KR102256447B1 - Network management apparatus and method for latency critical services on software defined network environment - Google Patents

Abstract

According to the present invention, provided are a network management device for a delay-sensitive service in a software-defined network environment and a method thereof. The network management device of the present invention comprises: a link latency obtaining unit for monitoring and obtaining the latencies of a target link to be predicted on a network and at least one peripheral link separately connected to two target nodes connected by the target link for a predetermined time, and arranging the obtained latencies in a predetermined patterned matrix shape to obtain a latency feature; a link influence calculating unit for obtaining port information by monitoring the target nodes, obtaining a correlation matrix indicating a correlation between the target link and the peripheral link from the obtained port information, and converting the correlation matrix obtained for a predetermined time into a predetermined shape to arrange the same to obtain an influence feature; a feature combining unit for obtaining a link feature for the target link by combining the latency feature and the influence feature; and a latency predicting unit for estimating the latency at a next time point of the target link by receiving the link feature by learning a pattern estimation method in advance. Therefore, the latency of the target link can be accurately predicted.

Description

A network management device and method for delay-sensitive services in a software-defined network environment {NETWORK MANAGEMENT APPARATUS AND METHOD FOR LATENCY CRITICAL SERVICES ON SOFTWARE DEFINED NETWORK ENVIRONMENT}

The present invention relates to a network management apparatus and method, and to a network management apparatus and method capable of predicting latency for a delay sensitive service in a software defined network environment.

As network traffic explosively increases due to the development of information and communication technology, the data processing capacity required for the network is also rapidly increasing, and the number of network equipment to be managed is also increasing significantly. However, in the past, there is a problem that network management is difficult as each of the numerous network devices on the network must be individually managed, and it is not easy to expand, update, change settings, etc. of the network.

In order to overcome these limitations, the concept of network virtualization was introduced in which a physical network infrastructure is commonly used, but an independent virtual network environment is created and provided according to the purpose of use. In particular, recently, research on software defined networking (SDN) that enables network control through software programming has been actively conducted. SDN has the advantage that the network control function is separated from the physical network such as a switch or router, and the network itself can be programmed in software, so that the network can be flexibly configured and change is easy. In other words, in SDN, a network administrator does not individually manage a plurality of network switches, and can easily change the control settings of a plurality of network switches through software programming.

When using SDN in this way, it is possible to dynamically control various devices on the network so that network reconfiguration is possible, but in the past, a network path that satisfies the Service Level Agreement (SLA) is set or bandwidth is allocated. Only the level of control was performed, and the network manager performed only passive monitoring of the network status.

However, there is an increasing demand for delay critical services, where speed is very important, such as Augmented Reality (AR), Virtual Reality (VR), emergency disaster communication, and V2I (Vehicle to Infrastructure). Therefore, it is desirable for the network manager to transmit data by predicting data transmission latency in advance in order to stably provide a delayed service.

Korean Registered Patent No. 10-1748750 (registered on June 13, 2017)

An object of the present invention is to provide a network management apparatus and method capable of accurately predicting in advance the latency of links between network nodes in a software defined network environment using an artificial neural network.

Another object of the present invention is to provide a network management apparatus and method capable of stably providing a delay-sensitive service based on a predicted latency.

The network management apparatus according to an embodiment of the present invention for achieving the above object determines the latency of a target link to predict latency on a network and at least one peripheral link connected to each of two target nodes connected by the target link. A link latency acquisition unit that monitors and acquires the acquired latency for a specified period of time, and obtains a latency characteristic by arranging the acquired latency in a matrix form of a predetermined pattern; The target node is monitored to obtain port information, a correlation matrix indicating a correlation between the target link and at least one neighboring link is obtained from the obtained port information, and a correlation matrix obtained for a predetermined period is previously A link influence degree calculation unit that converts and arranges in a designated form to obtain an influence degree characteristic; A feature combining unit that combines the latency feature and the influence feature to obtain a link feature for the target link; And a latency predictor for estimating a latency at a next point in time of the target link by receiving the link characteristic by learning a pattern estimation method in advance.

The latency acquisition unit includes an active monitoring unit that monitors a plurality of links on the network to acquire latency of the target link and the at least one peripheral link; And accumulating and storing the obtained latency of the target link and the at least one neighboring link, classifying each of the accumulated and stored latency of the target link and the latency of the at least one neighboring link and arranging them in chronological order to obtain the latency characteristics. It may include a latency feature acquisition unit.

The link influence calculation unit monitors at least one port of each of a plurality of nodes of the network using an open flow protocol of a software defined network to obtain transmission byte, reception byte, transmission packet, reception packet, and duration port information. Passive monitoring unit; And calculating the average packet rate change amount and the average transmission amount change amount in a predetermined manner using the port information, and at least one target link for each flow direction and each time point in which data is transmitted from the calculated average packet rate change amount and the average transmission amount change amount. A predetermined number of correlation values representing the degree of influence between each of the neighboring links of are calculated, and a correlation matrix having a correlation obtained by accumulating each of the calculated correlation values for each of the neighboring links as an element is obtained. The influence degree feature acquisition unit may be arranged to obtain the influence degree feature by arranging it in a designated shape.

)과 상기 평균 전송량 변화량(

)을 수학식The influence degree feature acquisition unit is the average packet rate change amount according to the flow direction (f) (

) And the amount of change in the average transmission amount (

) To the equation

) 또는 평균 전송량 변화량(

)을 나타내며, Y[]는 주변 링크(e_η)를 통해 전송되는 평균 패킷율 변화량(

) 또는 평균 전송량 변화량(

)을 나타낸다.)에 대입하여 주변 링크(e_η)로부터 타겟 링크(e_τ)로의 플로우에서 평균 패킷율 변화량의 상관값(

)과 평균 전송량 변화량(

)의 상관값(

), 타겟 링크(e_τ)에서 주변 링크(e_η)로의 플로우에서 평균 패킷율 변화량의 상관값(

) 및 평균 전송량 변화량(

)의 상관값(

)을 계산할 수 있다.(Where n represents the number of views from the current point in time (t) to the previously specified point in time (tk), and X[] is the average packet rate change amount transmitted through the _{target link (e τ ), respectively (}

) Or the amount of change in the average transmission amount (

), and Y[] is the amount of change in the average packet rate transmitted through the _{peripheral link (e η) (}

) Or the amount of change in the average transmission amount (

), and the correlation value of the average packet rate change amount in the flow from the peripheral link (e _η ) to the target link (e _{τ) (}

) And average transmission amount change (

) Of the correlation value (

), the correlation value of the average packet rate change in the flow from the target link (e _τ ) to the neighboring link (e _{η) (}

) And average transmission amount change (

) Of the correlation value (

) Can be calculated.

,

,

,

) 각각을 전체 주변 링크에 대해 각각 개별적으로 누적합하여 4개의 상관도를 원소로 포함하는 상기 상관도 행렬을 획득할 수 있다.The influence degree feature acquisition unit has four correlation values calculated for each of at least one neighboring link (

,

,

,

) It is possible to obtain the correlation matrix including four correlations as elements by individually accumulating each of each of the surrounding links.

The influence degree feature acquisition unit may obtain a transposed correlation matrix by transposing each of a plurality of correlation matrices obtained during a previously predetermined period, and obtain the influence degree feature by arranging the transposed correlation matrix in a row direction.

The influence degree feature acquisition unit may obtain a transposed correlation matrix by transposing each of a plurality of correlation matrices obtained during a previously determined period, and obtain the influence degree feature by arranging the transposed correlation matrix in a row direction. .

)과 후방향 레이턴시(

)를 이용하여 다음 시점에서의 레이턴시(

)를 계산하는 레이턴시 계산부를 포함할 수 있다.The latency prediction unit is implemented as a bi-directional LSTM including a plurality of bi-directional LSTM cells, and from the forward latency at the previous point in time and the reverse latency at the next time point applied from a latency feature at a corresponding time point in the link feature and an adjacent bidirectional LSTM cell. A latency estimating unit for estimating a forward latency and a reverse latency of a corresponding viewpoint according to a previously learned pattern estimation method; _{The relationship values for each point in time (α tk} , α _tk-1 , …, α _t-1 , which indicate the relationship between the latency of each point in the link feature and the forward and backward latency estimated at the point immediately before the delay of each point in time) a pattern analysis unit that acquires α _{t );} A relationship value that estimates the relationship value _{(α t+1} ) at the next time point from the relationship values (α _tk , α _tk-1 , …, α _t-1 , α _t ) for each time point according to a previously learned pattern estimation method Estimation unit; And the estimated relationship value (α _t+1 ) and the estimated forward latency (

) And backward latency (

Using ), the latency at the next point in time (

It may include a latency calculator that calculates ).

)와 이전 레이턴시 추정부(141)에서 이전 t-1 시점에서 추정된 전방향 레이턴시(

)와 후방향 레이턴시(

) 사이에 수학식 The pattern analysis unit has a latency (

) And the forward latency estimated at the previous time t-1 by the previous latency estimating unit 141 (

) And backward latency (

) Between equations

It can be obtained by calculating the relationship value α _{t that satisfies.}

)와 후방향 레이턴시(

)로부터 수학식 The latency calculation unit is the estimated relationship value (α _t+1 ) and the estimated omni-directional latency for the current time point (t) (

) And backward latency (

) From the equation

)를 계산할 수 있다.According to the latency at the next point in time (

) Can be calculated.

The latency estimating unit and the relationship value estimating unit may be learned by using the latency for each time point included in the link feature as learning data.

A network management method according to another embodiment of the present invention for achieving the above other object is to determine the latency of a target link to predict latency on a network and at least one peripheral link connected to each of two target nodes connected by the target link. Obtaining a latency characteristic by monitoring and obtaining the obtained latency for a predetermined period of time, and arranging the obtained latency in a matrix form of a predetermined pattern; The target node is monitored to obtain port information, a correlation matrix indicating a correlation between the target link and at least one neighboring link is obtained from the obtained port information, and a correlation matrix obtained for a predetermined period is previously Converting and arranging in a designated form to obtain an influence characteristic; Obtaining a link characteristic for the target link by combining the latency characteristic and the influence characteristic characteristic; And estimating a latency at a next point in time of the target link by receiving the link feature by learning a pattern estimation method in advance.

Therefore, the network management apparatus and method according to an embodiment of the present invention can determine the latency of the target link between network nodes in the SDN environment, the previously measured latency of the target link and the neighboring link, and the target link obtained through the open flow protocol of SDN. It can be obtained through an artificial neural network based on the degree of influence between neighboring links. Therefore, it is possible to stably provide a delay-sensitive service by using the latency of the target link from which the SDN is obtained.

1 shows a schematic structure of a network management apparatus according to an embodiment of the present invention.
2 shows an example of a target link, a peripheral node, and a peripheral link in a network.
3 is a diagram for explaining port information acquired using an open flow protocol.
4 shows an example of a detailed structure of the latency prediction unit of FIG. 1.
5 shows a network management method according to an embodiment of the present invention.
6 shows in detail an example of the latency prediction step of the present invention of FIG. 5.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the implementation of the present invention, reference should be made to the accompanying drawings illustrating preferred embodiments of the present invention and the contents described in the accompanying drawings.

Hereinafter, the present invention will be described in detail by describing a preferred embodiment of the present invention with reference to the accompanying drawings. However, the present invention may be implemented in various different forms, and is not limited to the described embodiments. In addition, in order to clearly describe the present invention, parts irrelevant to the description are omitted, and the same reference numerals in the drawings indicate the same members.

Throughout the specification, when a certain part "includes" a certain component, it means that other components may be further included, rather than excluding other components unless specifically stated to the contrary. In addition, terms such as "... unit", "... group", "module", and "block" described in the specification mean a unit that processes at least one function or operation, which is hardware, software, or hardware. And software.

1 shows a schematic structure of a network management apparatus according to an embodiment of the present invention, FIG. 2 shows an example of a target link, a neighboring node, and a neighboring link in a network, and FIG. 3 is a diagram obtained using an openflow protocol. It is a diagram for explaining port information.

Referring to FIG. 1, the network management apparatus according to the present embodiment may include a link latency acquisition unit 110, a link influence degree calculation unit 120, a feature combination unit 130, and a latency prediction unit 140. .

)을 획득한다.The link latency acquisition unit 110 includes a target link for predicting latency among a plurality of links connected by wire or wireless to transmit or receive data between a plurality of nodes on a network and two targets connected by the target link. Active monitoring of the latency of at least one peripheral link connected to each node, and using the monitored latency for a predetermined period of time, the latency characteristics (

).

The link latency acquisition unit 110 may include an active monitoring unit 111 and a latency feature acquisition unit 112. The active monitoring unit 111 monitors and acquires the latency _{of the target link e τ} and the peripheral link e _η = {e ₁ , e ₂ ,…, e _{δ }.}

2, the active monitoring unit 111 latency are two target nodes that send and receive data with each other via the target link (e _τ) and the target link (e _τ) to be predicted (v _7, v _8), and two target nodes (v _7, v ₈₎ and the other nodes (v ₁ ~ v ₆₎ to at least one peripheral connecting link (e ₁ ~ e ₆₎ by dynamic monitoring of the at least the target link (e _τ) One peripheral link (e ₁ ~ e ₆ ) each acquires the latency.

In FIG. 2, for convenience of explanation, only the target link (e _τ ) and six peripheral links (e ₁ to e ₆ ) based on the target link (e _τ ) are shown, but the active monitoring unit 111 You can monitor the latency for the link.

)을 획득한다. _{In addition, the latency feature acquisition unit 112 determines the latency of the target link (e τ} ) and the neighboring links (e ₁ to e ₆ ) for a previously predetermined period acquired by monitoring each link of the network by the active monitoring unit 111. By accumulating and arranging the accumulated latency in a matrix form of a predetermined pattern.

).

~

)를 누적하여 획득하고, 누적 획득된 타겟 링크(e_τ)의 레이턴시(

~

)를 기지정된 패턴으로 배열한다. 레이턴시 특징 획득부(112)는 일 예로 획득된 타겟 링크(e_τ)의 레이턴시(

~

)를 시간의 순서에 따라 1행에 순차적으로 배열할 할 수 있다.The latency feature acquisition unit 112 firstly has a latency from _{the previously predetermined k time point tk of the target link e τ} to the current time point (or step) t

~

) Is accumulated and acquired, and the latency of the _{accumulated acquired target link (e τ) (}

~

) Are arranged in a predetermined pattern. The latency feature acquisition unit 112 includes, for example, the latency of the _{acquired target link e τ}

~

) Can be arranged sequentially in one row according to the order of time.

~

), …, (

~

))를 누적하여 획득하고, 행 방향으로 배열된 타겟 링크(e_τ)의 레이턴시(

~

)와 마찬가지로, 획득된 δ개의 주변 링크(e₁ ~ e_δ)의 레이턴시((

~

), …, (

~

))를 각각 행단위로 시간의 순서에 따라 순차적으로 배열할 수 있다.And the latency feature acquisition unit 112 is the latency from the previously predetermined k time point (tk) to the current time point (t) for each of the _δ peripheral links (e ₁ ~ e _{δ) centered on the target link (e τ ).} ((

~

),… , (

~

)), and the latency of the target links (e _{τ) arranged in the row direction (}

~

), the latency of the acquired δ peripheral links (e ₁ to e _{δ) ((}

~

),… , (

~

)) can be arranged sequentially in the order of time in each row unit.

~

), (

~

), …, (

~

))를 매트릭스 형태로 배열함으로써 수학식 1에 나타난 바와 같이 레이턴시 특징(

)을 획득할 수 있다.That is, the latency feature acquisition unit 112 _{divides each of the target link (e τ} ) and the surrounding links (e ₁ to e _δ ) into rows, and each time point from the previously predetermined k time point (tk) to the current time point (t) Is divided into columns, and the accumulated latency of each of the _{target link (e τ} ) and the surrounding link (e ₁ ~ e _{δ) is calculated as ((}

~

), (

~

),… , (

~

)) in a matrix form, as shown in Equation 1, the latency characteristic (

) Can be obtained.

)은 타겟 링크(e_τ)의 주변 링크(e₁ ~ e_δ)의 개수가 δ개이고, 레이턴시가 이전 k 시점(t-k)으로부터 현재 시점(t)까지 누적되는 경우, (δ+1) X (k+1) 크기의 매트릭스로 획득될 수 있다.As shown in Equation 1, the latency characteristic (

) Is (δ+1) X (if the number of neighboring links (e ₁ to e _δ ) of the target link (e _τ ) is δ, and the latency is accumulated from the previous time point k (tk) to the current time point (t), (δ+1) X ( It can be obtained as a matrix of size k+1).

~

)가 제1 행에 배열되는 것으로 가정하였으나, 타겟 링크(e_τ)의 레이턴시(

~

)의 배열 위치 또한 가변될 수 있다. 일예로 타겟 링크(e_τ)의 레이턴시(

~

)는 레이턴시 특징(

)에서 가장 마지막 행에 배치될 수도 있다.In Equation 1, as an example, each of the target link (e _τ ) and δ neighboring links (e ₁ to e _δ ) is divided into rows, and each time point from the previously determined k time point (tk) to the current time point (t) is Although divided into columns, the present embodiment is not limited thereto. That is, each of the target link (e _τ ) and δ neighboring links (e ₁ ~ e _δ ) is divided into columns, and each time point from the previously specified time point k (tk) to the current time point (t) is divided into rows and arranged. May be. Also, the latency of the target link (e _{τ) (}

~

) Is assumed to be arranged in the first row, but the latency of the _{target link (e τ) (}

~

The arrangement position of) can also be changed. For example, the latency of the target link (e _{τ) (}

~

) Is the latency characteristic (

) May be placed on the last line.

~

), (

~

), …, (

~

))는 타겟 링크(e_τ)에서 데이터가 전송되는 방향, 즉 플로우 방향에 따라 획득된 단방향 레이턴시일 수 있다. 2개의 타겟 노드(v₇, v₈) 사이에 연결되는 타겟 링크(e_τ)에서는 네트워크 상태에 따라 제1 타겟 노드(v₇)로부터 제2 타겟 노드(v₈)로 데이터가 전달되는 제1 플로우 방향의 레이턴시와 제2 타겟 노드(v₈)로부터 제1 타겟 노드(v₇)로 데이터가 전달되는 제2 플로우 방향의 레이턴시가 동일하지 않은 경우가 빈번하게 발생한다. 즉 하나의 타겟 링크(e_τ)에서도 플로우 방향에 따라 레이턴시가 서로 상이할 수 있다. 그리고 네트워크 관리 장치는 2개의 타겟 노드(v₇, v₈) 중 타겟 링크(e_τ)를 통해 데이터를 전송할 타겟 노드를 미리 확인할 수 있다. 그러므로 타겟 링크(e_τ)에서 2개의 플로우 방향 각각에 대한 레이턴시를 모두 예측하는 것은 불필요하다.Here, the latency of the target link (e _τ ) and at least one peripheral link (e ₁ ~ e _δ ) accumulated by the latency feature acquisition unit 112 ((

~

), (

~

),… , (

~

)) may be a one-way latency obtained according to a direction in which data is transmitted from the target link e _{τ, that is, a flow direction.} Two target nodes (v _7, v ₈₎ in the target link (e _τ) which is connected between the first targets, depending on the network the node (v ₇₎ a second target node (v ₈₎ to the first data is being transferred from the It frequently occurs that the latency in the flow direction and the latency in the second flow direction in which data is transmitted from the second target node v ₈ to the first target node v _{7 are not the same.} That is, even in one target link e _τ , latency may be different from each other according to the flow direction. In addition, the network management apparatus may check in advance a target node to transmit data through a target link e _{τ among the two target nodes v 7} and v _8. Therefore, it is unnecessary to predict all of the latency for each of the two flow directions in the target link e _τ.

)을 획득할 수 있다.Accordingly, in this embodiment, the latency feature acquisition unit 112 includes data of the target link (e _τ ) and δ peripheral links (e ₁ to e _δ ) among the latencies of each of the plurality of links acquired by the active monitoring unit 111. By accumulating and arranging one-way latency according to the transmitted flow direction, the latency characteristics (

) Can be obtained.

)을 획득한다.On the other hand, the link impact calculation unit 120 _{passively monitors the target node (v 7} , v ₈ ) of the network to obtain port information, and based on the passively monitored port information, the target link (e _τ ) and the surrounding link ( Calculate the degree of influence between e ₁ and e _δ in a pre-specified manner, and use the calculated degree of influence to determine the degree of influence (

).

The link influence degree calculation unit 120 may include a passive monitoring unit 121 and an influence degree feature acquisition unit 122.

The passive monitoring unit 121 obtains port information by monitoring each of at least one port of each of a plurality of nodes of a network using the open flow protocol of SDN. Here, the port information may include transmitted bytes, received bytes, transmitted packets, received packets, duration time, and the like.

The influence feature acquisition unit 122 calculates the change amount Δpr of the average packet rate and the change amount ΔΘ of the average throughput using the port information obtained from the passive monitoring unit 121 Then, the influence between links is calculated from the calculated average packet rate change amount (Δpr) and the average transmission amount change amount (ΔΘ).

)과 평균 전송량 변화량(

)은 포트 정보에 포함된 송신 바이트, 수신 바이트, 송신 패킷, 수신 패킷 및 지속 시간으로부터 계산될 수 있으며, 평균 패킷율 변화량(

)과 평균 전송량 변화량(

)을 계산하는 기법은 공지된 기술이므로 여기서는 상세하게 설명하지 않는다.Average packet rate change (

) And the amount of change in the average transmission amount (

) Can be calculated from the transmitted byte, received byte, transmitted packet, received packet, and duration included in the port information, and the average packet rate change amount (

) And the amount of change in the average transmission amount (

The technique of calculating) is a known technique, so it is not described in detail here.

)과 평균 전송량 변화량(

)를 계산할 수 있다. 도 3에서

는 링크(e_η)를 통해 노드(v)로 전달되는 데이터의 평균 전송량 변화량(

)을 나타내고,

는 노드(v)로부터 링크(e_τ)를 통해 전달되는 데이터의 평균 전송량 변화량(

)을 나타낸다.As shown in FIG. 3, the influence feature acquisition unit 122 uses the port information based on at least one node v connected to at _{least one link e η of the network, and the average packet rate change amount (}

) And average transmission amount change (

) Can be calculated. In Figure 3

Is the average amount of change in the amount of data transmitted to the node (v) through the link (e _{η) (}

),

Is the average amount of change in the amount of data transmitted through the _{link (e τ} ) from the node (v) (

).

)과 평균 전송량 변화량(

)이 정규 분포(normal distribution)를 따른다는 가정하에서 특정 시점(t)에서 플로우 방향에 따른 영향도를 나타내는 4개의 상관값들(

,

,

,

)을 각각 계산할 수 있다.Influence feature acquisition unit 122 is the average packet rate change in a specific node (v) (

) And the amount of change in the average transmission amount (

Under the assumption that) follows a normal distribution, four correlation values representing the degree of influence according to the flow direction at a specific time point (t) (

,

,

,

) Can be calculated separately.

는 노드(v)에서 데이터가 전달되는 플로우가 링크(e_η)로부터 링크(e_τ) 방향인 경우의 평균 패킷율 변화량의 상관값을 나타내고,

는 링크(e_τ)로부터 링크(e_η)로 데이터가 전달되는 플로우에서 평균 패킷율 변화량의 상관값을 나타낸다. 그리고

는 링크(e_η)로부터 링크(e_τ)로 데이터가 전달되는 플로우에서 평균 전송량 변화량의 상관값을 나타내고,

는 링크(e_τ)로부터 링크(e_η)로 데이터가 전달되는 플로우에서의 평균 전송량 변화량의 상관값을 나타낸다.here

Represents the correlation value of the average packet rate change amount when the flow through which data is transmitted from the node (v) is from the link (e _η ) to the link (e _{τ ),}

Denotes the correlation value of the average packet rate change amount in the flow in which data is transferred from the link (e _τ ) to the link (e _{η ).} And

Represents the correlation value of the average transmission amount change in the flow in which data is transferred from the link (e _η ) to the link (e _{τ ),}

Denotes the correlation value of the average transmission amount change amount in the flow in which data is transferred from the link (e _τ ) to the link (e _{η ).}

Here, f() denotes the flow direction, and ρ is a correlation function for calculating the degree of influence, and may be calculated as in Equation 2 according to the flow direction between the _{link e τ} and the link e _η.

) 또는 평균 전송량 변화량(

)을 나타내며, Y[]는 링크(e_η)를 통해 전송되는 평균 패킷율 변화량(

) 또는 평균 전송량 변화량(

)을 나타낸다.Here, n represents the number of viewpoints from the current point in time (t) to a previously specified point in time, and X[] is the average packet rate change amount transmitted through _{each link (e τ) (}

) Or the amount of change in the average transmission amount (

), and Y[] is the amount of change in the average packet rate transmitted through the _{link (e η) (}

) Or the amount of change in the average transmission amount (

).

,

,

,

)을 수학식 2에 기반으로 수학식 3과 같이 계산하여 4개의 상관도(λ₁, λ₂, λ₃, λ₄)를 원소로 포함하는 상관도 행렬(λ = {λ₁, λ₂, λ₃, λ₄})을 획득한다.Since it is inefficient for the influence feature acquisition unit 122 to calculate the degree of influence between links for all links on the network, four correlations between the _{target link (e τ} ) and the neighboring link (e _{η) among the plurality of links on the network} Values(

,

,

,

) Is calculated as in Equation 3 based on Equation 2, and a correlation matrix (λ = (λ ₁ , λ ₂ , _{) containing four correlations (λ 1} , λ ₂ , λ ₃ , λ _{4) as elements} λ ₃ , λ ₄ }) are obtained.

)을 획득할 수 있다.Equation 3 is the correlation matrix (λ = {λ ₁ , λ ₂ , λ ₃ , λ ₄ }) at the viewpoint t, and the influence characteristic acquisition unit 122 is the current viewpoint from the previous k viewpoint tk _{Each of the four correlations (λ 1} , λ ₂ , λ ₃ , and λ ₄ ) is calculated at each time point up to (t) to obtain k+1 correlation matrixes. Since the correlation matrix (λ(tk), λ(t-(k-1)),…, λ(t-1), λ(t)) at each viewpoint is obtained in the form of a 1 X 4 matrix, each viewpoint Transpose the correlation matrix (λ(tk), λ(t-(k-1)),…, λ(t-1), λ(t)) in 4 X 1 matrix form Obtain (λ ^T (tk), λ ^T (t-(k-1)),…, λ ^T (t-1), λ ^T (t)), and k+1 transposed correlation matrix (λ ^T (tk), λ ^T (t-(k-1)),…, λ ^T (t-1), λ ^T (t)) are arranged in the row direction as shown in Equation 4, and the influence characteristics (

) Can be obtained.

)이 획득되고, 링크 영향도 계산부(120)에서 영향도 특징(

)이 획득되면, 특징 결합부(130)는 레이턴시 특징(

)과 영향도 특징(

)을 인가받아 수학식 5와 같이 결합(concatenate)하여 타겟 링크(e_τ)에 대한 링크 특징(

)을 획득하여, 레이턴시 예측부(140)로 인가한다.In the link latency acquisition unit 110, the latency feature (

) Is obtained, and the influence degree feature (

) Is obtained, the feature combining unit 130 is a latency feature (

) And impact characteristics (

) Is authorized and concatenated as in Equation 5, and the link feature for _{the target link (e τ) (}

) Is obtained and applied to the latency prediction unit 140.

)의 패턴을 추정하여, 다음 시점(t+1)에서 타겟 링크(e_τ)를 통해 전달되는 데이터의 레이턴시(

)를 예측한다. 이때, 레이턴시 예측부(140)는 데이터가 전달되는 플로우 방향에 따라 제1 타겟 노드(v₇)로부터 제2 타겟 노드(v₈)로 타겟 링크(e_τ)에서의 레이턴시(

)를 예측할 수 있다.The latency prediction unit 140 is implemented as a neural network in which the pattern estimation method is learned, and the link feature applied from the feature combiner 130 (

) By estimating the pattern of the data transmitted through the _{target link (e τ} ) at the next point in time (t+1) (

) Is predicted. At this time, the latency predictor 140 may determine the latency in the target link e _{τ from the first target node v 7} to the second target node v ₈ according to the flow direction in which data is transmitted.

) Can be predicted.

)을 인가받고, 인가된 링크 특징(

)의 패턴을 추정하여, 레이턴시(y*_t+1)를 예측할 수 있다. 그러나 본 실시예에서 레이턴시 예측부(140)는 레이턴시(y*_t+1)를 더욱 정확하게 예측할 수 있도록 도 4와 같은 구성의 인공 신경망으로 구현될 수도 있다.As an example, the latency prediction unit 140 is implemented with a pre-learned Long Short Term Memory (LSTM) to provide a link feature (

) Is authorized, and the authorized link feature (

By estimating the pattern of ), the latency (y* _t+1 ) can be predicted. However, in this embodiment, the latency prediction unit 140 may be implemented as an artificial neural network having the configuration as shown in FIG. 4 to more accurately predict the _{latency (y*t+1).}

4 shows an example of a detailed structure of the latency prediction unit of FIG. 1.

Referring to FIG. 4, the latency estimation unit 140 may include a latency estimation unit 141, an estimation pattern analysis unit 142, a relationship value estimation unit 143, and a latency calculation unit 144.

)을 인가받고, 미리 학습된 패턴 추정 방식에 따라 링크 특징(

)의 패턴을 추정하여 타겟 링크(e_τ)의 이전 레이턴시를 추정한다. 이때 레이턴시 추정부(141)는 일 예로 양방향 LSTM(Bidirectional-LSTM: 이하 B-LSTM)으로 구현되어, 타겟 링크(e_τ)의 양방향 플로우에 각각에 대한 레이턴시(

,

)를 추정한다.The latency estimation unit 141 is a link feature (

) Is authorized, and the link feature (

) To estimate the previous latency of the target link (e _{τ ).} At this time, the latency estimating unit 141 is implemented as, for example, a bidirectional LSTM (Bidirectional-LSTM: hereinafter B-LSTM), and the latency for each of the bidirectional flows of the _{target link e τ}

,

) Is estimated.

As shown in FIG. 4, the B-LSTM includes a plurality of bidirectional LSTM cells 1410 to 141k. In addition, each of the plurality of bidirectional LSTM cells 1410 to 141k can be considered to functionally include two LSTM layers LSTML1 and LSTML2. Here, the two LSTM layers (LSTML1, LSTML2) included in each of the plurality of bidirectional LSTM cells 1410 to 141k may actually be implemented as two LSTM layers (LSTML1, LSTML2), but an artificial neural network that is generally implemented in software. It may be a functional block divided according to the flow of operation processing in the.

In addition, in the bidirectional LSTM cells 1410 to 141k, the first LSTM layer (LSTML1) is a forward LSTM cell that transmits and processes information in the forward direction based on time, and the second LSTM layer (LSTML2) is a backward LSTM cell. It can be viewed as a reverse LSTM cell that transfers and processes information.

)을 출력한다. 한편 제2 LSTM 레이어(LSTML2)는 각각 이전 k-t 시점으로부터 t시점까지 획득된 레이턴시 특징(X_(t-k), X_(t-k-1), … , X_(t-1), X_t) 중 대응하는 시점의 레이턴시 특징을 인가받고, 다음 시점의 제2 LSTM 레이어(LSTML2)에서 출력되는 역방향 상태 정보와 함께 패턴을 추정하여 역방향 레이턴시(

)를 출력한다.Accordingly, in the latency estimation unit 141 implemented as B-LSTM, the first LSTM layer (LSTML1) of the plurality of bidirectional LSTM cells 1410 to 141k is each obtained from the previous time point kt to the time point t (X _(tk) , X _(tk-1) ,…, X _(t-1) , X _t ) receive the latency feature of the corresponding viewpoint, and learn with forward state information output from the first LSTM layer (LSTML1) of the previous viewpoint The forward latency (

) Is displayed. On the other hand, the second LSTM layer (LSTML2) is the corresponding time point among the latency features (X _(tk) , X _(tk-1) ,…, X _(t-1) , X _t ) acquired from the previous time point kt to time point t, respectively. Is applied, and the pattern is estimated together with the reverse state information output from the second LSTM layer (LSTML2) at the next point in time.

) Is displayed.

)와 역방향 레이턴시(

)를 각각 수학식 6 및 7에 따라 계산할 수 있다.For example, in the case of the bidirectional LSTM cell 1410 for time t among a plurality of bidirectional LSTM cells 1410 to 141k, the forward latency (

) And reverse latency (

) Can be calculated according to Equations 6 and 7, respectively.

,

,

,

는 학습에 의해 획득된 B-LSTM의 순방향, 역방향 및 출력 가중치이고,

,

는 순방향, 역방향 및 출력 바이어스 값이며, σ는 활성화 함수(activation function)로서 시그모이드(sigmoid) 함수가 이용될 수 있다.here

,

,

,

Is the forward, backward and output weights of the B-LSTM obtained by learning,

,

Is a forward, reverse, and output bias value, and σ is an activation function, a sigmoid function may be used.

The calculation methods of Equations 6 to 7 are known as general calculation methods for B-LTSM, and thus are not described in detail here.

)으로부터 k-t부터 t까지 각 시점까지 각 시점에서 실제 모니터링된 레이턴시를 이용하여 추정된 레이턴시를 분석하여, 시점별 관계값(α_t-k, α_t-k-1, …, α_t-1,α_t)을 획득한다.When the forward latency and the reverse latency for each time point from kt to t are obtained by the latency estimation unit 141, the estimation pattern analysis unit 142 performs the forward and reverse latency at each time point, and the link characteristics (

) To each time point from kt to t, by analyzing the estimated latency using the actual monitored latency at each time point, the relationship values for each time point (α _tk , α _tk-1 , …, α _t-1 , α _t ) are calculated. Acquire.

)으로부터 t시점에서 측정된 레이턴시(

)를 확인하고, 확인된 레이턴시(

)와 레이턴시 추정부(141)에서 이전 t-1 시점에서 추정된 순방향 레이턴시(

)와 역방향 레이턴시(

) 사이의 관계를 나타내는 관계값(α_t)을 수학식 8에 따라 계산할 수 있다.As an example, the estimation pattern analysis unit 142 includes a link feature (

The latency measured at time t from) (

), and the confirmed latency (

) And the forward latency (

) And reverse latency (

A relationship value (α _t ) representing the relationship between) can be calculated according to Equation 8.

) 및 역방향 레이턴시(

)와 실제 측정된 레이턴시(

) 사이의 관계를 나타내는 관계값(α_t)을 도출할 수 있다.In Equation 8, tanhh ^-1 () is a function for inversely transforming the measured latency into the form of an output value of the LSTM to which the activation function is applied. That is, the estimated pattern analysis unit 142 estimates the forward latency (

) And reverse latency (

) And the actual measured latency (

A relationship value (α _t ) representing the relationship between) can be derived.

Then, the relationship values (α _tk , α _tk-1 , …, α _t-1 , α _t ) derived at each time point are transferred to the relationship value estimation unit 143.

The relationship value estimating unit 143 may be implemented as an LSTM including a plurality of LSTM layers (LSTML), and each of the plurality of LSTM layers (LSTML) is a relationship value (α _tk , α _tk-1 , …, α _{t- 1} , α _t ) of the corresponding relationship value is applied. And the LSTM layer (LSTML) at the last stage among the multiple LSTM layers (LSTML) is the relationship at time t+1 according to the pattern of the _{relationship values (α tk} , α _tk-1 , …, α _t-1 , α _t) You can output the value (α _{t+1 ).}

)와 역방향 레이턴시(

)를 수학식 8을 변형한 수학식 9에 적용함으로써, 타겟 링크(e_τ)의 다음 t+1 시점에서의 레이턴시(

)를 예측한다.The latency calculation unit 144 includes the relationship value α _t+1 estimated by the relationship value estimating unit 143 and the forward latency estimated by the latency estimating unit 141 (

) And reverse latency (

) To Equation 9 modified from Equation 8, the latency at the next point t+1 of the _{target link e τ}

) Is predicted.

~

)를 이용하여 미리 학습될 수 있다. 그러나 실시간으로 변화하는 네트워크 환경에 적응적으로 이용될 수 있도록, 링크 레이턴시 획득부(110)에서 획득된 타겟 링크(e_τ)의 레이턴시(

~

)의 업데이트에 따라 실시간으로 반복적으로 학습될 수 있다.Here, the artificial neural network of the latency prediction unit 140 is the latency of the target link e _{τ obtained by the link latency acquisition unit 110}

~

Can be learned in advance using ). However, in order to be adaptively used in a network environment that changes in real time, the latency of the target link (e _τ ) obtained by the link latency acquisition unit 110 (

~

) Can be repeatedly learned in real time according to the update.

)가 예측되면, 네트워크 관리 장치는 타겟 노드(v₇, v₈) 사이의 타겟 링크(e_τ)를 통해 전송해야하는 데이터가 존재하는 경우, 예측된 레이턴시(

)를 고려하여 데이터 전송 타이밍을 조절함으로써, SDN이 지연 민감 서비스를 안정적으로 제공하도록 할 수 있다.The latency at the next point in time (t+1) of the target link (e _{τ) (}

) Is predicted, when there is data to be transmitted through the target link (e _{τ ) between the target nodes (v 7} , v _{8 ), the predicted latency (}

) In consideration of the data transmission timing, so that the SDN can stably provide a delay-sensitive service.

5 shows a network management method according to an embodiment of the present invention.

)을 획득한다(S12).Referring to FIGS. 1 to 3, the network management method of FIG. 5 will be described, first, the latency of _{the target link e τ} on the network and the peripheral link e _η = {e ₁ , e ₂ ,…, e _{δ })} Is obtained by monitoring (S11). In addition, the latency of the target link (e _τ ) and the neighboring links (e ₁ ~ e ₆ ) for the previous predetermined period is accumulated, and the accumulated latency is arranged in a matrix form of a predetermined pattern, and the latency characteristic (

) Is obtained (S12).

In addition, the target node (v ₇ , v ₈ ) connected through the target link (e _τ ) in the network is passively monitored to obtain port information (S21). In addition, a correlation matrix indicating a correlation between the _{target link e τ} and at least one peripheral link e _η is obtained from the obtained port information (S22).

, λ₂ =

, λ₃ =

, λ₄ =

)을 수학식 2에 따라 계산한다(S22). 그리고 타겟 노드(v₇, v₈)에 연결되는 모든 주변 링크(e_η)와 타겟 링크(e_τ) 사이의 상관값을 누적하여, 4개의 상관도(λ₁, λ₂, λ₃, λ₄)를 계산한다. 4개의 상관도(λ₁, λ₂, λ₃, λ₄)이 계산되면, 4개의 상관도(λ₁, λ₂, λ₃, λ₄)를 원소로 포함하는 상관도 행렬(λ = {λ₁, λ₂, λ₃, λ₄})을 획득한다.To obtain the correlation matrix, first calculate the average packet rate change amount (Δpr) and the average transmission amount change amount (ΔΘ) from the acquired port information in a known manner, and the calculated average packet rate change amount (Δpr) and the average transmission amount change amount (ΔΘ) Assuming that) follows a normal distribution, _{four correlation values (λ 1} =) representing the degree of influence according to the flow direction between the _{target link (e τ} ) and the surrounding link (e _η)

, λ ₂ =

, λ ₃ =

, λ ₄ =

) Is calculated according to Equation 2 (S22). And by accumulating the correlation values between all the peripheral links (e _η ) connected to the target nodes (v ₇ , v ₈ ) and the target links (e _{τ ), four correlations (λ 1} , λ ₂ , λ ₃ , λ) ₄ ) is calculated. When four correlations (λ ₁ , λ ₂ , λ ₃ , λ ₄ ) are calculated, _{a correlation matrix containing four correlations (λ 1} , λ ₂ , λ ₃ , λ ₄ ) as elements (λ = { λ ₁ , λ ₂ , λ ₃ , λ ₄ }) are obtained.

)을 획득한다(S23). _{When the correlation matrix (λ = {λ 1} , λ ₂ , λ ₃ , λ ₄ }) at each time point is obtained during the previously specified period, each of the obtained correlation matrix is transposed to the transposed correlation matrix (λ ^T (tk), λ ^T (t-(k-1)),…, λ ^T (t-1), λ ^T (t))

) Is obtained (S23).

)과 영향도 특징(

)이 획득되면, 획득된 레이턴시 특징(

)과 영향도 특징(

)을 결합하여 하여 타겟 링크(e_τ)에 대한 링크 특징(

)을 획득한다(S30).Latency feature (

) And impact characteristics (

) Is acquired, the acquired latency characteristic (

) And impact characteristics (

) By combining the link feature for the target link (e _{τ) (}

) Is obtained (S30).

)을 이전 획득된 타겟 링크(e_τ)의 레이턴시(

~

)를 학습 데이터로 이용하여 미리 학습된 인공 신경망의 입력으로 인가함으로써, 타겟 링크(e_τ)의 다음 시점(t+1)에서의 레이턴시(

)를 예측한다(S40).And the obtained link feature (

) To the latency of the previously acquired target link (e _{τ) (}

~

) The latency of the application by using the learning data to the input of the artificial neural network learning in advance, a target link _(τ e) following the time (t + 1) of the (

) Is predicted (S40).

6 shows in detail an example of the latency prediction step of the present invention of FIG. 5.

)을 입력하여 타겟 링크(e_τ)의 다음 시점(t+1)에서의 레이턴시(

)를 예측할 수 있으나, 더욱 정확한 레이턴시 예측을 위해서는 다수의 인공 신경망을 이용하여 레이턴시(

)를 예측할 수도 있다.As described above, in this embodiment, the link feature (

) By entering the latency at the next point in time (t+1) of the target link (e _{τ) (}

) Can be predicted, but for more accurate latency prediction, a number of artificial neural networks are used to predict the latency (

) Can also be predicted.

)을 입력하여, 각 시점별 순방향 레이턴시와 역방향 레이턴시를 추정한다(S41). 그리고 링크 특징(

)으로부터 각 시점에서 측정된 레이턴시와 측정된 레이턴시 직전 시점에서 추정된 순방향 레이턴시와 역방향 레이턴시 사이의 관계를 나타내는 관계값(α_t-k, α_t-k-1, …, α_t-1,α_t)을 수학식 8에 따라 각 시점별로 계산한다(S42).In this case, first of all, link features to the artificial neural network implemented by B-LSTM (

) To estimate the forward and reverse latency for each time point (S41). And the link feature (

) From the measured latency at each time point and the relationship between the forward and reverse latency estimated at the time point immediately before the measured latency (α _tk , α _tk-1 , …, α _t-1 , α _t ) It is calculated for each time point according to Equation 8 (S42).

When the relationship values for each time point (α _tk , α _tk-1 , …, α _t-1 , α _t ) are calculated, the calculated relationship values (α _tk , α _tk-1 , …, α _t-1 ,α _t ) Is applied to the artificial neural network implemented by LSTM, and the relationship value α _t+1 at the next time point t+1 is estimated (S43).

)와 역방향 레이턴시(

)를 수학식 8에 적용하여 타겟 링크(e_τ)의 다음 t+1 시점에서의 레이턴시(

)를 계산한다(S44). _{When the relationship value (α t+1} ) at the next time point (t+1) is estimated, the estimated relationship value (α _t+1 ) and the estimated forward latency for the current time point (t) (

) And reverse latency (

) Is applied to Equation 8, and the latency at the next point t+1 of _{the target link (e τ) (}

) Is calculated (S44).

The method according to the present invention can be implemented as a computer program stored in a medium for execution on a computer. Here, the computer-readable medium may be any available medium that can be accessed by a computer, and may also include all computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, and ROM (Read Dedicated memory), RAM (random access memory), CD (compact disk)-ROM, DVD (digital video disk)-ROM, magnetic tape, floppy disk, optical data storage device, and the like.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those of ordinary skill in the art will understand that various modifications and equivalent other embodiments are possible therefrom.

Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

110: link latency acquisition unit 111: active monitoring unit
112: latency feature acquisition unit 120: link influence calculation unit
121: passive monitoring unit 122: influence feature acquisition unit
130: feature combining unit 140: latency predicting unit
141: latency estimation unit 142: estimation pattern analysis unit
143: relationship value estimation unit 144: latency calculation unit

Claims (19)

The latency of the target link for which the latency is to be predicted on the network and the latency of at least one neighboring link connected to each of the two target nodes connected by the target link is monitored for a predetermined period of time, and the obtained latency is obtained in the form of a matrix of a predetermined pattern. A link latency acquisition unit arranged to obtain a latency characteristic;
The target node is monitored to obtain port information, a correlation matrix indicating a correlation between the target link and at least one neighboring link is obtained from the obtained port information, and a correlation matrix obtained for a predetermined period is previously A link influence degree calculation unit that converts and arranges in a designated form to obtain an influence degree characteristic;
A feature combining unit that combines the latency feature and the influence feature to obtain a link feature for the target link; And
A network management apparatus in a software-defined network environment, comprising: a latency predictor for estimating a latency at a next point in time of the target link by receiving the link characteristic by learning a pattern estimation method in advance. The method of claim 1, wherein the latency acquisition unit
An active monitoring unit that monitors a plurality of links on the network to obtain latency of the target link and the at least one peripheral link; And
The latency of accumulating and storing the obtained latency of the target link and the at least one neighboring link, and separating each of the accumulated and stored latency of the target link and the latency of at least one neighboring link and arranging in chronological order to obtain the latency characteristic A network management device in a software-defined network environment including a feature acquisition unit. The method of claim 1, wherein the link influence calculation unit
A passive monitoring unit that monitors at least one port of each of a plurality of nodes of the network using an open flow protocol of a software defined network to obtain transmission byte, reception byte, transmission packet, reception packet, and duration port information; And
The average packet rate change amount and the average transmission amount change amount are calculated in a predetermined manner using the port information, and from the calculated average packet rate change amount and the average transmission amount change amount, the target link for each flow direction and each time point in which data is transmitted, and at least one A predetermined number of correlation values representing the degree of influence between each of the neighboring links are calculated, and a correlation matrix having a correlation obtained by accumulating each of the calculated correlation values for each of the neighboring links as an element is obtained. A network management device in a software-defined network environment comprising an influence degree feature acquisition unit arranged in a form to obtain the influence degree feature. delete delete delete The method of claim 1, wherein the latency prediction unit
Implemented as a bidirectional LSTM including a plurality of bidirectional LSTM cells, a pattern learned in advance from the forward latency at the previous point in time and the reverse latency at the next time point applied from the latency feature at the corresponding point in the link feature and the adjacent bidirectional LSTM cell A latency estimating unit for estimating a forward latency and a reverse latency of a corresponding viewpoint according to the estimation method;
_{The relationship values for each point in time (α tk} , α _tk-1 , …, α _t-1 , which indicate the relationship between the latency of each point in the link feature and the forward and backward latency estimated at the point immediately before the delay of each point in time) a pattern analysis unit that acquires α _{t );}
A relationship value that estimates the relationship value _{(α t+1} ) at the next time point from the relationship values (α _tk , α _tk-1 , …, α _t-1 , α _t ) for each time point according to a previously learned pattern estimation method Estimation unit; And
The estimated relationship value (α _t+1 ) and the estimated forward latency for the current point in time (t) (

) And backward latency (

Using ), the latency at the next point in time (

A network management device in a software-defined network environment including a latency calculation unit that calculates ). The method of claim 7, wherein the pattern analysis unit
Latency determined from link characteristics at time (t) (

) And the forward latency estimated at the previous time t-1 by the previous latency estimating unit 141 (

) And backward latency (

) Between
Equation

A network management device in a software defined network environment that calculates and obtains a relationship value (α _{t) that satisfies.} The method of claim 8, wherein the latency calculator
The estimated relationship value (α _t+1 ) and the estimated forward latency for the current point in time (t) (

) And backward latency (

) From the equation

According to the latency at the next point in time (

A network management device in a software-defined network environment that calculates ). The method of claim 7, wherein the latency estimating unit and the relationship value estimating unit
A network management device in a software-defined network environment that learns by using the latency for each time point included in the link feature as learning data. The latency of the target link for which the latency is to be predicted on the network and the latency of at least one neighboring link connected to each of the two target nodes connected by the target link is monitored for a predetermined period of time, and the obtained latency is obtained in the form of a matrix of a predetermined pattern. Arranging them to obtain a latency characteristic;
The target node is monitored to obtain port information, a correlation matrix indicating a correlation between the target link and at least one neighboring link is obtained from the obtained port information, and a correlation matrix obtained for a predetermined period is previously Converting and arranging in a designated form to obtain an influence characteristic;
Obtaining a link characteristic for the target link by combining the latency characteristic and the influence characteristic characteristic; And
And estimating a latency at a next point in time of the target link by learning a pattern estimation method in advance and receiving the link feature. The method of claim 11, wherein obtaining the latency characteristic comprises:
Monitoring a plurality of links on the network to obtain latency of the target link and the at least one peripheral link; And
Accumulating and storing the acquired latency of the target link and the at least one neighboring link, classifying each of the accumulated and stored latency of the target link and the latency of at least one neighboring link, and arranging in chronological order to obtain the latency characteristic Network management method of a software-defined network environment comprising a. The method of claim 11, wherein acquiring the influence degree feature comprises:
Monitoring at least one port of each of a plurality of nodes of the network using an openflow protocol of a software-defined network to obtain transmission byte, reception byte, transmission packet, reception packet, and duration port information; And
Calculating an average packet rate change amount and an average transmission amount change amount in a predetermined manner using the port information;
Calculating a correlation value of a predetermined number from the calculated average packet rate change amount and the average transmission amount change amount indicating an influence between each flow direction in which data is transmitted and between a target link for each time point and each of at least one neighboring link;
Obtaining a correlation matrix having a correlation obtained by accumulating each of the calculated correlation values for each of the surrounding links as an element; And
A network management method of a software-defined network environment comprising the step of arranging the obtained correlation matrix in a predetermined form. delete delete delete The method of claim 11, wherein estimating the latency
To a bidirectional LSTM including a plurality of bidirectional LSTM cells for which the pattern estimation method is pre-learned, the latency feature at the time point corresponding to the link feature and the forward latency at the previous time point output from the adjacent bidirectional LSTM cell and the reverse latency at the next time point are applied to the bidirectional LSTM cell. Thus, estimating a forward latency and a reverse latency of a corresponding viewpoint;
_{The relationship values for each point in time (α tk} , α _tk-1 , …, α _t-1 , which indicate the relationship between the latency of each point in the link feature and the forward and backward latency estimated at the point immediately before the delay of each point in time) obtaining α _t );
_{Estimating a relationship value (α t+1} ) at a next time point from the relationship values (α _tk , α _tk-1 , ..., α _t-1 , α _t ) according to a previously learned pattern estimation method; And
The estimated relationship value (α _t+1 ) and the estimated forward latency for the current point in time (t) (

) And backward latency (

Using ), the latency at the next point in time (

A network management method of a software-defined network environment comprising the step of calculating ). The method of claim 17, wherein the obtaining of the relationship value for each time point comprises:
Latency determined from link characteristics at time (t) (

) And the forward latency estimated at the previous time t-1 by the previous latency estimating unit 141 (

) And backward latency (

) Between
Equation

Network management method of a software-defined network environment that is obtained by calculating a relationship value (α _{t) that satisfies the.} The method of claim 18, wherein calculating the latency
The estimated relationship value (α _t+1 ) and the estimated forward latency for the current point in time (t) (

) And backward latency (

) From the equation

According to the latency at the next point in time (

) To calculate the network management method of a software-defined network environment.

Images