JP2004088153A - How to identify network bottlenecks - Google Patents

Abstract

A bottleneck identification method capable of identifying a bottleneck in a network with higher accuracy is provided.
According to the present invention, a resource use status Xi is measured for a plurality of locations i of a network, and a service quality Y at a specific location of the network is measured. Next, a model function of the regression equation is determined, a regression analysis is performed on the service quality of the service using the regression equation Yi = fi (Xi), and the service quality obtained in the regression analysis step is obtained. Determining the contribution ratio Ri = var (Yi) / var (Y) by using the estimated value Yi of the above and the measured value Y of the service quality, and specifying a position where the contribution ratio Ri is maximum as a bottleneck. I can.
[Selection diagram] Fig. 1

Description

ここで、ｍｉはｆｉの自由度であり、多項式の場合は次数＋１、
Ｌｉは尤度であり、
【数２】

これらＡＩＣ、ＢＩＣ、ＡＩＣＣを用いてボトルネックを特定する場合、求められた値が最小となる箇所ｉをボトルネックとして特定することができる。
なお、ここで求めたボトルネックの識別番号ｉをＩとおく。
【００１６】
ステップ３　トラヒックが確率的に発生すると考えた場合、各測定値も確率的に変動するため、ボトルネックを誤判定する確率も存在する。また、Ｘ_ＩとＸ_ｊ（ｊ∈｛１，−−−，Ｎ｝，ｊ≠Ｉ）間で相関が強い場合、回帰式や相関係数が近いものとなるため、さらにボトルネックを誤判定する確率が増加するおそれがある。そこで、本発明の一例として、確率検定手法を用いてボトルネックの特定が誤判定か否かを判断する工程を具える。そして、この判断工程において、構成要素Ｉ以外の構成要素についてボトルネックとなる確率が構成要素Ｉと同程度であるかどうかの判定を次の様に行う。
各ｊ（ｊ∈｛１，−−−，Ｎ｝，ｊ≠Ｉ）に開し、Ｅ_ｊ，Ｅ_Ｉが正規分布をなしていると仮定し、ｆｉの自由度をｄｆ_ｉとおくと、Ｆ_ｊ，Ｉ　＝ｖａｒ　（Ｅ_ｊ）／ｖａｒ（Ｅ_Ｉ）は自由度（Ｔ−ｄｆ_ｊ，Ｔ−ｄｆ_Ｉ）のＦ分布に従う。そこでＦが事前に決めた有意度より大きければＥ_ｊと　Ｅ_Ｉは同一分布でありｆ_ｊ，ｆ_Ｉは同程度に精度良く回帰できている。そのため、構成要素ｊ又はＩのどちらがボトルネックであるかは測定結果からは判定できないと判断し、構成要素ｊもボトルネック候補として出力する。
回帰モデルとして局所回帰など、モデルの自由度が容易に求められない場合は、ｄｆ_ｉとして等価自由度を用いることによりＦ_ｊ，Ｉ　＝ｖａｒ　（Ｅ_{ｊ） 　９}ｖａｒ（Ｅ_Ｉ）に対して同様の検定を行うことができる。従って、箇所ｉ及びｊの回帰の残差Ｅｉ及びＥｊの分散が同一か否かをＦ検定を用いて判断し、得られたＦ値の結果に基づいてボトルネックの特定が誤判定か否かを判断することができる。
なお、回帰式として線形モデル（１次多項式）を用いた場合、ＹとＸ_ｉの相関係数をρＹ，Ｘ_ｉ　とおくと、寄与率はρ^２Ｙ，Ｘ_ｉと表される。よって、寄与率に基づいた順位づけは相関係数に基づいた順位づけと同義である。
【００１７】
【発明の実施の形態】
シミュレーションにより測定したデータに対して、本発明を適用した結果について説明する。
［シミュレーションモデル］
シミュレーションモデルとして用いたネットワークの構成を図１に示す。事業者は、ルータ１、ルータ２に接続された顧客にサーバからファイルをダウンロードするための回線１、回線２を提供している。管理対象は回線１、回線２であり、リソースの使用状況として回線の使用率、サービス品質としてルータ２に接続された疑似顧客が測定したダウンロード時間を測定している。その他の条件を表１に示す。
【００１８】
【表１】

【００１９】
分析には、サービス品質の測定値（Ｙ）として１０秒毎に３ＫＢｙｔｅのファイルをダウンロードするのに要する時間を測定したものの１分間の平均値を用い、リソース使用状況の測定値（Ｘ_１，Ｘ_２）として１分毎の各回線の使用帯域を用いた。シミュレーションは７２００秒行い、サンプル数（Ｔ）は１２０である。
【００２０】
ソースの使用状況である回線使用率の測定値は、回線１についてはＸ１＝３９．１％であり、回線２についてはＸ２＝３７．１％であった。
【００２１】
尚、実際のボトルネックを明らかにするため、回線１の帯域を十分大きいと考えられる２２５Ｍｂｐｓとした場合、回線２の帯域を十分大きいと考えられる２２５Ｍｂｐｓとした場合のそれぞれについて個別にシミュレーションを行った。その結果、サービス品質の平均値は表２に示す様になった。この結果より、回線２を十分大きくした場合の方がサービス品質の改善が大きく、回線２がボトルネックであることがわかる。しかし表１に示す条件で行ったシミュレーションでは、回線１、回線２の使用率はそれぞれ３９．１％、３７．１％であり、単純な回線使用率の比較では回線１がボトルネックであると誤った判断をするおそれがある。すなわち、回線使用率だけを用いて判断したのでは、ボトルネックを正確に特定できないことになる。
【００２２】
【表２】
個別シュミレーションによるボトルネックの特定
回線１の帯域を充分大きくした場合　　０．０８３７ｓｅｃ
回線２の帯域を充分大きくした場合　　０．００４１３ｓｅｃ
【００２３】
次に、回帰分析の結果について説明する。本例では、ネットワークの複雑性に柔軟に対応するため、５次の回帰式を用いて回帰分析を行う。ここで、回線１及び２の利用率の測定値Ｘ１及びＸ２を用いて得られたサービス品質の推定値Ｙ１及びＹ２は以下の回帰式から得た。
Ｙ_１＝ｆ_１（Ｘ_１）＝−２．９３＋（３．３５ｅ−０６）Ｘ_１＋（−１．３１ｅ−１２）Ｘ_１ ^２＋（２．３８ｅ−１９）Ｘ_１ ^３＋（−１．９５ｅ−２６）Ｘ_１ ^４＋（５．９０ｅ−３４）Ｘ_１ ^５
Ｙ_２＝ｆ_２（Ｘ_２）＝０．１６３＋（８．２９ｅ−０８）Ｘ_２＋（−１．９６ｅ−１３）Ｘ_２ ^２＋（２．５９ｅ−１９）Ｘ_２ ^３＋（−１．１８ｅ−２５）Ｘ_２ ^４＋（２．２２ｅ−３２）Ｘ_２ ^５
【００２４】
回帰分析の結果により得られたサービス品質の回帰曲線、すなわち回帰分析により得られたサービス品質の推定値を測定値と共に図２、３に示す。図２は回線１の回線利用率Ｘ１に回帰した結果を示し、図３は回線２の回線利用率Ｘ２に回帰した結果を示す。図２及び図３において、横軸は回線使用率を示し、縦軸はサービス品質すなわちファイルのダウンロード所要時間を示し、十字印は測定値を示し、破線は回帰曲線を示す。これらの図からＸ２の方がＹと相関が強く回線２の使用帯域がボトルネックであることは直感的にもわかるが、以降では数値計算によりそれを示す。
【００２５】
次に、サービス品質の測定値と回帰分析により得られたサービス品質の推定値について寄与率の計算結果を示す。回線１の回線利用率Ｘ１に回帰して得られた推定値に対する寄与率Ｒ１及回線２の回線利用率Ｘ２に回帰して得られた推定値に対する寄与率びＲ２は以下の通りである。
Ｒ_１＝０．３７４　　　　　　　　　　　　　　　　（３）
Ｒ_１＝０．９０８　　　　　　　　　　　　　　　　（４）
この結果から、回線２の寄与率が最大となり、回線２がボトルネックである旨を出力する。この寄与率から得られた結果は、前述した回線帯域を極めて大きくした個別シミュレーション結果に強く相関しており、ネットワーク中で発生したボトルネックの箇所を高い精度で特定することができることが実証された。
【００２６】
Ｘ_１も　Ｘ_２と同様にボトルネックであるかどうかを、回帰の残差Ｅ_１とＥ_２が同一分布であるかどうか検定することにより行うこともできる。「Ｅ_１とＥ_２は同一分布である」と言う帰無仮説に対して回帰の残差のＦ値Ｆ_１，２を求めると、

有意度１％で両側検定するときのＦの値は１．６２５であり、帰無仮説は十分棄却でき、Ｅ_１とＥ_２の差には十分優位な差があると言える。よって、回線１はボトルネックではないと判断する。
【００２７】
次に、本発明によるネットワークのボトルネック特定装置について説明する。本発明による装置は、管理すべきネットワークの複数の箇所のリソース使用状況、例えばネットワーク中の複数の回線の回線使用率を測定する手段と、ネットワーク中の特定の箇所のサービス品質を測定する手段、例えばネットワークに接続した模擬端末装置とを具える。そして、性能管理すべきネットワークに応じてモデル関数を決定し、回帰式を用いて測定されたリソースの使用状況について回帰処理を実行し、得られたサービス品質の推定値とサービス品質の測定値との間の寄与率を求め、寄与率が最大となる箇所にボトルネックが発生したものと特定する手段とを具える。
【００２８】
【発明の効果】
以上説明したように、本発明によれば、管理すべきネットワーク中の複数箇所のリソース使用状況を測定し、回帰式を用いて得られた測定値に対して回帰分析処理を実行し、得られたサービス品質の推定値とサービス品質の測定値とを用いて寄与率を算出し、寄与率が最大になる箇所をボトルネックとして特定しているので、一層高い精度でボトルネックの発生箇所を特定することができる。
【図面の簡単な説明】
【図１】本発明によるボトルネック特定方法を実証するためのシミュレーションモデルとして用いたネットワークの構成を示す線図である。
【図２】シミュレーションモデルにおいてサービス品質Ｙを回線１の測定値に回帰した場合の回帰分析結果及び測定値を示すグラフである。
【図３】シミュレーションモデルにおいてサービス品質Ｙを回線２の測定値に回帰した場合の回帰分析結果及びサービス品質の測定値を示すグラフである。
【符号の説明】
Ｓ　サーバ
Ｃ　顧客端末装置
Ｐ　模擬顧客端末装置[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a computer system including a plurality of elements and a performance management of a computer network, and more particularly to a method and apparatus for identifying a bottleneck of a network.
[0002]
[Prior art]
In computer systems and computer networks, performance increases when the load increases, resulting in deterioration in the quality of service provided by them. Therefore, in order to maintain the quality of service provided by the built system,
1. It constantly monitors system performance and detects performance degradation.
2. When the performance degradation is detected, a location where the performance is most greatly improved by improvement is identified, that is, a bottleneck location is specified, and an improvement measure is taken.
It is necessary to perform the following two points.
[0003]
[Method for regularly monitoring system performance]
Methods for constantly monitoring the performance of the system can be roughly classified into the following two methods.
(Monitor resource usage status)
A method that collects data on resource usage in each component of the system, and estimates and monitors the performance of the component and the quality of the entire service from the data.
In this method, no new traffic is generated for the measurement (passive measurement), so that the load on the system is light and the performance of other users is hardly degraded. Standardization of protocols, items, etc. for data collection is progressing, and necessary functions are often implemented as standard, and have been widely used so far.
[0004]
In order to use this method, it is necessary to clarify in advance the relationship between the resource usage status of each component and the quality of service provided by the system. The relationship can be obtained by modeling the system and applying analysis or simulation to the model. At this time, the relationship obtained on the model may be different from the relationship on the real system because the model is complicated and it is difficult to derive the relationship or the model is not sufficiently modeled. In particular, recent research results have shown that Internet traffic has self-similarity, which indicates that the performance of the system tends to be lower than the results of conventional studies.
[0005]
(Estimation of service quality itself)
A method of measuring and monitoring the performance of the entire system to be managed, that is, the quality of the service provided to the user. For this measurement, it is necessary to use a measuring device that simulates a user or to introduce a mechanism for reporting the quality when the user uses the service to a system administrator.
When a measurement device simulating a user is used, it is necessary to flow traffic for measurement (active measurement). As a result, the performance of the system may be reduced, and more equipment must be installed, which increases the cost. However, in recent years, systems that play an important role in society and systems that provide services that require strict service quality are increasing, and cases where this method is used to accurately grasp service quality Is increasing.
When introducing a mechanism for reporting the quality when a user uses a service to a system administrator, there is no general-purpose method. Since the service provider needs to distribute or modify the device and software on the user side, it can be applied only to some limited services.
[0006]
[Method to identify bottleneck location that caused performance degradation]
If only the measured value of the service quality is measured without using the data for each component of the system, the bottleneck cannot be specified. On the other hand, if we only monitor the usage of resources of the system components and do not measure the service quality, the relationship between the resource usage status of each component obtained in advance and the quality of service provided by the system Thus, the magnitude of the influence of each element on the service quality is estimated.
[0007]
[Problems to be solved by the invention]
However, as described above, since the model is complicated, it is often difficult to derive the relationship between the resource usage status and the service quality. Furthermore, since the model cannot be sufficiently modeled, there are many differences between the relationship obtained on the model and the relationship with the actual system, and there is a disadvantage that reliability is low.
[0008]
Further, there is a method of dividing the service quality of the entire system into each part, and comparing the results of the quality measurement performed on each part to identify a bottleneck. However, this method increases the disadvantages of active measurement (possibility of deteriorating system performance, the cost of installing many devices), and cannot identify a bottleneck more than the minimum unit of service. For example, when a part is encrypted on a communication path, it is impossible to perform active measurement only on a device that performs decryption under the same conditions as an actual service.
[0009]
Accordingly, an object of the present invention is to provide a bottleneck identification method capable of identifying a bottleneck in a network with higher accuracy.
[0010]
[Means to solve the problem]
It has been found that, when the service quality is reduced, a change in the usage state of the bottleneck resource and a change in the service quality are correlated, such as an increase in the resource usage rate at the bottleneck. Therefore, a location where the correlation between the usage status of each resource and the service quality is strongest can be set as a bottleneck.
[0011]
The present invention is based on the premise that a set of a service quality at a certain time and a numerical value representing the use state of each resource exists for many time sections. Note that these numerical values may be measured values themselves or processed values as long as they reflect the respective values. The processed value is exemplified by a statistical value such as an average value of many measurements. In addition, the presence of “many” data means that the degree of freedom of the residual is not zero in the regression analysis performed below.
[0012]
The service quality of a specific portion of the network to be performance-managed is measured, the usage of resources at a plurality of locations in the network is measured, and the components of the system in which the usage of these resources is measured (network locations) Are sequentially assigned an identification number, and the number is set to i. Represents a measure of service quality Y, attributes reflecting the usage and queue length, such as resource usage Xi, representing the measured value y (t), x i ( t). Here, if N is the number of system components measuring the resources, and T is the number of samples, i {1,2, ---, N}, t {1,2, ---, T}.
[0013]
Step 1 After determining the regression model function performs regression analysis to X _i of Y For each i, derive the regression equation Y = f i a _{(X i).} Further, the estimated value Yi of the service quality obtained by the regression equation is defined as Y _i = f _i (X _i ). Further, the residuals Ei _regression, and E _i = _Y-Y i. Further, let the values at each time t be Y _i (t) and E _i (t).
[0014]
Step 2 The relevance equation is used to determine the degree of conformity Ri of the value Yi obtained for each point to the measured value Y of the service quality, and the point determined to be the most suitable by the degree of conformity is specified as a bottleneck. For example, a contribution rate can be used as the degree of adaptation, and a contribution rate R _i = var (Y _i ) / var (Y) is obtained for each point i of the network. Here, var represents an unbiased estimator of the variance. Identify and R _i is made describes well Y with X _i when the maximum, i.e. be a most bottleneck has a high point, and outputs.
[0015]
It is clear that when a different model function is used for each location or component i of the network, particularly when a model function having a different degree of freedom is used, the use of a model function having a higher degree of freedom is more applicable and better. Therefore, when the contribution rate is used as an evaluation criterion, the degree of freedom is affected. Therefore, when the degrees of freedom of the respective portions are different, an evaluation criterion in which the influence of the degrees of freedom such as AIC, BIC, and AICC is reduced can be used instead. Here, AIC, BIC, and AICC at each point i of the network are represented by the following equations.
(Equation 1)

Here, mi is the degree of freedom of fi, and in the case of a polynomial, the order is +1;
Li is the likelihood,
(Equation 2)

When a bottleneck is specified using these AIC, BIC, and AICC, the location i where the obtained value is minimum can be specified as the bottleneck.
The bottleneck identification number i obtained here is set to I.
[0016]
Step 3 When it is considered that traffic occurs stochastically, each measurement value also fluctuates stochastically, and there is also a probability of erroneously determining a bottleneck. Further, _{X I} and _{X j (j∈ {1, ---} , N}, j ≠ I) When is a strong correlation between, for regression equation and correlation coefficient is close, the more bottlenecks misjudgment There is a possibility that the probability of doing this will increase. Therefore, as an example of the present invention, the method includes a step of determining whether or not the bottleneck is erroneously determined using a probability test method. Then, in this determination step, it is determined whether or not the probability of becoming a bottleneck with respect to the components other than the component I is about the same as that of the component I as follows.
Each j (j∈ {1, ---, N}, j ≠ I) is _opened, E j, assuming _{E I} have no normal distribution, placing the freedom of fi and df _i, F _{j, I} = var (E _j ) / var (E _I ) follows an F distribution with degrees of freedom (T-df _j , T-df _I ). Therefore, if F is greater than the significance determined in advance, E _j and E _I have the same distribution, and f _j and f _I can be regressed to the same degree of accuracy. Therefore, it is determined that it is impossible to determine which of the components j or I is the bottleneck from the measurement result, and the component j is also output as a bottleneck candidate.
Such as local regression as a regression model, if the degree of freedom of the model is not easily sought is, _F j by using the equivalent freedom as df _i, similarly with respect to _{I = var (E j) 9} var (E I) Can be tested. Therefore, it is determined using the F test whether or not the variances of the regression residuals Ei and Ej of the regression at the points i and j are the same, and whether the bottleneck is erroneously determined based on the obtained F-value result. Can be determined.
In the case of using a linear model (first order polynomial) as a regression equation, RoY correlation coefficients Y and _{X i,} by placing the _{X i,} the contribution rate is expressed as [rho ² Y, _{X i.} Therefore, the ranking based on the contribution rate is synonymous with the ranking based on the correlation coefficient.
[0017]
BEST MODE FOR CARRYING OUT THE INVENTION
The result of applying the present invention to data measured by simulation will be described.
[Simulation model]
FIG. 1 shows the configuration of a network used as a simulation model. The business operator provides a line 1 and a line 2 for downloading a file from a server to a customer connected to the router 1 and the router 2. The line 1 and line 2 are managed, and the line usage rate is measured as the resource use status, and the download time measured by the pseudo customer connected to the router 2 is measured as the service quality. Table 1 shows other conditions.
[0018]
[Table 1]

[0019]
In the analysis, the measured value of the resource use status (X ₁ , X ₁ , X _{1) The} bandwidth used for each line every minute was used as ₂ ). The simulation was performed for 7200 seconds, and the number of samples (T) was 120.
[0020]
The measured value of the line utilization rate, which is the source use status, was X1 = 39.1% for line 1, and X2 = 37.1% for line 2.
[0021]
Note that in order to clarify the actual bottleneck, simulations were individually performed for the case where the bandwidth of the line 1 was set to 225 Mbps, which is considered to be sufficiently large, and for the case where the band of the line 2 was set to 225 Mbps, which was considered to be sufficiently large. . As a result, the average value of the service quality was as shown in Table 2. From this result, it can be seen that when the line 2 is sufficiently large, the improvement in service quality is greater, and the line 2 is a bottleneck. However, in the simulations performed under the conditions shown in Table 1, the usage rates of the line 1 and the line 2 are 39.1% and 37.1%, respectively, and a simple comparison of the line usage rates indicates that the line 1 is a bottleneck. There is a risk of making an incorrect decision. In other words, a bottleneck cannot be specified accurately by making a determination using only the line usage rate.
[0022]
[Table 2]
0.0837 sec when the bandwidth of the specific line 1 which is the bottleneck due to individual simulation is made sufficiently large
When the bandwidth of line 2 is sufficiently increased 0.00413 sec
[0023]
Next, the result of the regression analysis will be described. In this example, a regression analysis is performed using a fifth-order regression equation in order to flexibly respond to the complexity of the network. Here, the estimated values Y1 and Y2 of the service quality obtained using the measured values X1 and X2 of the utilization rates of the lines 1 and 2 were obtained from the following regression formula.
_{Y 1 = f 1 (X 1} ) = - 2.93+ (3.35e-06) X 1 + (- 1.31e-12) X 1 2 + (2.38e-19) X 1 3 + (- 1 ^{_{.95e-26) X 1 4 +}} (5.90e-34) X 1 5
_{Y 2 = f 2 (X 2} ) = 0.163 + (8.29e-08) X 2 + (- 1.96e-13) X 2 2 + (2.59e-19) X 2 3 + (- 1. ^{_{18e-25) X 2 4 +}} (2.22e-32) X 2 5
[0024]
The regression curve of the service quality obtained by the result of the regression analysis, that is, the estimated value of the service quality obtained by the regression analysis is shown in FIGS. FIG. 2 shows the result of returning to the line utilization X1 of the line 1, and FIG. 3 shows the result of returning to the line utilization X2 of the line 2. 2 and 3, the horizontal axis indicates the line usage rate, the vertical axis indicates the service quality, that is, the time required for downloading the file, the cross mark indicates the measured value, and the broken line indicates the regression curve. From these figures, it can be intuitively understood that X2 has a stronger correlation with Y and the band used by the line 2 is a bottleneck, but it will be shown by numerical calculations hereinafter.
[0025]
Next, the calculation results of the contribution ratio for the measured value of the service quality and the estimated value of the service quality obtained by the regression analysis will be described. The contribution ratio R1 to the estimated value obtained by returning to the line utilization ratio X1 of the line 1 and the contribution ratio R2 to the estimated value obtained by returning to the line utilization ratio X2 of the line 2 are as follows.
R ₁ = 0.374 (3)
R ₁ = 0.908 (4)
From this result, the contribution rate of the line 2 is maximized, and it is output that the line 2 is a bottleneck. The results obtained from this contribution rate are strongly correlated with the individual simulation results described above in which the line bandwidth was extremely large, and it was demonstrated that the location of the bottleneck that occurred in the network could be specified with high accuracy. .
[0026]
Whether X ₁ is a bottleneck as well as X ₂ can be determined by testing whether the regression residuals E ₁ and E ₂ have the same distribution. When the F-values F ₁ and F ₂ of the regression residual are obtained for the null hypothesis that “E ₁ and E ₂ have the same distribution”,

The value of F when the two-tailed test with a significance level of 1% is 1.625, the null hypothesis is sufficiently rejected, it can be said that the difference between E ₁ and E ₂ have enough significant difference. Therefore, it is determined that the line 1 is not a bottleneck.
[0027]
Next, a network bottleneck identification device according to the present invention will be described. The apparatus according to the present invention comprises: means for measuring resource use status of a plurality of points in a network to be managed, for example, a line utilization rate of a plurality of lines in the network; and means for measuring a service quality of a specific point in the network. For example, a simulation terminal device connected to a network is provided. Then, a model function is determined according to the network to be performance-managed, regression processing is performed on the resource usage measured using the regression equation, and the obtained service quality estimation value and service quality measurement value are obtained. And a means for determining that a bottleneck has occurred at a position where the contribution ratio is maximum.
[0028]
【The invention's effect】
As described above, according to the present invention, a resource use situation at a plurality of locations in a network to be managed is measured, and a regression analysis process is performed on a measurement value obtained by using a regression equation. The contribution rate is calculated using the estimated value of service quality and the measured value of service quality, and the location where the contribution rate becomes maximum is identified as the bottleneck, so the location where the bottleneck occurred can be identified with higher accuracy can do.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a network used as a simulation model for demonstrating a bottleneck identification method according to the present invention.
FIG. 2 is a graph showing a regression analysis result and a measured value when service quality Y is regressed to a measured value of a line 1 in a simulation model.
FIG. 3 is a graph showing a regression analysis result and a measured value of service quality when the service quality Y is regressed to a measured value of the line 2 in the simulation model.
[Explanation of symbols]
S Server C Customer terminal device P Simulated customer terminal device

Claims (8)

A method for identifying network bottlenecks to be performance-managed,
Measuring the resource usage Xi for a plurality of locations i in the network;
Measuring the quality of service Y at a particular point in the network;
Determining a model function of the regression equation, and performing a regression analysis of the service quality resource usage state Xi using the regression equation Yi = fi (Xi);
A step of determining a fitness level Ri of the value Yi determined for each location with respect to the service quality measurement value Y using the regression equation, and identifying a location determined to be most suitable by the fitness level Ri as a bottleneck; A method for identifying a bottleneck in a network, comprising: 2. The method according to claim 1, wherein when the degree of freedom of each portion i of the network is the same, a contribution ratio = var (Yi) / var (Y) is determined as the fitness Ri. A method for identifying a bottleneck of a network, characterized in that a portion where the maximum is determined is identified as a bottleneck. 2. The network bottleneck identification method according to claim 1, wherein when the degrees of freedom of the respective parts of the network are different, AICi, BICi, or AICCi of each part is obtained as the fitness Ri, and the obtained AICi, BICi are obtained. , Or a location where the value of AICCi is minimum is identified as a bottleneck. 4. The method according to claim 1, wherein regression analysis is performed using a polynomial regression equation in the regression analysis step. 5. 5. The method according to claim 1, wherein a line usage rate is used as the resource use status, and a time required for downloading data of a predetermined capacity is used as the service quality. 6. How to identify network bottlenecks. The method for identifying a bottleneck in a network according to any one of claims 1 to 5, further comprising a step of determining whether the bottleneck is erroneously determined using a probabilistic test method. How to identify network bottlenecks. 7. The method according to claim 6, wherein the variances of Ei and Ej at the points i and j are the same if the residual Ei of the regression at the point i in the network is Ei = Y−Yi. Is determined using an F test, and it is determined whether or not the bottleneck is erroneously determined based on the result of the obtained F value [Fi, j = var (Ej) / var (Ei)]. How to identify network bottlenecks. An apparatus for identifying a network bottleneck for performance management,
Means for measuring resource usage Xi at a plurality of locations in the network;
Means for measuring the quality of service Y at a particular point in the network;
The regression equation Yi = fi (Xi) is used to perform a regression analysis on the service usage Xi of the service quality Yi, and the value Yi obtained for each location using the regression equation is compared with the measured value Y of the service quality. A bottleneck identification device, comprising: a processing unit that obtains a degree of fitness Ri and identifies a location determined to be most suitable by the degree of fitness Ri as a bottleneck.

Images