Monitoring data analyzing apparatus, monitoring data analyzing method, and monitoring data analyzing program

[0001] This application is a National Stage of International Application No. PCT/JP2011/078714 filed Dec. 12, 2011, claiming priority based on Japanese Patent Application No. 2010-288529 filed Dec. 24, 2010, the contents of all of which are incorporated herein by reference in their entirety.

[0002] The present invention relates to a monitoring data analyzing apparatus, a monitoring data analyzing method, and a monitoring data analyzing program.

[0003] There is a technique for, in carrying out performance management for a system, regressively analyzing monitoring data in the past in the monitoring target system to calculate a regression model and predicting a performance value of the system using the regression model. For example, Patent Document 1 described below discloses a technique concerning a performance monitoring apparatus for a WWW site.

[0004] Patent Document 1: Patent Publication JP-A-2002-268922

[0005] Some systems have a plurality of patterns of use such as, for example, being used frequently for data retrieval such as browsing processing in the daytime and used frequently for data update such as batch processing at night. In such systems, when a utilization rate of a disk with respect to the number of operation requests for, for example, readout, write, and the like is predicted as a performance value, a regression model is generated using monitoring data in all the patterns of use and prediction is performed in the conventional performance managing method. However, since the utilization rate of the disk is substantially different depending on the patterns of use. Therefore, in the conventional method, an error that occurs between a prediction value based on the regression model and an actual value increases and prediction accuracy for a performance value is deteriorated.

[0006] The present invention has been devised to solve the problems and an object of the present invention is to provide a monitoring data analyzing apparatus, a monitoring data analyzing method, and a monitoring data analyzing program that can improve prediction accuracy for a performance value of a monitoring target system even if the performance value is substantially different for each of patterns of use.

[0007] A monitoring data analyzing apparatus according to the present invention includes a regression-model recalculating section configured to combine regression models, which are generated for each of a plurality of groups into which log data including monitoring data in a monitoring target system set as a target of performance management is classified, using the log data belonging to the groups corresponding to the regression models and test target log data, which is the log data set as a target of a performance test, to recalculate the regression models.

[0008] A monitoring data analyzing method according to the present invention includes the step of combining regression models, which are generated for each of a plurality of groups into which log data including monitoring data in a monitoring target system set as a target of performance management is classified, using the log data belonging to the groups corresponding to the regression models and test target log data, which is the log data set as a target of a performance test, to recalculate the regression models.

[0009] A monitoring data analyzing program according to the present invention causes a computer to execute the step included in the monitoring data analyzing method.

[0010] According to the present invention, it is possible to improve prediction accuracy for a performance value of a monitoring target system even if the performance value is substantially different for each of patterns of use.

[0023] Preferred embodiments of a monitoring data analyzing apparatus, a monitoring data analyzing method, and a monitoring data analyzing program according to the present invention are explained below with reference to the accompanying drawings. The monitoring data analyzing apparatus according to the present invention is an apparatus that analyzes monitoring data in a monitoring target system set as a target of performance management.

[0024] [First Embodiment]

[0025] First, the configuration of a monitoring data analyzing apparatus in a first embodiment is explained.

[0026] The monitoring data analyzing apparatus physically includes, for example, a CPU (Central Processing Unit), a storage device, and an input/output interface. The storage device includes components such as a ROM (Read Only Memory) and a HDD (Hard Disk Drive) that store programs and data processed by the CPU and a RAM (Random Access Memory) mainly used as various work areas for control processing. These components are connected to one another via a bus. The CPU executes the programs stored in the ROM and processes messages received via the input/output interface and data expanded in the RAM, whereby functions of sections in the monitoring data analyzing apparatus explained below can be realized.

[0027] As shown in FIG. 1, a monitoring data analyzing apparatus 1 functionally includes, for example, a characteristic-data-item extracting section 13, a regression-model recalculating section 14, and a performance-value calculating section 15. The monitoring data analyzing apparatus 1 includes, as files that the sections refer to, a log data file 21 in which log data is accumulated and a regression model file 22 in which regression models are stored. The log data is data in which monitoring data in a monitoring target system is recorded in time series. The regression models are generated by regressively analyzing respective log data classified into groups for each of patterns of use. For example, a data retrieval type (a readout type) and a data update type (a write type) correspond to the patterns of use. In this embodiment, the regression analysis is performed in advance and a plurality of regression models are stored in the regression model file 22.

[0028] A data configuration of the log data file 21 is explained with reference to FIG. 2. The log data file 21 includes, as data items, for example, a time segment item, a disk utilization rate item, the number of requests item, the number of times of write item, the number of times of readout item, and an average data length item. In the log data file 21, one record (hereinafter referred to as “log record”.) is generated for each of time segment items.

[0029] In the time segment item, segment information for identifying a period of time to which respective data of the log record belongs. In the disk utilization rate item, a utilization rate of a disk (e.g., a HDD) in the monitoring target system is stored. In this embodiment, a resource utilization rate of the monitoring target system is explained using a disk utilization rate. However, the resource utilization rate is not limited to the disk utilization rate. As the resource utilization rate, data indicating a status of use of components of the monitoring target system can be used. Specifically, besides the disk utilization rate, for example, a utilization rate of the CPU, a utilization rate of a memory (e.g., the RAM), a utilization rate of a network, and the like correspond to the resource utilization rate.

[0030] In the number of requests item, the number of requests received by the monitoring target system is stored. The number of requests item is not limited to the number of requests. For example, a ratio of arrival of requests may be used.

[0031] In the number of times of write item, the number of times data is written in the disk of the monitoring target system is stored. In the number of times of readout item, the number of times data is read out from the disk of the monitoring target system is stored. In the average data length item, an average value in a target period of time of data lengths of requests received in the target period of time is stored.

[0032] The characteristic-data-item extracting section 13 shown in FIG. 1 extracts, out of the data items included in the log data, data items related to the explanatory variable (the number of requests λ) of the regression model as characteristic data items.

[0033] The characteristic-data-item extracting section 13 calculates, for each of the groups classified during the regression analysis, concerning data items included in the log data forming the group, dependencies on the explanatory variable of the data items. The characteristic-data-item extracting section 13 extracts, as the characteristic data items, data items having the dependencies higher than a predetermined threshold.

[0034] The dependencies on the explanatory variable of the data items can be calculated, for example, as explained below.

[0035] As indicated by Expression (1) below, the characteristic-data-item extracting section 13 calculates, targeting all groups X classified during the regression analysis, for each of data items a_j of the log data, variances d_j of values obtained by dividing values x.a_j of the data item a_j by a value (the number of requests) x.λ of the explanatory variable of the regression model (hereinafter referred to as “variances d_j in the entire groups”.).

[0036] $\begin{array}{cc}\left[\mathrm{Math}.1\right]& \\ d_j=V\left(\frac{x·a_j}{x·\lambda }|x\in X\right)& \mathrm{Expression}\left(1\right)\end{array}$

[0037] As indicated by Expression (2) below, the characteristic-data-item extracting section 13 calculates, by targeting each of groups X_i classified during the regression analysis, for each of data items a_j of the log data belonging to the groups X_i, variances d_ji of values obtained by dividing the values x.a_j of the data item a_j by the value (the number of requests) x.λ of the explanatory variable of the regression model.

[0038] $\begin{array}{cc}\left[\mathrm{Math}.2\right]& \\ d_{\mathrm{ji}}=V\left(\frac{x·a_j}{x·\lambda }|x\in X_i\right)& \mathrm{Expression}\left(2\right)\end{array}$

[0039] As indicated by Expression (3) below, the characteristic-data-item extracting section 13 totalizes, in all the groups X, the variances d_ji of the data items a_j calculated for each of the groups X_i to calculate a total d_j′ in the entire groups of the variances d_ji in the groups X_i (hereinafter referred to as “sum d_j′ of the variances in the groups”.).

[0040] $\begin{array}{cc}\left[\mathrm{Math}.3\right]& \\ d_j^{\prime }=\sum _i{d}_{\mathrm{ji}}=\sum _{i}V\left(\frac{x·a_j}{x·\lambda }|x\in X_i\right)& \mathrm{Expression}\left(3\right)\end{array}$

[0041] When the value obtained by dividing the sum d_j′ of the variances in the groups by the variances d_j of the entire groups is smaller than a predetermined threshold k (d_j′/d_j<k), the characteristic-data-item extracting section 13 determines that the data items a_j satisfying the condition have high dependencies on the explanatory variable and extracts the data items a_j as the characteristic data items.

[0042] In FIG. 3, a diagram representing, as a bar graph, a value of (d_j′/d_j) per the data item a_j is shown. In FIG. 3, the data items a_j are arranged on the abscissa and a value of (d_j′/d_j) is shown on the ordinate. The threshold k is set to 0.3. In this case, data items having values of (d_j′/d_j) smaller than 0.3 are “the number of times of write item” and “the number of times of readout item”. Therefore, “the number of times of write item” and “the number of times of readout item” are extracted as the characteristic data items.

[0043] The regression-model recalculating section 14 shown in FIG. 1 combines a plurality of regression models stored in the regression model file 22, referring to the characteristic data items extracted by the characteristic-data-item extracting section 13 to calculate (recalculate) a regression model used in performing a performance test of the monitoring target system.

[0044] When the characteristic data items are known in advance, the regression-model recalculating section 14 can use the known data items instead of referring to the characteristic data items extracted by the characteristic-data-item extracting section 13. For example, it is known that, in a Web system, processing performance is substantially different the time when write in a database is present and the time when write in the database is absent. Therefore, in such a case, it is known in advance that “the number of times of write item” and “the number of times of readout item” are the characteristic data items.

[0045] Specifically, the regression-model recalculating section 14 calculates, for each of the groups, ratios of values of the characteristic data items with respect to the explanatory variable and combines the regression models of the groups using the calculated ratios to calculate regression models per characteristic data item, which are regression models concerning the characteristic data items. The regression-model recalculating section 14 combines the calculated regression models per characteristic data item according to appearance ratios of the characteristic data items included in target log data of a performance test in the monitoring target system (hereinafter referred to as “test target log data”) to calculate regression models.

[0046] The regression-model recalculating section 14 calculates the regression models per characteristic data item, for example, as explained below.

[0047] First, when the number of classified groups is represented as m and the number of extracted characteristic data items is represented as n, the regression-model recalculating section 14 generates a matrix A of m rows and n columns. In elements (i, j) of the matrix A, average values in the groups X_i of values obtained by dividing values of the characteristic data items a_j by the explanatory variable (the number of requests λ) are arrayed.

[0048] For example, as shown in FIG. 4, when the classified groups are two groups, i.e., “a group 1” and “a group 2” and the extracted characteristic data items are two data items, i.e., “the number of times of write item” and “the number of times of readout item”, the matrix A is a matrix of 2 rows×2 columns. In data fields shown in FIG. 4, average values in the groups of values obtained by dividing values of the characteristic items by the number of requests λ are stored. That is, the values of the data fields are the elements (i, j) of the matrix A. The matrix A in this case is represented as shown below.

[0049] $\left[\mathrm{Math}.4\right]$ $\begin{array}{cc}A=\left(\begin{array}{cc}0.87& 0.13\\ 0.20& 0.80\end{array}\right)& \end{array}$

[0050] Subsequently, the regression-model recalculating section 14 calculates a pseudo inverse matrix A⁺ of the matrix A. As a method of calculating the pseudo inverse matrix, for example, a method described in Reference Document 1 described below can be used.

[0051] [Reference Document 1]

[0052] D. A. Harvill, “Matrix Algebra From a Statistician's Perspective: Second Volume (2)”, Chapter 20 Moore-Penrose Inverse, Springer Japan KK

[0053] For example, the pseudo inverse matrix A⁺ of the matrix A illustrated above is represented as follows.

[0054] $\left[\mathrm{Math}.5\right]$ $\begin{array}{cc}{A}^{+}=\left(\begin{array}{cc}1.2& -0.2\\ -0.3& 1.3\end{array}\right)& \end{array}$

[0055] Subsequently, when a value of a certain characteristic data item a_j is set to “1” and a value of the other characteristic data items is set to “0”, the regression-model recalculating section 14 calculates regression models h_aj(λ) using Expression (4) below to calculate the regression models per characteristic data item. In Expression (4), f_i(λ) represents regression models of the groups Xi generated by the regression-model generating section 12.

[0056] $\begin{array}{cc}\left[\mathrm{Math}.6\right]& \\ \left(\begin{array}{c}⋮\\ h_{a_j}\left(\lambda \right)\\ ⋮\end{array}\right)=A^{+}\left(\begin{array}{c}⋮\\ f_i\left(\lambda \right)\\ ⋮\end{array}\right)& \mathrm{Expression}\left(4\right)\end{array}$

[0057] It is assumed that the regression model of “the group 1” shown in FIG. 4 is “f₁(λ)=0.0041λ+0.0951” and the regression model of “the group 2” shown in FIG. 7 is “f₂(λ)=0.0019λ+0.1188”. In this case, a regression model per characteristic data item of “the number of times of write item a₁” is calculated as “h_a1(λ)=1.2(0.0041λ+0.0951)−0.2(0.0019λ+0.1188)” using Expression (4) above. A regression model per characteristic data item of “the number of times of write item a₂” is calculated as “h_a2(λ)=−0.3(0.0041λ+0.0951)+1.3(0.0019λ+0.1188)” using Expression (4) above.

[0058] A procedure in recalculating regression models using the regression models per characteristic data item is explained below.

[0059] First, the regression-model recalculating section 14 calculates average values b_j in the test target log data of values obtained by dividing values of the characteristic data items a_j included in the test target log data by the explanatory variable (the number of requests λ).

[0060] Subsequently, the regression-model recalculating section 14 generates a regression model f(λ) using Expression (5) below. That is, the regression-model recalculating section 14 recalculates, using a plurality of existing regression models, regression models used in performing a performance test on the basis of the test target log data.

[0061] $\begin{array}{cc}\left[\mathrm{Math}.7\right]& \\ f\left(\lambda \right)=\left(\dots b_j\dots \right)\left(\begin{array}{c}⋮\\ h_{a_j}\left(\lambda \right)\\ ⋮\end{array}\right)& \mathrm{Expression}\left(5\right)\end{array}$

[0062] It is assumed that the test target log data is log data shown in FIG. 5. In this case, an average value b₁ of values obtained by dividing values of “the number of times of write item a₁”, which is the characteristic data item, by the number of requests λ is 90/100=0.9. An average value b₂ of values obtained by dividing values of “the number of times of readout item a₂” by the number of requests λ is 10/100=0.1.

[0063] Therefore, a regression model used in performing the performance test on the basis of the test target log data shown in FIG. 5 is calculated as “f(λ)=0.9h_a1(λ)+0.1h_a2(λ)” from Expression (5) above.

[0064] The performance-value calculating section 15 shown in FIG. 1 calculates the disk utilization rate ρcorresponding to the number of requests λ using the regression model f(λ) recalculated by the regression-model recalculating section 14. Consequently, it is possible to calculate a performance value based on the test target log data.

[0065] For example, when the recalculated regression model is “f(λ)=0.9h_a1(λ)+0.1h_a2(λ)”, the regression model per characteristic item of the number of times of write item a₁″ is “h_a1(λ)=1.2(0.0041λ+0.0951)−0.2(0.0019λ+0.1188)”, and the regression model per characteristic data item of the number of times of readout item a₂ is “h_a2(λ)=−0.3(0.0041λ+0.0951)+1.3(0.0019λ+0.1188)”, the disk utilization rate ρ at the time when the assumed number of requests λ is set to “400” is calculated.

[0066] In this case, the disk utilization rate ρis calculated as ρ=f(400)=0.9h_a1(400)+0.1h_a2(400)=1.77. That is, in this example, the disk utilization rate ρ exceeds 1. Therefore, in this case, an administrator takes measures for, for example, reinforcing disk resources.

[0067] The operation of the monitoring data analyzing apparatus in the first embodiment is explained. An operation in calculating a disk utilization rate using a regression model is explained with reference to FIG. 6.

[0068] First, the characteristic-data-item extracting section 13 executes characteristic data item extraction processing explained below (step S101). Subsequently, the regression-model recalculating section 14 executes regression model recalculation processing explained below (step S102). Subsequently, the performance-value calculating section 15 calculates a disk utilization rate corresponding to an assumed number of requests using a regression model f(λ) recalculated by the regression-model recalculating section 14 (step S103). The administrator can arbitrarily set the assumed number of requests.

[0069] The characteristic data item extraction processing executed in step S101 is explained with reference to FIG. 7.

[0070] Processing from steps S201 to S204 explained below is executed on all the data items a_j included in the log data while being sequentially looped for each of the data items a_j.

[0071] First, the characteristic-data-item extracting section 13 calculates, using Expression (1) above, targeting the log data of all the groups X classified during the regression analysis, the variances d_j of values obtained by dividing values of the target data items a_j by the number of requests λ (step S201).

[0072] Subsequently, the characteristic-data-item extracting section 13 calculates, using Expression (2) above, targeting the log data of the groups X_i, for each of the groups X_i, the variances d_ji, of values obtained by dividing the values of the target data items a_j by the number of requests λ. The characteristic-data-item extracting section 13 further calculates the sum d_j′ in the entire groups of the variances d_ji using Expression (3) above (step S202).

[0073] Subsequently, the characteristic-data-item extracting section 13 determines whether a value obtained by dividing the sum d_j′ calculated in step S202 by the variances d_j calculated in step S201 is smaller than the threshold k (step S203). When it is determined that the value is not smaller than the threshold k (NO in step S203), the characteristic-data-item extracting section 13 shifts the processing to the subsequent stage of step S204.

[0074] On the other hand, when it is determined in step S203 that the value of (d_j′/d_j) is smaller than the threshold k (YES in step S203), the characteristic-data-item extracting section 13 extracts the target data items a_j as the characteristic data items (step S204).

[0075] The regression model recalculation processing executed in step S102 is explained with reference to FIG. 8.

[0076] First, the regression-model recalculating section 14 sets, as elements of a matrix, average values in the groups X_i of values obtained by dividing values of the characteristic data items a_j extracted in step S101 by the explanatory variable (the number of requests λ) and generates the matrix A of the number of groups m rows×the number of characteristic data items n columns (step S301).

[0077] Subsequently, the regression-model recalculating section 14 calculates the pseudo inverse matrix A⁺ of the matrix A (step S302).

[0078] Subsequently, the regression-model recalculating section 14 calculates the regression models per characteristic data item h_aj(λ) using Expression (4) above (step S303).

[0079] Subsequently, the regression-model recalculating section 14 calculates average values b_j in the test target log data of values obtained by dividing values of the characteristic data items a_j included in the test target log data by the number of requests λ (step S304).

[0080] Subsequently, the regression-model recalculating section 14 substitutes the regression models per characteristic data item h_aj(λ) calculated in step S303 and the average values b_j calculated in step S304 in Expression (5) above and calculates the regression model f(λ) to generate (recalculate) the regression model f(λ) for the test target log data (step S305).

[0081] As explained above, according to the monitoring data analyzing apparatus according to the first embodiment, it is possible to combine a plurality of existing regression models and generate (recalculate) a new regression model according to ratios of values of characteristic data items included in the test target log data. Consequently, it is possible to generate a regression model adapted to an assumed pattern of use and predict a resource utilization rate such as a disk utilization rate using the regression model. Therefore, even if the resource utilization rate such as the disk utilization rate is substantially different for each of use patterns, it is possible to improve prediction accuracy for the resource utilization rate and take appropriate measures according to the resource utilization rate.

[0082] The characteristic-data-item extracting section 13 in the first embodiment extracts the characteristic data items using the dependencies on the explanatory variable of the data items. However, a method of extracting the characteristic data items is not limited to this. For example, an extracting method explained below can be used.

[0083] The characteristic-data-item extracting section 13 calculates, for each of the groups X_i classified during the regression analysis, concerning the data items a_j included in the log data forming the group, average values in the groups X_i of correlation coefficients between the data items a_j and the explanatory variable (the number of requests λ). The characteristic-data-item extracting section 13 extracts, as the characteristic data items, the data items a_j having the average values equal to or larger than a predetermined threshold k′.

[0084] The average values of the correlation coefficients between the data items a_j and the explanatory variable can be calculated by, for example, a procedure explained below.

[0085] First, the characteristic-data-item extracting section 13 calculates, using Expression (6) below, for each of the groups X_i, correlation coefficients r_ji between the data items a_j and the explanatory variable (the number of requests λ).

[0086] $\begin{array}{cc}\left[\mathrm{Math}.8\right]& \\ r_{\mathrm{ji}}=\frac{\sum _k\left(p^{\left(k\right)}·a_j-\stackrel{\_}{p·a_j}\right)\left(p^{\left(k\right)}·\lambda -\stackrel{\_}{p·\lambda }\right)}{\sqrt{\sum _k{\left(p^{\left(k\right)}·a_j-\stackrel{\_}{p·a_j}\right)}^2{\left(p^{\left(k\right)}·\lambda -\stackrel{\_}{p·\lambda }\right)}^2}}\stackrel{\_}{p·a_j}=\frac{1}{m}\sum _k{p}^{\left(k\right)}·a_j\stackrel{\_}{p·\lambda }=\frac{1}{m}\sum _k{p}^{\left(k\right)}·\lambda & \mathrm{Expression}\left(6\right)\end{array}$

[0087] In Expression (6), p^(k).a_j represents values of the data items a_j of kth log data belonging to the groups X_i, p^(k).λ represents values of the number of requests item of a kth log record belonging to the groups X_i, and m represents the number of groups.

[0088] Subsequently, the characteristic-data-item extracting section 13 calculates average values r_j of the correlation coefficients between the data items a_j and the explanatory variable by dividing a value obtained by totalizing the correlation coefficients r_ji for each of the groups X_i in all the groups X by the number of groups m.

[0089] Subsequently, the characteristic-data-item extracting section 13 extracts the data items a_j having |r_j| equal to or larger than the threshold k′ as the characteristic data items.

[0090] In the first embodiment explained above, the characteristic data items are extracted by the characteristic-data-item extracting section 13. However, a method of obtaining the characteristic data items is not limited to this. For example, when the characteristic data items are evident, the administrator may set the characteristic data items by, for example, registering the characteristic data items in a memory in advance. In this case, the characteristic-data-item extracting section 13 can be made unnecessary. When the characteristic-data-item extracting section 13 is made unnecessary, the regression-model recalculating section 14 only has to acquire values of the characteristic data items referring to the memory.

[0091] In the first embodiment explained above, the performance-value calculating section 15 is provided in the apparatus in which the characteristic-data-item extracting section 13 and the regression-model recalculating section 14 are provided. However, the performance-value calculating section 15 may be provided in an external apparatus.

[0092] [Second Embodiment]

[0093] A second embodiment of the present invention is explained. First, the configuration of a monitoring data analyzing apparatus in the second embodiment is explained.

[0094] As shown in FIG. 9, the monitoring data analyzing apparatus in the second embodiment is different from the monitoring data analyzing apparatus in the first embodiment in that the monitoring data analyzing apparatus in the second embodiment further includes a data classifying section 11, a regression-model generating section 12, and an abnormality determining section 16 in addition to the various functions of the monitoring data analyzing apparatus in the first embodiment (see FIG. 1). The other components are the same as the components of the monitoring data analyzing apparatus in the first embodiment. Therefore, the components are denoted by the same reference numerals and signs and explanation of the components is omitted. In the following explanation, differences from the first embodiment are mainly explained.

[0095] The data classifying section 11 classifies log data stored in the log data file 21 into groups for each of patterns of use (e.g., a data retrieval type and a data update type).

[0096] As a method of classifying log data into groups of each of patterns of use, for example, a publicly-known clustering method such as a shortest distance method, a longest distance method, a group average method, or a ward method can be used.

[0097] An example of a procedure in classifying log data into a plurality of groups using the clustering method is explained below. In this embodiment, as a characteristic in classifying log data into groups, a characteristic of a disk utilization rate ρ with respect to the number of requests λ is used.

[0098] First, log data is arranged on a coordinate plane shown in FIG. 10 in a unit of log record. The abscissa of FIG. 10 indicates the number of requests λ and the ordinate indicates the disk utilization rate ρ.

[0099] Subsequently, the data arranged on the coordinate plane is sequentially collated using the clustering method on the basis of the characteristic of the disk utilization rate ρ with respect to the number of requests λ. Finally, the data is classified into a first group G1 and a second group G2 shown in FIG. 10. In this case, log data belonging to a use pattern of a data update type is classified in the first group G1. Log data belonging to a use pattern of a data retrieval type is classified in the second group G2.

[0100] The regression-model generating section 12 shown in FIG. 9 executes a regression analysis of the log data for each of the classified groups and generates a regression model. As a method of generating the regression model, for example, a method described in Reference Document 2 described below can be used.

[0101] [Reference Document 2]

[0102] Hidekazu Nagahata, “Tahenryo-Kaiseki e no Suteppu” (“Step to a Multivariate Analysis”), Chapter 2 Regression Analysis, KYORITSU SHUPPAN CO., LTD.

[0103] Specifically, the regression-model generating section 12 performs the regression analysis, for example, using the disk utilization rate ρas an objective variable of the regression analysis and using the number of requests λ as an explanatory variable of the regression analysis. For example, when the log data shown in FIG. 10 is subjected to the regression analysis, a regression model represented by Expression (7) below is generated as a regression model M1 of the log data belonging to the first group G1. A regression model represented by Expression (8) below is generated as a regression model M2 of the log data belonging to the second group G2.
Regression model M1: f₁(λ)=ρ=0.0041λ+0.0951  Expression (7)
Regression model M2: f₂(λ)=ρ=0.0019λ+0.1188  Expression (8)

[0104] The regression-model generating section 12 stores the generated regression model M1 and the generated regression model M2 in a regression model file 22.

[0105] The abnormality determining section 16 shown in FIG. 9 determines, for each of log records of the test target log data, whether a difference (|f(λ′)−ρ′|) between a value f(λ′) obtained by substituting a value λ′ of the number of requests item in the regression model f(λ) and a value ρ′ of the disk utilization rate item is larger than a predetermined threshold. When (|f(λ′)−ρ′|) is larger than the predetermined threshold, the abnormality determining section 16 determines that an abnormality occurs in the monitoring target system.

[0106] Specifically, presence or absence of an abnormality in the monitoring target system can be determined as explained below.

[0107] Concerning an objective variable (the disk utilization rate ρ) of the regression model f(λ) recalculated by the regression-model recalculating section 14, the abnormality determining section 16 calculates variance V(ρ) and calculates a standard deviation (√V) of the variance. The abnormality determining section 16 calculates |f(λ′)−ρ′| and, when a value obtained by dividing this value by the standard deviation (√V) is larger than a threshold a, determines that an abnormality occurs in the monitoring target system. A value of the threshold a can be arbitrarily set. In other words, the abnormality determining section 16 determines presence or absence of an abnormality on the basis of a degree of a difference of the value of |f(λ′)−ρ′| from the standard deviation (√V).

[0108] The variance V(ρ) can be obtained by, for example, calculating Expression (10) below derived from Expression (9) below. The regression model f(λ) of Expression (9) below can be derived from Expression (4) and Expression (5) above. The variance V(ρ) of Expression (10) below can be derived from Expression (9) below.

[0109] $\begin{array}{cc}\left[\mathrm{Math}.9\right]& \\ f\left(\lambda \right)=\left(\dots b_j\dots \right)A^{+}\left(\begin{array}{c}⋮\\ f_i\left(\lambda \right)\\ ⋮\end{array}\right)=\sum _j{e}_j{f}_i\left(\lambda \right)& \mathrm{Expression}\left(9\right)\\ V\left(\rho \right)=\sum _i{e}_j^{2}V\left({\rho }_j\right)& \mathrm{Expression}\left(10\right)\end{array}$

[0110] V(ρ_j) of Expression (10) above represents variances concerning the objective variable (the disk utilization rate ρ) of the regression model calculated for each of the groups X_i. ρ_j represents probability variables indicating distributions of the objective variable in the groups X_i.

[0111] The operation of the monitoring data analyzing apparatus in the second embodiment is explained. An operation in calculating a disk utilization rate using a regression model is the same as the operation of the monitoring data analyzing apparatus in the first embodiment explained above (see FIGS. 6 to 8). Therefore, explanation of the operation is omitted.

[0112] First, an operation in classifying log data into a plurality of groups and generating a regression model for each of the groups is explained with reference to FIG. 11.

[0113] First, the data classifying section 11 clusters the log data stored in the log data file 21 and classifies the log data into a plurality of groups (step S401).

[0114] Subsequently, the regression-model generating section 12 executes a regression analysis of the log data for each of the classified groups and generates a regression model (step S402).

[0115] As explained below, the generated regression model can be used, for example, when a resource utilization rate such as a disk utilization rate is calculated 2 0 and when an abnormality of the monitoring target system is determined.

[0116] Abnormality determination processing for the monitoring target system is explained with reference to FIG. 12. The abnormality determination processing is executed after step S102 in FIG. 6 explained above is executed. The abnormality determination processing may be executed in parallel to step S103 in FIG. 6 or may be executed before or after step S103.

[0117] First, the abnormality determining section 16 calculates the standard deviation (√V) concerning the objective variable (the disk utilization rate ρ) referring to the recalculated regression model f(λ) recalculated by the regression-model recalculating section 14 (step S501).

[0118] Processing from steps S502 and S503 explained below is executed targeting all the log records included in the test target log data while being sequentially looped for each of the log records.

[0119] Subsequently, the abnormality determining section 16 calculates a difference between a value f(λ′) obtained by substituting the value λ′ of the number of requests item of the test target log data in the recalculated regression model f(λ) and a value ρ′ of the disk utilization rate item of the test target log data and calculates a value obtained by dividing the difference by the standard deviation (√V) calculated in step S501 (step S502).

[0120] Subsequently, the abnormality determining section 16 determines whether the value calculated in step S502 is larger than the threshold a (step S503). When it is determined that the value is not larger than the threshold a (NO in step S503), the abnormality determining section 16 shifts the processing to the subsequent stage.

[0121] On the other hand, when it is determined in step S503 that the value calculated in step S502 is larger than the threshold a (YES in step S503), the abnormality determining section 16 determines that an abnormality occurs in the monitoring target system and notifies the administrator of the abnormality (step S504). For example, a method of outputting a message indicating the abnormality, a method of outputting warning sound, and the like correspond to a method of notifying the administrator of the abnormality.

[0122] As explained above, according to the monitoring data analyzing apparatus in the second embodiment, in addition to the effects by the monitoring data analyzing apparatus in the first embodiment, it is possible to combine a plurality of existing regression models and generate (recalculate) a new regression model according to ratios of values of characteristic data items included in the test target log data.

[0123] Consequently, it is possible to generate a regression model adapted to an assumed pattern of use and predict a resource utilization rate such as a disk utilization rate using the regression model. Accordingly, even if the resource utilization rate such as the disk utilization rate is substantially different for each of use patterns, it is possible to improve prediction accuracy of the resource utilization rate and take appropriate measures according to the resource utilization rate.

[0124] Further, it is possible to find an abnormality of the monitoring target system on the basis of a difference between a regression model and the regression model adapted to the assumed pattern of use. Therefore, it is possible to improve accuracy in finding an abnormality.

[0125] [Modifications of Embodiments]

[0126] The embodiments are merely illustrations and do not exclude application of various modifications and techniques not clearly described in the embodiments. That is, the present invention can be modified and carried out in various forms without departing from the spirit of the present invention.

[0127] For example, in the embodiments, the present invention is carried out by one monitoring data analyzing apparatus. However, the present invention can be carried out by a plurality of apparatuses. In this case, the functions of the monitoring data analyzing apparatus in the embodiments only have to be distributed to the plurality of apparatuses to cause groups of the plurality of apparatuses to function in the same manner as the monitoring data analyzing apparatus in the embodiments.

[0128] A part or all of the embodiments can be described as indicated by notes below. However, the present invention is not limited to the below description.

[0129] (Note 1) A monitoring data analyzing apparatus comprising a regression-model recalculating section configured to combine regression models, which are generated for each of a plurality of groups into which log data including monitoring data in a monitoring target system set as a target of performance management is classified, using the log data belonging to the groups corresponding to the regression models and test target log data, which is the log data set as a target of a performance test, to recalculate the regression models.

[0130] (Note 2) The monitoring data analyzing apparatus according to note 1, further comprising: a data accumulating section configured to accumulate the log data; a data classifying section configured to classify the log data into the groups on the basis of characteristics of use status data included in the log data and indicating statuses of use of components of the monitoring target system; and a regression-model generating section configured to execute a regression analysis of the log data and generate a regression model for each of the groups.

[0131] (Note 3) The monitoring data analyzing apparatus according to note 1, further comprising a characteristic-data-item extracting section configured to extract, out of data items included in the log data, the data items related to an explanatory variable of the regression model as characteristic data items.

[0132] (Note 4) The monitoring data analyzing apparatus according to note 3, wherein the characteristic-data extracting section calculates, for each of the groups, dependencies on the explanatory variable in the data items of the data forming the group, and extracts data items having the calculated dependencies higher than a predetermined threshold as the characteristic data items.

[0133] (Note 5) The monitoring data analyzing apparatus according to note 3, wherein the characteristic-data extracting section calculates, for each of the groups, correlation coefficients between the data items of the data forming the group and the explanatory variable, and extracts data items having average values of the calculated correlation coefficients larger than a predetermined threshold as the characteristic data items.

[0134] (Note 6) The monitoring data analyzing apparatus according to any one of notes 3 to 5, wherein the regression-model recalculating section calculates, for each of the groups, ratios of values of the characteristic data items with respect to the explanatory variable and combines the regression models of the groups using the calculated ratios to calculate regression models per characteristic data item, which are the regression models concerning the characteristic data items, and combines the calculated regression models per characteristic data item according to appearance ratios of values of the characteristic data items included in the test target log data to recalculate the regression models.

[0135] (Note 7) The monitoring data analyzing apparatus according to any one of notes 1 to 6, further comprising a performance-value calculating section configured to calculate a performance value, which is an objective variable of the regression model, by using the regression model recalculated by the regression-model recalculating section.

[0136] (Note 8) The monitoring data analyzing apparatus according to any one of notes 1 to 6, further comprising an abnormality determining section configured to determine presence or absence of an abnormality in the monitoring target system on the basis of a difference between a value obtained by substituting a value of a data item corresponding to an explanatory variable of the regression model included in the test target log data in the regression model recalculated by the regression-model recalculating section, and a value of a data item corresponding to an objective variable of the regression model included in the test target log data.

[0137] (Note 9) A monitoring data analyzing method comprising the step of combining regression models, which are generated for each of a plurality of groups into which log data including monitoring data in a monitoring target system set as a target of performance management is classified, using the log data belonging to the groups corresponding to the regression models and test target log data, which is the log data set as a target of a performance test, to recalculate the regression models.

[0138] (Note 10) A monitoring data analyzing program for causing a computer to execute the step described in note 9.

[0139] The monitoring data analyzing apparatus, the monitoring data analyzing method, and the monitoring data analyzing program according to the present invention is suitable for improving prediction accuracy for a performance value of a monitoring target system even if the performance value is substantially different for each of patterns of use.

[0140] 1: monitoring data analyzing apparatus, 11: data classifying section, 12: regression-model generating section, 13: characteristic-data-item extracting section, 14: regression-model recalculating section, 15: performance-value calculating section, 16: abnormality determining section, 21: log data file, 22: regression model file