JPWO2013183425A1 - Performance prediction apparatus, performance prediction method, and program - Google Patents

Abstract

Performs performance prediction that reflects the actual behavior of the target system for performance prediction. The performance prediction apparatus clusters the input of the performance prediction target system as a target of performance prediction into a plurality of clusters in consideration of the type and interaction of the inputs, and the performance prediction target corresponding to each of the plurality of clusters Based on the performance prediction model of the system, the predicted output of the performance prediction model is calculated with respect to the input, and the plurality of the plurality of the output values so that the difference between the calculated predicted output and the actual output of the performance prediction target system is smaller. The performance prediction model of the performance prediction target system corresponding to each of the clusters is adjusted.

Description

  The present invention relates to a technique for predicting system performance by reflecting actual behavior of the system.

  Regarding system performance prediction technology, Patent Document 1 discloses a method for predicting performance indexes such as throughput, response, and resource usage of a parallel computer in a multitasking environment by converting the parallel computer system into a queuing model. Proposed. In the prediction based on this queuing model, parameters relating to the system to be constructed (for example, processing time (Demand) per request) are given in advance, and prediction is performed using the parameters. As a technique for adjusting this parameter, Patent Document 2 discloses that the log output of each of the application and the application model is compared, and the parameter of the application model is automatically adjusted and reflected. As a result, it is possible to select appropriate parameters according to the execution environment of the application and improve the accuracy of performance prediction.

  In Patent Document 3, the entire server computer system is made a black box, a measurement transaction is given, and the number of simultaneous processes is estimated by a simple establishment calculation different from the model based on the queuing network. Moreover, in patent document 4, in order to estimate the performance of a machine, each of several data groups is used as learning data of several models. Patent Document 5 discloses a technique related to a model construction method.

JP 2000-298593 A JP 2002-215423 A JP-A-10-187495 JP 2005-242803 A JP 2010-079325 A

  However, in the performance prediction disclosed in Patent Document 1, since parameter adjustment is not performed, the parameter value given in advance may be different from the parameter value of the actually constructed target system. Therefore, there is a problem that the performance index predicted using the model deviates from the performance index of the target system actually constructed.

  In the performance prediction disclosed in Patent Documents 2 and 3, parameters and the like are adjusted using measured data. However, when predicting the performance of a system in which multiple different types of requests are input, the ratio of each type of request in the input data to the measurement data used for adjustment, and each type of request in the input data in the situation to be predicted If the ratio is different, there is a problem that the accuracy of prediction is lowered.

  In Patent Document 4, a plurality of different types of driving environment data are classified into a plurality of data groups. However, the classification method simply divides each dimension according to a predetermined value. Therefore, even if this method is applied to the performance prediction of a system in which a plurality of different types of requests are input, the performance prediction considering the competition / interference between requests when these requests are executed simultaneously in the system. I can't do it.

  The present invention has been made to solve the above-described problems, and its purpose is to provide a performance prediction apparatus, a performance prediction method, and a program capable of performing system performance prediction reflecting the actual behavior of the system. Is to provide.

  In order to achieve the above object, the performance prediction apparatus according to the present invention classifies an input to a system subject to performance prediction into one of a plurality of clusters according to the type of the input and the interaction between the inputs. Clustering means, a predicted output calculating means for calculating a predicted output for the input using a performance prediction model of the system corresponding to the classified cluster, the calculated predicted output, and an actual output of the system Model adjustment means for adjusting parameters of the performance prediction model of the system corresponding to the classified clusters.

  The performance prediction method according to the present invention is a method implemented in an information processing apparatus having a control unit, wherein the control unit inputs an input to a system that is a target of performance prediction, between the type of the input and the input between the inputs. A clustering step of classifying the cluster into one of a plurality of clusters according to the interaction, and the control unit calculates a predicted output for the input using a performance prediction model of the system corresponding to the classified cluster A predicted output calculation step, and the control unit sets parameters of the system performance prediction model corresponding to the classified cluster such that a difference between the calculated predicted output and the actual output of the system is reduced. And a model adjustment step for adjustment.

  A program according to the present invention causes a computer to function as the performance prediction apparatus according to the present invention.

  ADVANTAGE OF THE INVENTION According to this invention, the performance prediction apparatus, the performance prediction method, and program which can perform the performance prediction of a system reflecting the actual behavior of a system can be provided.

It is a block diagram which shows the function structure of the performance prediction apparatus in one Embodiment. It is a figure which shows the structure of the database which has accumulate | stored the model in one Embodiment. It is a flowchart which shows an example of the process by the performance prediction apparatus in one Embodiment. It is a figure for demonstrating an example of the input data in one Embodiment. It is a figure for demonstrating an example of the input data in one Embodiment. It is a block diagram which shows the function structure of the performance prediction apparatus in other embodiment. It is a figure for demonstrating an example of the input data in other embodiment. It is a figure for demonstrating an example of the input data in other embodiment. It is a figure which shows the structure of the database which has accumulate | stored the model in other embodiment. It is a figure which shows the structure of the database which has accumulate | stored the model in other embodiment. It is a figure which shows the structure of the database which has accumulate | stored the model in other embodiment. It is a flowchart which shows an example of the process by the performance prediction apparatus in other embodiment.

  Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings. However, the constituent elements described in the following embodiments are merely examples, and are not intended to limit the technical scope of the present invention only to them.

[First Embodiment]
FIG. 1 is a block diagram illustrating a functional configuration example of the information processing apparatus 1 according to the first embodiment of the present invention. The information processing apparatus 1 functions as a performance prediction apparatus. The information processing apparatus 1 is an apparatus that operates according to a program such as a computer. The information processing apparatus 1 includes a model accumulation unit 101, a clustering unit 102, a performance prediction unit 103, and a model adjustment unit 104.

The model storage unit 101 is a database (hereinafter referred to as DB) that stores a model of the relationship between input and output of performance indexes of a system that is a target of performance prediction.
Here, the input is, for example, the number of requests that the system must process within a unit time. The output is, for example, system throughput or response time. However, the input and output are not limited to these, and if the model can be described as the relationship between the independent variable and the dependent variable, the independent variable may be the input and the dependent variable may be the output.

  Here, the model is a system performance prediction model, and determines the prediction of the output from the system by performing calculation or simulation according to a predetermined procedure with respect to the input to the system. . For example, a queue that imitates the operation of each component of a computer such as a CPU (Central Processing Unit) can be used as a model. In addition, a mechanism (for example, a neural network, a hidden Markov model, a polynomial function approximation, or the like) that can determine an appropriate output for an input by means such as learning or regression can be used as a model. These models are merely examples, and the models are not limited to these.

  The clustering unit 102 clusters the input to the system subject to performance prediction into one of a plurality of clusters in consideration of the input type and the interaction (relationship) between the inputs. Here, clustering is a process of classifying data such that data having a relationship with each other is in the same group (cluster).

  For example, when the input to the performance prediction target system includes a plurality of different types of requests, each cluster is given at a ratio for each type of request in the input (defined according to the ratio). For example, when typical ratios at a certain time such as weekdays, holidays, and the end of the month are known, the ratios at each time may be defined as clusters. In this case, it is determined to which cluster the ratio of input requests to the performance prediction target system at the time of prediction belongs.

  The performance prediction unit 103 refers to the model storage unit 101 to identify a model corresponding to the cluster to which the input to the system belongs, and uses the relationship between the input and output in the model to predict the input of the performance prediction target system. Calculate the output.

  The model adjustment unit 104 adjusts the parameters of the model based on the difference between the prediction output calculated by the performance prediction unit 103 and the actual output of the performance prediction target system for a model corresponding to a certain cluster. Here, the model parameters are, for example, a processing time per request in the case of a queue model, and a synaptic load in the case of a neural network. Hereinafter, adjusting a parameter included in the model is expressed as adjusting the model.

  The actual output value of the performance prediction target system can be obtained by any method. For example, the actual output value of the performance prediction target system may be acquired from a storage unit included in the information processing device 1 or may be acquired from an external device of the information processing device 1. The adjusted parameters are held by the performance prediction unit 103. However, the present invention is not limited to this, and the adjusted parameters may be held by the model storage unit 101 or other elements. The performance prediction unit 103 calculates a predicted output by a model using the adjusted parameters.

  Specifically, the algorithm for adjusting the model adjusts the model by using a method such as a learning or optimization method so that the predicted output of the calculated model approaches the actual output of the performance prediction target system. For example, when the model is a neural network, the synaptic load is corrected by a learning algorithm such as backpropagation. When the model is a polynomial approximation, the coefficient of each term is corrected by the least square method or the like. If the model is a queue, the request processing time is corrected by a Kalman filter or the like. In addition, as a function to evaluate the difference between the actual output of the performance prediction target system and the predicted output, the model parameters can be corrected using reinforcement learning, genetic algorithm, Monte Carlo method, linear programming, etc. The method for adjusting the model is not limited to these examples.

  In one embodiment, the information processing apparatus 1 includes a control unit, a storage unit, and an input / output unit as a hardware configuration. The control unit is configured by a CPU or the like. The storage unit includes a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disc Drive), and the like. The input / output unit includes various interfaces such as an operation unit, a display unit, and a communication interface. The functions provided in the information processing apparatus 1 described above are realized by the cooperation of these hardware and various programs stored in the storage unit.

  With reference to FIG. 2, the configuration of data stored in the model storage unit 101 will be described. The prediction target system ID 151 is an identifier for identifying a system that is a target of performance prediction. The cluster ID 152 is an identifier for identifying a cluster to which an input to the prediction target system belongs. The model 153 is a type of system performance prediction model. The model adjustment algorithm 154 is a type of algorithm used for adjusting the model. The model adjustment condition 155 is a condition for ending the model adjustment.

  The cluster ID 152 is stored in association with the prediction target system ID 151. Similarly, the model 153, the model adjustment algorithm 154, and the model adjustment condition 155 are stored in association with the black box ID 151 and the cluster ID 152.

  With respect to the model adjustment condition 155, for example, the model adjustment unit 104 compares the actual output of the performance prediction target system with the prediction output predicted by the model, and the difference is determined in advance (that is, the model adjustment condition 155). If the value is equal to or less than the stored value, the adjustment is completed. When the adjustment is completed in this way, the model can predict the behavior of the real system within a predetermined accuracy.

  As a modification, the difference between the actual output of the performance prediction target system and the predicted output predicted by the model does not change after a certain number of adjustments (ie, the adjustment can be considered to be in an equilibrium state). ) May be completed. Alternatively, the adjustment may be completed when the adjustment is performed a predetermined number of times.

  Note that the same model, model adjustment algorithm, and model adjustment condition may be applied to all clusters, as in the case where the prediction target system ID 151 in FIG. 2 is “001”. Alternatively, different models, model adjustment algorithms, and model adjustment conditions may be applied to each cluster as in the case where the prediction target system ID 151 is “002”.

  Next, an example of the flow of processing by the information processing apparatus 1 will be described with reference to FIG. The control of this process is performed by developing and executing a program stored in the storage unit by a control unit provided in the information processing apparatus 1.

  First, in step S201, the clustering unit 102 clusters (classifies) input data to the performance prediction target system into any cluster. This clustering is performed according to the type of input data to the performance prediction target system and the interaction between the input data.

  For example, request interaction depends on the ratio of request types in the input request (ie, requests that the system processes simultaneously). Therefore, clustering according to the type and interaction of the input data can be performed by clustering the input data according to the ratio of the type of input data in the input data to the performance prediction target system.

  FIG. 4A shows an example of input data to the performance prediction target system. Here, the input data includes two types of requests, and the number of each request per unit time is shown in a graph. The horizontal axis represents time, and the vertical axis represents the number of requests per unit time. In this example, it is assumed that clusters 1 and 2 are defined according to the ratio of two types of requests. Cluster 1 is a cluster in which the ratio of the number of requests 1 to the number of requests 2 is 1: 2. Cluster 2 is a cluster in which the ratio of the number of requests 1 to the number of requests 2 is 4: 3. Clustering is performed by determining to which cluster the ratio of requests included in the input data is close.

  4B shows a graph in which the input data shown in FIG. 4A is plotted with the number of requests 1 on the horizontal axis and the number of requests 2 on the vertical axis. In the graph of FIG. 4B, in addition to the input data, a straight line having a request ratio of 1: 2 and a straight line having a 4: 3 ratio are plotted. In FIG. 4B, the input data represented by x is clustered in cluster 1 and the input data represented by ◯ is clustered in cluster 2. Here, the index of the proximity between the input data and the cluster is noticed, for example, in the plot of the ratio of the input data and the ratio of each cluster (such as Euclidean distance) and / or the number of requests of a plurality of types as shown in FIG. 4B. The distance between the current data point and the cluster represented by a straight line can be used. In addition, it is possible to use the inner product of unit vectors representing the slopes of the data point of interest, the straight line passing through the origin, and the straight line representing the cluster. However, the index of the proximity between the input data and the cluster is not limited to these examples.

  4A and 4B exemplify the case of two types of requests and two types of clusters. However, it is obvious that the application can be easily performed even in the case of a plurality of types of requests and a plurality of types of clusters.

  Next, in step S202, the performance prediction unit 103 refers to the model storage unit 101, and uses the relationship between the input and output of the model corresponding to the cluster clustered in S201 to input to the performance prediction target system. The predicted output for is calculated.

  In step S <b> 203, the model adjustment unit 104 determines whether the model adjustment is completed based on the model adjustment condition 155 of the model storage unit 101. Here, whether or not the model adjustment is completed is determined based on the model adjustment condition 155 of FIG. 2 for each model corresponding to each cluster. If it is determined that the model adjustment has been completed, the process ends. If it is not determined that the adjustment has been completed, the process proceeds to step S204.

  In step S204, the model adjustment unit 104 adjusts the model based on the predicted output of the model calculated in step S202 and the actual output of the performance prediction target system. Here, the model adjustment is performed based on the model adjustment algorithm 154 of FIG. 2 for each model corresponding to each cluster. After adjusting the model in step S204, the process returns to step S203, and the performance prediction unit 103 calculates the prediction output predicted by the adjusted model again.

  Next, the effect of this embodiment will be described. In this embodiment, the input of the performance prediction target system is assigned to any of a plurality of clusters in consideration of the type of input and interaction (relationship) (for example, the ratio for each type of request included in the input). Clustering. Then, for the input of the performance prediction target system, the prediction output predicted by the model corresponding to the cluster is calculated, and the model is adjusted so that the difference between the actual output of the performance prediction target system and the prediction output is small. . As a result, according to the present embodiment, by adjusting the models corresponding to a plurality of clusters with respect to the performance prediction target system in which the performance varies due to the interaction of multiple types of inputs, Performance prediction that reflects actual behavior can be performed. For example, a plurality of clusters may be defined by the ratio of each request included in the input to the system. Request interaction depends on the ratio of requests (ie, requests that the system processes simultaneously). Therefore, adjusting the model for each cluster is a reflection of the actual interaction of requests.

[Second Embodiment]
FIG. 5 is a block diagram illustrating a functional configuration example of the information processing apparatus 3 according to the second embodiment of the present invention. Among the configurations of the information processing device 3, the same components as those of the information processing device 1 in the first embodiment are denoted by the same reference numerals. Here, only a part different from the information processing apparatus 1 in the configuration of the information processing apparatus 3 will be described. Description of other configurations and operations thereof is omitted. Note that the hardware configuration of the information processing apparatus 3 is the same as that of the first embodiment.

  The clustering unit 302 creates a plurality of clusters according to the type and interaction of inputs to the performance prediction target system, and clusters the inputs into any of the plurality of clusters. Here, creation of a cluster can be performed by, for example, applying multiple regression equations to a plurality of types of requests included in the input.

  FIG. 6 shows an example of input data to the performance prediction target system. In this example, two types of requests are included in the input data. FIG. 6 shows a graph in which the horizontal axis plots the number of requests 1 and the vertical axis plots the number of requests 2 for each unit time. In FIG. 6, in addition to the input data, two straight lines obtained by applying the input data to the regression equation are plotted. Each straight line (ie, regression equation) represents each cluster. Input data is clustered into one of two types of clusters. FIG. 6 illustrates the case of two types of requests and two clusters, but in general, a plurality of multiple regression equations may be applied to a plurality of types of requests. A method of applying a plurality of multiple regression equations may be, for example, the method described in Patent Document 5 or another method. Note that all of the multiple regression equations need not have the same dimension.

  FIG. 7 shows an example in which there are three types of requests. Each axis represents the number of unit times of each request. Assume that the numbers of requests 1, 2, and 3 are N1, N2, and N3, respectively. At this time, the cluster 1 is a two-dimensional planar cluster represented by a multiple regression equation of N3 = A * N1 + B * N2 (A and B are coefficients). Cluster 2 is a one-dimensional linear cluster represented by a multiple regression equation of N3 = C * N2 and N2 = D * N1 (C and D are coefficients). In this way, clusters of different dimensions may be created. The number of clusters to be created may be determined in advance, or an index (Akaike information criterion: AIC or Bayesian information criterion: BIC, etc.) that evaluates the goodness of the model from the degree of freedom and accuracy of the created regression equation ) May be calculated, and as many clusters as the number of the values may be created.

The model creation unit 305 creates a model corresponding to each cluster created by the clustering unit 302.
FIG. 8A shows the configuration of the DB storing the models at the time before creating the cluster. Since the cluster is not created, there is only one cluster (ie all inputs belong to the same cluster). After creating the cluster, the model creation unit 305 creates a model for each cluster, for example, by duplicating the model before cluster creation as shown in FIG. 8A. FIG. 8B shows a state in which three clusters are created for the prediction target system ID 151 of “001” and a corresponding model is created. As illustrated in FIG. 8B, the same model is created for all clusters having the prediction target system ID 151 of “001”.

  Alternatively, for example, the model creation unit 305 may create a model for each cluster by incorporating the properties of the clusters. FIG. 8C illustrates a case where a model is created by taking in the properties of each cluster after duplicating the model before cluster creation as shown in FIG. 8A. As illustrated in FIG. 8C, models for each cluster having the prediction target system ID 151 of “001” are different types of neural networks.

  As a method of incorporating the properties of the cluster, when the cluster is a multiple regression equation, the dimension can be reflected. For example, it is assumed that the number of input neurons of the neural network at the time of FIG. 8A is the number of request types, and the number of unit times of the corresponding type of request is input to each input neuron. Consider a case where such a neural network is duplicated and corresponds to each cluster. Considering the case of FIG. 7 as an example, since the cluster 1 is a two-dimensional planar cluster, two independent variables are sufficient to specify any one point in the cluster. Therefore, two new independent basis vectors in this cluster can be taken and used as input variables to the model. Therefore, the number of input neurons of the neural network corresponding to this cluster may be two, which is the number of independent variables, from three, which is the number of types of requests. Similarly, since the cluster 2 is a one-dimensional linear cluster, the number of input neurons of the neural network corresponding to this cluster may be one. Thus, the number of input neurons can be reduced by considering the dimension of the cluster. It will be obvious that such types of input variables can be reduced with general models, not limited to neural networks.

  Next, the flow of processing by the information processing apparatus 3 will be described. FIG. 9 is a flowchart showing processing of the information processing apparatus 3. The control of this process is performed by the control unit of the information processing device 3 developing and executing a program stored in the storage unit. In FIG. 9, the same steps as those in FIG. 3 of the first embodiment are denoted by the same reference numerals, and these steps are processed in the same manner as in the first embodiment. Omitted.

  In step S401, the clustering unit 302 creates a cluster according to the input to the performance prediction target system. In step S <b> 402, the model generation unit 305 creates a model corresponding to each of the clusters created by the clustering unit 302. In step S402, the created model and the accompanying model adjustment algorithm / model adjustment condition may be manually corrected while referring to the corresponding cluster.

  Next, effects of the second embodiment will be described. The information processing apparatus 3 as the performance prediction apparatus according to the present embodiment is different from the first embodiment in that a cluster is created from an input and a model is created for each cluster. According to the present embodiment, since the cluster is created from the input, even if the cluster is not generated when performing the performance prediction, the cluster can be generated according to the contents of the input, and the generated cluster Each model can correspond.

  In addition, by creating clusters using multiple regression equations, it is possible to adjust models that incorporate input interactions. For example, in the case of a one-dimensional linear cluster, the ratio of requests of each type is constant, and the model can be adjusted in such a case. In the case of a two-dimensional planar cluster (ie, N3 = A * N1 + B * N2), if request 1 occurs (1 / A) times, request 3 occurs once, or request 2 ( The average behavior of the request that the request 3 occurs once when it occurs 1 / B) times, and the model can be adjusted in such a case. The same applies to a three-dimensional or higher cluster.

  Furthermore, according to the present embodiment, a model corresponding to each cluster can be created by reflecting the cluster information. For example, reflecting the dimension of the cluster can reduce the dimension of the input variable as described above. In general, the smaller the dimension, the higher the learning speed.

  The specific configuration of the present invention is not limited to the above-described embodiment, and modifications within a range not departing from the gist of the present invention are included in the present invention.

  This application claims the priority on the basis of Japanese application Japanese Patent Application No. 2012-129164 for which it applied on June 6, 2012, and takes in those the indications of all here.

  While the present invention has been described with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  A part or all of the above embodiment can be described as in the following supplementary notes, but is not limited thereto.

(Appendix 1) Clustering means for classifying the input to the system subject to performance prediction into one of a plurality of clusters according to the type of the input and the interaction between the inputs,
Predicted output calculation means for calculating a predicted output for the input using a performance prediction model of the system corresponding to the classified cluster;
Model adjustment means for adjusting parameters of the performance prediction model of the system corresponding to the classified cluster so that a difference between the calculated predicted output and the actual output of the system is reduced. A characteristic performance prediction device.

  (Additional remark 2) The said clustering means produces the several cluster about the said input according to the kind of input to the system used as the object of performance prediction, and the interaction between inputs, The additional remark 1 characterized by the above-mentioned. Performance prediction device.

  (Additional remark 3) The performance prediction apparatus of Additional remark 1 or 2 provided with the model creation means which produces the performance prediction model of the said system corresponding to each of these several clusters.

  (Additional remark 4) The said model preparation means produces | generates the said performance prediction model reflecting the information of the said cluster, The performance prediction apparatus of Additional remark 3 characterized by the above-mentioned.

  (Additional remark 5) The said cluster is defined according to each ratio of the kind of request contained in the said input, The performance prediction apparatus as described in any one of Additional remark 1 to 4 characterized by the above-mentioned.

  (Supplementary note 6) The performance prediction apparatus according to any one of supplementary notes 1 to 4, wherein the cluster is defined by applying a multiple regression equation to the input.

(Appendix 7) A performance prediction method implemented in an information processing apparatus having a control unit,
A clustering step in which the control unit classifies the input to the system that is the target of performance prediction into one of a plurality of clusters according to the type of the input and the interaction between the inputs;
A predictive output calculating step in which the control unit calculates a predictive output for the input using a performance prediction model of the system corresponding to the classified cluster;
A model adjustment step in which the control unit adjusts parameters of the performance prediction model of the system corresponding to the classified cluster so that a difference between the calculated predicted output and an actual output of the system is reduced; A performance prediction method comprising:

  (Additional remark 8) The program for functioning a computer as a performance prediction apparatus as described in any one of Additional remark 1 to 7.

1, 3 Information processing apparatus 101 Model accumulation unit 102, 302 Clustering unit 103 Performance prediction unit 104 Model adjustment unit 305 Model generation unit

Claims (8)

Clustering means for classifying the input to the system subject to performance prediction into one of a plurality of clusters according to the type of the input and the interaction between the inputs;
Predicted output calculation means for calculating a predicted output for the input using a performance prediction model of the system corresponding to the classified cluster;
Model adjustment means for adjusting parameters of the performance prediction model of the system corresponding to the classified cluster so that a difference between the calculated predicted output and the actual output of the system is reduced. A characteristic performance prediction device.   2. The performance prediction according to claim 1, wherein the clustering unit creates a plurality of clusters for the input in accordance with a type of an input to a system to be a performance prediction target and an interaction between the inputs. apparatus.   The performance prediction apparatus according to claim 1, further comprising model creation means for creating a performance prediction model of the system corresponding to each of the plurality of clusters.   The performance prediction apparatus according to claim 3, wherein the model creation unit generates the performance prediction model by reflecting the cluster information.   The said cluster is defined according to each ratio of the kind of request contained in the said input, The performance prediction apparatus of any one of Claim 1 to 4 characterized by the above-mentioned.   5. The performance prediction apparatus according to claim 1, wherein the cluster is defined by applying a multiple regression equation to the input. 6. A performance prediction method implemented in an information processing apparatus having a control unit,
A clustering step in which the control unit classifies the input to the system that is the target of performance prediction into one of a plurality of clusters according to the type of the input and the interaction between the inputs;
A predictive output calculating step in which the control unit calculates a predictive output for the input using a performance prediction model of the system corresponding to the classified cluster;
A model adjustment step in which the control unit adjusts parameters of the performance prediction model of the system corresponding to the classified cluster so that a difference between the calculated predicted output and an actual output of the system is reduced; A performance prediction method comprising:   The program for functioning a computer as a performance prediction apparatus of any one of Claim 1 to 6.

Images