procedure corresponding to the acquired combination information, a procedure information storage unit that stores the created or updated restoration procedures in a certain order, and a filtering storage unit that stores the recovery procedure to be excluded from a use target among the stored recovery procedures.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technique for creating an operation procedure for recovering a network failure by machine learning or correcting the created operation procedure.

  Conventionally, when machine learning is performed, old information and new information are learned with the same weight. In the information used for learning, learning corresponding to time changes is performed by learning in consideration of the time point (Patent Document 1).

JP 2014-167744 A

  However, in Patent Document 1, while it is possible to cope with a change in the situation with respect to time, there is no processing such as correction that is triggered by a change in the state, and it is desired to immediately correct the learning content when the state changes. The case has not been addressed.

  The present invention has been made in view of such circumstances, and when correction of learned content is required, it is measured whether there is a change in the state of the network, and an optimal solution for the state of the network is measured. The purpose is to provide a failure recovery procedure optimization system that immediately reflects the correction contents in the recovery procedure by filtering as unnecessary information even when there is no change in the network status. To do.

  (1) In order to achieve the above object, the present invention takes the following measures. That is, the failure recovery procedure optimizing system of the present invention creates an operation procedure for recovering a network failure by machine learning, corrects the created operation procedure, and optimizes the created or corrected operation procedure. An optimization system that acquires each network configuration information, quantifies and standardizes the acquired network configuration information, calculates a feature amount, and includes the calculated feature amount and the feature amount A procedure learning / creating unit that acquires combination information of failure alarm types that have occurred in the network, creates or updates a recovery procedure corresponding to the acquired combination information, and stores the created or updated recovery procedure in a probable order An information storage unit and a filter for storing a recovery procedure to be excluded from the use target among the stored recovery procedures; A storage unit, and when it is determined that the restoration procedure needs to be corrected, the network configuration information is acquired again, and the feature amount is calculated. When the new feature value is different from any existing feature value, the calculated combination information of the new feature value and the failure alarm type and the restoration procedure for the combination information are stored in the filtering storage unit.

  In this way, each network configuration information is acquired, the acquired network configuration information is quantified and standardized, the feature amount is calculated, and the combination information of the calculated feature amount and the failure alarm type generated in the network having the feature amount is obtained. Acquire and create or update the recovery procedure corresponding to the acquired combination information, store the created or updated recovery procedure in the most probable order, store the recovery procedures to be excluded from the use among the stored recovery procedures, If it is determined that the restoration procedure needs to be corrected, the network configuration information is obtained again and the feature values are calculated. As a result, the calculated feature values in the network are different from any existing feature values. If it is an amount, filter the combination information of the calculated new feature amount and failure alarm type and the recovery procedure for the combination information Since stored in ring storage portion, it becomes possible to reflect the recovery procedure failure in accordance with the status of the network immediately.

  (2) In addition, the failure recovery procedure optimization system of the present invention obtains network configuration information again when it is determined that correction is necessary for the recovery procedure, and calculates the feature value. When the calculated feature value does not change, the restoration procedure is stored in the filtering storage unit.

  As described above, when it is determined that the restoration procedure needs to be corrected, the network configuration information is obtained again and the feature amount is calculated. As a result, if the calculated feature amount does not change in the network, Since the recovery procedure is stored in the filtering storage unit, when the same failure occurs next time, it is possible to exclude the recovery procedure that is determined to be corrected this time by filtering.

  (3) Also, in the failure recovery procedure optimization system of the present invention, when it is determined that correction is necessary for the recovery procedure, the network configuration information is acquired again and the feature amount is calculated. When the calculated feature value changes and is the same as any of the existing feature values, the restoration procedure is stored in the filtering storage unit.

  As described above, when it is determined that the restoration procedure needs to be corrected, the network configuration information is obtained again and the feature amount is calculated. As a result, the calculated feature amount changes in the network, and any existing If it is the same as the feature amount, the recovery procedure is stored in the filtering storage unit, so when the same failure occurs next time, it is excluded by filtering the recovery procedure that is determined to require correction this time It becomes possible to do.

  (4) Also, in the failure recovery procedure optimization system of the present invention, after the combination information of the calculated new feature quantity and failure alarm type and the recovery procedure for the combination information are stored in the filtering storage unit, the filtering storage unit It is characterized by deleting the previous recovery procedure stored in

  In this way, the combination information of the calculated new feature quantity and failure alarm type and the recovery procedure for the combination information are stored in the filtering storage unit, and then the previous recovery procedure stored in the filtering storage unit is deleted. The restored procedure can be executed again according to the subsequent change in the network status.

  (5) Also, the failure recovery procedure optimization method of the present invention creates an operation procedure for recovering a network failure by machine learning or modifies the created operation procedure, and optimizes the created or modified operation procedure. A method for optimizing a failure recovery procedure, wherein each of the network configuration information is acquired in a feature amount calculation unit, the obtained network configuration information is digitized and standardized, and a feature amount is calculated, and a procedure learning / creation unit A step of acquiring combination information of the calculated feature amount and a failure alarm type generated in the network having the feature amount, and creating or updating a recovery procedure corresponding to the acquired combination information; and a procedure information storage unit A step of storing the created or updated recovery procedures in a probable order and a filtering storage unit; Storing at least a recovery procedure to be excluded from use among the stored recovery procedures, and when it is determined that correction is necessary for the recovery procedure, obtain network configuration information again, If the calculated feature value is a new feature value different from any existing feature value in the network as a result of calculating the feature value, the combination information of the calculated new feature value and the fault alarm type and the combination A recovery procedure for information is stored in the filtering storage unit.

  In this way, the feature amount calculation unit acquires each network configuration information, the acquired network configuration information is quantified and standardized, the feature amount is calculated, and the procedure learning / creation unit calculates the calculated feature amount and the feature amount. The combination information of the failure alarm type that occurred in the network that has, the recovery procedure corresponding to the acquired combination information is created or updated, and in the procedure information storage unit, the created or updated recovery procedure is stored in order of probability, The filtering storage unit stores the recovery procedures that are excluded from the use from the stored recovery procedures, and if it is determined that the recovery procedures need to be corrected, the network configuration information is acquired again, and the feature quantity As a result, the calculated feature value in the network is a new feature value that is different from any existing feature value. In this case, the combination information of the calculated new feature quantity and failure alarm type and the recovery procedure for the combination information are stored in the filtering storage unit, so that the failure recovery procedure according to the network status can be reflected immediately. .

  According to the present invention, when the learned content needs to be corrected, it is measured whether or not the state of the network has changed, and an optimal solution for the state of the network is recreated. Even if there is no change, by filtering the recovery procedure that required correction as unnecessary information, the correction contents can be immediately reflected in the recovery procedure. As a result, depending on the details of the network failure It is possible to provide an appropriate recovery procedure.

It is a figure which shows schematic structure of the failure recovery procedure optimization system which concerns on this embodiment. It is a figure which shows an example of the information preserve | saved at the network information storage part. It is a figure which shows an example of the information stored in the feature-value storage part. It is a figure which shows an example of the information stored in the procedure information storage part. It is a figure which shows an example of the information stored in the filtering storage part. It is a flowchart which shows operation | movement of a failure recovery phase. It is a flowchart which shows operation | movement of a procedure learning and correction phase.

  In the recovery procedure in a network failure, the present inventors pay attention to the fact that the correction of the recovery procedure due to a change in the state of the network resource or the like is not immediately reflected, and when it is necessary to correct the learned content, Measure whether there has been a change in the network status, re-create the optimal solution for the network status, and even if there is no change in the network status, filter it as unnecessary information, so that the recovery procedure As a result, it was found that the correction contents can be reflected immediately, and as a result, an appropriate recovery procedure can be provided according to the contents of the failure of the network.

  That is, the failure recovery procedure optimizing system of the present invention creates an operation procedure for recovering a network failure by machine learning, corrects the created operation procedure, and optimizes the created or corrected operation procedure. An optimization system that acquires each network configuration information, quantifies and standardizes the acquired network configuration information, calculates a feature amount, and includes the calculated feature amount and the feature amount A procedure learning / creating unit that acquires combination information of failure alarm types that have occurred in the network, creates or updates a recovery procedure corresponding to the acquired combination information, and stores the created or updated recovery procedure in a probable order An information storage unit and a filter for storing a recovery procedure to be excluded from the use target among the stored recovery procedures; A storage unit, and when it is determined that the restoration procedure needs to be corrected, the network configuration information is acquired again, and the feature amount is calculated. When the new feature value is different from any existing feature value, the calculated combination information of the new feature value and the failure alarm type and the restoration procedure for the combination information are stored in the filtering storage unit.

  As a result, the present inventors have made it possible to immediately reflect the failure recovery procedure according to the network status. Embodiments of the present invention will be specifically described below with reference to the drawings.

  FIG. 1 is a diagram showing a schematic configuration of a failure recovery procedure optimization system according to the present embodiment. The failure recovery procedure optimization system 1 includes a network information storage unit 11, a monitoring unit 13, a feature amount calculation unit 15, a feature amount storage unit 17, an input value creation unit 21, a procedure information storage unit 23, a result output unit 25, and a filtering storage. Unit 27, procedure learning / creating unit 29, and procedure result confirming unit 31. The usage form of the failure recovery procedure optimization system 1 according to the present embodiment is composed of four phases: prior learning, failure recovery, procedure learning / correction, and filter correction. In the following, each function will be described while explaining these four modes of use step by step.

[1. Prior learning]
First, the user inputs a network name and IP address information related to the network. The input network name and IP address information related to the network are stored in the network information storage unit 11. FIG. 2 is a diagram illustrating an example of information stored in the network information storage unit 11.

  Next, the monitoring unit 13 refers to the IP address information stored in the network information storage unit 11, acquires resource information (CPU usage rate, memory usage rate, traffic information, etc.) of each network device, and calculates a feature amount. To the unit 15.

  Next, the feature amount calculation unit 15 calculates a network state as a feature amount. Here, for example, it is assumed that the feature quantity of each network is calculated using machine learning or the like, and machine learning such as SOM (Self Organizing Map) or Auto encoder may be used. After calculating the feature amount of each network, the feature amount storage unit 17 stores information regarding each network and the feature amount of each network. FIG. 3 is a diagram illustrating an example of information stored in the feature amount storage unit 17.

  Next, the input value creation unit 21 combines the feature amount of each network and the alarm type of the failure that has occurred in each network, and acquires it as an input value. Then, the input value creation unit 21 transmits the restoration procedure used in the failure as an output value, that is, a solution for the feature value and the alarm type as the input value, to the procedure learning / creation unit 29. The procedure learning / creation unit 29 performs learning (supervised learning) based on the acquired feature value and alarm type that are input values and the recovery procedure that is an output value, and stores the information as a procedure information storage unit 23. Store in FIG. 4 is a diagram illustrating an example of information stored in the procedure information storage unit 23. The procedure information storage unit 23 stores the recovery procedures in a probable order for each combination of network feature amount and alarm type.

[2. Disaster recovery]
FIG. 6 is a flowchart showing the operation of the failure recovery phase. When a failure occurs in the network, a failure alarm is issued from the network in which the failure has occurred to the input value creation unit 21, and the failure is detected (step S101). The input value creation unit 21 refers to the information in the procedure information storage unit 23 and confirms the network in which the failure has occurred and the network feature amount (steps S102 and S103). As a result of the confirmation, it is determined whether or not the fault has occurred so far (step S104).

  In step S104, if the failure that has occurred is a failure that has already occurred and recovery procedure information exists, the input value creation unit 21 notifies the result output unit 25 to that effect. The result output unit 25 refers to the information in the filtering storage unit 27 and the procedure information storage unit 23 (step S107), and the information stored in the filtering storage unit 27 among the information stored in the procedure information storage unit 23. The corresponding restoration procedure other than is output to the user (step S108).

  On the other hand, if the input value creation unit 21 determines in step S104 that the failure that has occurred is a failure that has occurred for the first time, the combination of the feature amount of the network and the alarm is transmitted as an input value to the procedure learning / creation unit 29. (Step S105).

  Next, the procedure learning / creating unit 29 calculates a restoration procedure from the input value acquired from the input value creating unit 21, and uses the calculated restoration procedure as an output value for the input value, such as a feature amount, a network name, an alarm type, and the like. The information is stored together with the information in the procedure information storage unit 23 (step S106). Then, the calculated restoration procedure is output to the user via the result output unit 25 (step S108).

[3. Procedure learning / correction]
FIG. 7 is a flowchart showing the operation of the procedure learning / correction phase. First, after executing the recovery procedure output from the result output unit 25, the user checks whether or not the recovery procedure needs to be corrected (step S201), and whether or not the recovery procedure needs to be corrected. Is determined (step S202). The user inputs the determination result to the procedure result confirmation unit 31.

  If there is no need to modify the restoration procedure, the procedure learning / creating unit 29 receives the network feature value and alarm information as input values, and the executed procedure information as an output value. Procedure information is transmitted and learning is performed (step S209). The learning result is stored (updated) in the procedure information storage unit 23, and the process ends (step S210).

  If it is determined in step S202 that the procedure needs to be corrected, the procedure result confirmation unit 31 requests the monitoring unit 13 to perform re-monitoring (step S203). This is aimed at exploring the possibility that the status of the network will change and the previous procedures will not be available.

  The monitoring unit 13 refers to the network information storage unit 11, acquires the resource information of the network device existing on the network, and transmits the acquired resource information to the feature amount calculation unit 15.

  Next, the feature amount calculation unit 15 calculates a feature amount based on the acquired resource information (step S204). As a result of calculating the feature amount, if the feature amount has not changed (step S205), the recovery procedure executed this time is stored (updated) in the filtering storage unit 27 as filtering in the feature amount of the network in which the failure has occurred, It is not used next time (step S208).

  In addition, if the feature quantity has changed as a result of calculating the feature quantity in step S205, but has already existed (step S206), the restoration procedure used this time is used as filtering for the feature quantity. A new input (update) is made to the filtering storage unit 27 so that it will not be used next time (step S208).

  In step S206, if the feature value changes and is the first feature value, the information is stored in the feature value storage unit 17, and the filtering is performed on the feature value together with the restoration procedure used this time. Then, it is newly input (updated) to the filtering storage unit 27 so that it is not used next time (step S207).

  After the processing in steps S205 to S208 is completed, the procedure learning / creating unit receives the feature value and the alarm information as input values, and uses a restoration procedure that does not require procedure correction or a newly created restoration procedure. The output value is learned (step S209), and the information is stored (updated) in the procedure information storage unit 23 (step S210).

[4. Filter correction]
In the procedure learning / correction phase, when learning progresses and it is determined that the procedure stored in the filtering storage unit is not the procedure information determined to be the most unnecessary, the procedure information is deleted from the filtering storage unit. For example, in FIG. 5, two procedure information items such as a feature value “1” and an alarm type “error” are stored in the filtering storage unit as filtering (exclusion) targets. > 6] is no longer the first recovery procedure to be filtered (excluded), the recovery procedure [4->5-> 6] is deleted from the filtering storage unit. When a recovery procedure is stored in the filtering storage unit, the recovery procedure is no longer used. However, depending on the situation, a recovery procedure that is not first in the exclusion target may be effective. Therefore, when the recovery procedure is not ranked first among the recovery procedures to be filtered (excluded), the recovery procedure is deleted.

  As described above, according to the present embodiment, when the learned content needs to be corrected, it is measured whether the network state has changed, and an optimal solution for the network state is recreated. In addition, even when there is no change in the network status, by filtering the recovery procedure that required correction as unnecessary information, the correction contents can be immediately reflected in the recovery procedure. An appropriate recovery procedure can be provided according to the content of the network failure.

DESCRIPTION OF SYMBOLS 1 Failure recovery procedure optimization system 11 Network information storage part 13 Monitoring part 15 Feature-value calculation part 17 Feature-value storage part 21 Input value preparation part 23 Procedure information storage part 25 Result output part 27 Filtering storage part 29 Procedure learning and preparation part 31 Procedure result confirmation section

Claims (5)

A failure recovery procedure optimization system that creates an operation procedure that recovers a network failure by machine learning, modifies the created operation procedure, and optimizes the created or modified operation procedure,
A feature amount calculation unit that acquires each network configuration information, digitizes and standardizes the acquired network configuration information, and calculates a feature amount;
A procedure learning / creating unit that acquires combination information of the calculated feature value and a failure alarm type that has occurred in the network having the feature value, and creates or updates a recovery procedure corresponding to the acquired combination information;
A procedure information storage unit for storing the created or updated recovery procedures in an ascending order;
Among the stored recovery procedures, a filtering storage unit that stores recovery procedures excluded from use, and
When it is determined that the restoration procedure needs to be corrected, the network configuration information is acquired again, and the feature amount is calculated. As a result, the calculated feature amount is the same as any existing feature amount in the network. A failure recovery procedure optimizing system that stores the combination information of the calculated new feature amount and failure alarm type and the recovery procedure for the combination information in the filtering storage unit when they are different new feature amounts.   When it is determined that correction is necessary for the restoration procedure, the network configuration information is obtained again, and the feature amount is calculated. As a result, in the network, when the calculated feature amount has not changed, The failure recovery procedure optimization system according to claim 1, wherein the recovery procedure is stored in the filtering storage unit.   When it is determined that the restoration procedure needs to be corrected, the network configuration information is acquired again, and the feature amount is calculated. As a result, the calculated feature amount changes in the network, and any existing The failure recovery procedure optimization system according to claim 1 or 2, wherein the recovery procedure is stored in the filtering storage unit when the feature amount is the same.   The combination information of the calculated new feature quantity and failure alarm type and the recovery procedure for the combination information are stored in the filtering storage unit, and then the previous recovery procedure stored in the filtering storage unit is deleted. The failure recovery procedure optimization system according to claim 1. A failure recovery procedure optimization method that creates an operation procedure that recovers a network failure by machine learning or modifies the created operation procedure and optimizes the created or modified operation procedure,
In the feature amount calculating unit, acquiring each network configuration information, quantifying and standardizing the acquired network configuration information, and calculating a feature amount;
In the procedure learning / creating unit, acquiring the combination information of the calculated feature amount and the failure alarm type generated in the network having the feature amount, and creating or updating a recovery procedure corresponding to the acquired combination information;
In the procedure information storage unit, storing the created or updated recovery procedure in the most likely order;
The filtering storage unit includes at least a step of storing a recovery procedure to be excluded from the use target among the stored recovery procedures,
When it is determined that the restoration procedure needs to be corrected, the network configuration information is acquired again, and the feature amount is calculated. As a result, the calculated feature amount is the same as any existing feature amount in the network. A failure recovery procedure optimizing method comprising: storing the combination information of the calculated new feature amount and failure alarm type and the recovery procedure for the combination information in the filtering storage unit when the feature values are different new feature amounts.

Images