Patent Application
20140188446
The present invention relates to a technology conducting performance prediction which reflects an actual behavior (action) of a target system.
In the above technical field, Patent Literature 1 (Japanese Patent Laid Open No. 2000-298593) discloses a method for predicting performance metrics, such as a throughput, a response, a resource usage rate, of a parallel computer in a multi-task environment by changing a parallel computer system into a queue model. In the prediction by the queue model, a parameter related to a system to be configured (e.g. processing time per one request (Demand)) is given in advance, and the performance metrics are predicted by using the parameter.
As a technology adjusting the parameter, Patent Literature 2 discloses a method for comparing an application with a log output of an application model corresponding to the application. In the method described in Patent Literature 2, a parameter of the application model is automatically adjusted and the adjusted parameter is reflected to the application model (hereinafter, referred to as “model” in some cases). In Patent Literature 2, the method is proposed, that improves accuracy of the performance prediction of the application by choosing an appropriate parameter corresponding to an execution environment of the application.
A device described in Patent Literature 3 changes a whole server computer system into a black box, provides a measurement transaction, and predicts the number of simultaneous processing by a simple established calculation which is different from a model based on a queue network.
[Patent Literature 1]
[Patent Literature 2]
[Patent Literature 3]
In Patent Literature 1, the value given in advance may differ from a value in an actually configured target system (i.e. system to be targeted for performance prediction). The method disclosed in Patent Literature 1, therefore, includes the problem that a difference occurs between the performance metrics predicted by using the application model and performance metrics of the actually configured target system.
The method described in Patent Literature 2 (Japanese Patent Laid Open No. 2002-215423) aims at simulating a target system and adjusts the parameter of the model which is configured based on known information. Therefore, the technology described in Patent Literature 2 only discloses an adjustment technique for a parameter in the configured model.
Patent Literature 3 (Japanese Patent Laid Open No. 1998(H10)-187495) discloses a technology in which the whole target system is substituted with the black box and evaluation data on the model is predicted. In Patent Literature 3, a targeted model for evaluation is a simulation model of the target system which is configured based on the known information, like Patent Literature 2. Therefore, the device described in Patent Literature 3 cannot solve the problem that the difference occurs between the performance index predicted by using the model and the performance index of the target system actually configured.
A main object of the invention is to provide a technology solving the problem described above.
In order to solve the above problem, the information processing device of the invention includes, I/O measurement means for measuring input and output of a performance prediction target system that is a subject of performance prediction, adjustment part substitution means for, with respect to a system model of the performance prediction target system that is configured from a plurality of partial system models, substituting a designated partial system model with a black box that is connected to input and output of the partial system model, predicted output calculation means for, on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box by the adjustment part substitution means, calculating predicted output of the system model for the input measured by the I/O measurement means, and model adjustment means for adjusting a relation between the input and the output in the black box such that a difference between the output of the performance prediction target system measured by the I/O measurement means and the predicted output of the system model calculated by the predicted output calculation means is made smaller.
In order to solve the above problem, a system performance prediction method of the invention that is conducted by the information processing device, includes measuring input and output of a performance prediction target system that is a subject of performance prediction; for a system model of the performance prediction target system that is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model; on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box, in the substitution, calculating predicted output of the system model for the input measured for the performance prediction target system; and adjusting a relation between input and output in the black box such that a difference between the output measured for the performance prediction target system and the calculated predicted output of the system model is made smaller.
In order to solve the above problem, a control program (computer program) of the invention causing a computer to execute the functions of a I/O measurement function for measuring input and output of a performance prediction target system which is a subject of performance prediction; an adjustment part substitution function, for a system model of the performance prediction target system which is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model; a predicted output calculation function, on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box by the adjustment part substitution function, calculating predicted output of the system model for the input measured by the I/O measurement function; and a model adjustment function for adjusting a relation between the input and the output in the black box such that a difference between the output of the performance prediction target system measured by the I/O measurement function and the predicted output of the system model calculated by the predicted output calculation function is made smaller.
In order to solve the above problem, a system performance prediction method of the invention that is conducted by the information processing device, includes measuring input and output of a performance prediction target system which is a subject of performance prediction; for a system model of the performance prediction target system which is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model; on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box, in the substitution, calculating predicted output of the system model for the input measured for the performance prediction target system; and adjusting a relation between input and output in the black box such that a difference between the output measured for the performance prediction target system and the calculated predicted output of the system model is made smaller; and setting the predicted output measured on the system model as a result of performance prediction on the performance prediction target system, if the difference between the output measured for the performance prediction target system and the calculated predicted output of the system model becomes the smallest on the basis of the adjustment on the black box.
Further, the object is achieved not only by the method, the device, and the computer program having the above configuration, but by a computer-readable recording medium storing the computer program.
In the invention, performance prediction which reflects an actual behavior of a target system for the performance prediction can be conducted.
Referring to drawings, exemplary embodiments of the invention are described below in detail as examples. Constituent elements described in the exemplary embodiments are only examples, and the technical scope of the invention is not limited to those.
An information processing device 100 of a first exemplary embodiment of the invention is described by referring
As shown in
The I/O measurement unit 101 measures input and output of a performance prediction target system 110. The performance prediction target system 110 is a subject whose performance is predicted by using the information processing device 100. A system model 105 shown in
With respect to the system model 105, the adjustment part substitution unit 102 substitutes a designated partial system model 105a with a black box 106a connected to input and output of the partial system model 105a. A substitution model 106 shown in
The prediction output calculation unit 103 calculates, based on the substitution model 106, a predicted output 103a from the substitution model 106 with respect to an input 101a (i.e. measured input) measured by the I/O measurement unit 101.
The model adjustment unit 104 adjusts a relation between input and output in the black box 106a such that a difference between an output 101b of the performance prediction target system 110 measured by the I/O measurement means 101 (i.e. measured output) and the predicted output 103a from the substitution model 106 calculated by the predicted output calculation unit 103 is made smaller.
In the exemplary embodiment, the performance prediction which reflects an actual behavior of a target system for the performance prediction can be conducted.
Next, an information processing device of a second exemplary embodiment of the invention based on the above first exemplary embodiment is described. The information processing device of the exemplary embodiment substitutes a partial system model, among partial system models included a system model, designated through an operator's instruction, with a black box. The information processing device of the exemplary embodiment inputs a measured input of the performance prediction target system into a system model including the black box, and monitors a predicted output thereof. The information processing device of the exemplary embodiment adjusts the black box such that a difference between the measured output of the performance prediction target system and the predicted output becomes small.
In the exemplary embodiment, even though a fluctuation of performance occurs due to change of environment in the system model depending on the operator's instruction, the information processing device can substitute the designated partial system model with an appropriate black box in response to the fluctuation based on an operator's instruction. The information processing device of the exemplary embodiment can promptly conduct performance prediction reflecting an actual behavior of a target system for performance prediction
In
The performance prediction target system 210 is a subject for performance prediction in the exemplary embodiment. The performance prediction target system 210 is configured by one or more devices, e.g. one or more computers, which work in accordance with a program. When a plurality of devices are used, communication between the devices may be performed through the network 220 or through a communication cable directly connected. The network 220 may be Internet or a LAN (Local Area Network). The network 220 may take any configuration which enables communication between the information processing device 200 and the performance prediction target system 210.
The information processing device 200 includes a communication control unit 201, an I/O measurement unit 202, a performance prediction unit 203, a model adjustment unit 204, a model accumulation unit 205, an adjustment part substitution unit 206, and a substitution part reception unit 207.
The model accumulation unit 205 is a database (hereinafter, referred to as “DB”) which accumulates a system model representing a relation between input and output of a performance index of the performance prediction target system 210. The input is e.g. the number of requests which a system has to handle in a unit of time. The output is e.g. throughput of the system or a response time. However, the input and the output are not limited to those. If the input and the output can be described by a model as a relation between an independent variable and a dependent variable, the independent variable may be employed as the input and the dependent variable may be employed as the output.
The communication control unit 201 communicates with the performance prediction target system 210 through the network 220. The I/O measurement unit 202 has a capability to access the performance prediction target system 210 through the communication control unit 201 and measure input and output of the performance prediction target system 210.
The substitution part reception unit 207 receives a designation of a partial system model to be substituted with a black box through an operator's operation. The substitution part reception unit 207 may choose the partial system model to be substituted with the black box from an input device, like a keyboard, in accordance with the operator's operation. The substitution part reception unit 207 may acquire the partial system model to be substituted with the black box from a recording medium, like a hard disc drive (HDD) in a computer. The substitution part reception unit 207 may acquire the partial system model to be substituted with the black box from a server through a communication network, like Internet. The substitution part reception unit 207 substitutes the partial system model acquired by the substitution part reception unit 207 with the black box.
The black box in the exemplary embodiment is a mechanism which enables determination of an appropriate output for input based on learning or regression. For example, the mechanism may be achieved by a neural network or a hidden Markov model. The mechanism may be achieved by approximation by a polynomial function or a non-parametric regression function.
By using a relation between input and output of a system model in which the designated partial system model is substituted with the black box by the adjustment part substitution unit 206, the performance prediction unit 203 calculates a predicted output predicted by the system model with respect to the input of the performance prediction target system 210 measured by the I/O measurement unit 202.
The model adjustment unit 204 adjusts the black box based on the input and the output of the performance prediction target system 210 measured by the I/O measurement unit 202 and the predicted output predicted by the system model including the black box calculated by the performance prediction unit 203.
An example of the system model and an example of substituting the partial system model with the black box in the exemplary embodiment are described below.
In the example shown in
The module receives one or more inputs, conducts specified processing, and determines one or more outputs. The specified processing may be represented by an equation, e.g. y=exp (u) (u is input and y is output), or by a queue. The module just has to determine an output which is calculated or simulated, according to the predetermined procedure with respect to the received input, and not limited to the example described above.
In the example shown in
When the system model is graphically displayed, like in
In the system model 320 shown in
In following descriptions and drawings, the application server 340 may be described as “AP server”, the DB server 350 may be described as “DB server”, and the disc (storage device) may be described as “DK”.
The Web server 330 includes a CPU 331, two DKs 332, 333, and a queue. The AP server 340 includes a CPU 341, two DKs 342, 343, and a queue. The DB server 350 includes a CPU 351, two DKs 352, 353, and a queue.
By using system models 302, 303 after substitution shown in
In
A RAM 440 is a random access memory which the CPU 410 uses as a work area for a temporary storage. The RAM 440 includes a data storage area required to realize the exemplary embodiment. A numeral number 441 denotes input data (measured input) transmitted from the performance prediction target system 210 (hereinafter, referred to as “real system” in some cases). A numeral number 442 denotes output data (measured output) transmitted from the real system.
A numeral number 443 denotes model prediction output data outputted from the system model. A numeral number 444 denotes an output data difference which is a difference between the output data 442 transmitted from the real system and the model prediction output data 443 outputted from the system model. A numeral number 445 denotes substitution instruction data indicating a partial system model which is instructed by an operator and substituted with a black box.
A numeral number 446 denotes a system model of the performance prediction target system. A numeral number 447 denotes a system model in which a partial model is substituted with a black box based on the substitution instruction data 445. A numeral number 448 denotes a black box which is used when a partial model is substituted.
A storage 450 stores database, parameters, or following data or programs required for achievement of the exemplary embodiment. A numeral number 451 denotes a system model DB configuring the model accumulation unit 205 (refer to
The storage 450 stores following programs. A numeral number 455 denotes an information processing program executing whole of processes. A numeral number 456 denotes an I/O measurement module measuring input-output of a real system. A numeral number 457 denotes, based on a system model in which a partial system model is substituted with a black box, an adjustment part control module which controls the black box, in the information processing program 455.
An input interface 460 mediates an operator's operation and data input. The input interface 460 connects to, for example, a key board 461, a mouse (registered trade mark) 462 and a recording medium 463. An output interface 470 mediates outputs of an operation instruction to an operator and processing results. The output interface 470 connects to, for example, a display unit 471 and a printer 472.
In
A system model ID 481 is an identifier identifying a system model. The system model DB 451, with respect to the system model ID 481, makes an association (connection) between a target model 482 targeted by the system model, an attribute 483 including the feature thereof, and input/output 484 indicating input and output of the system model, and stores them. Further the system model DB 451 includes a real system model 485.
The black box DB 452 stores a black box type 492 which is associated with a black box ID 491 which is an identifier identifying a black box. The model adjustment algorism 453 and the model adjustment condition 454 which are associated with the black box ID 491 are stored therein.
If the black box is the neural network, the model adjustment algorism 453 my choose each synapse weight as a parameter to be adjusted, and determine an initial value and an adjustment step for the parameter. If the black box is polynomial approximation, the model adjustment algorism 453 may choose each coefficient, as a parameter to be adjusted, and determine an initial value, an adjustment order, and an adjustment step for the parameter. The model adjustment algorism 453 may give a random value, as the initial value for the parameter. Also, the model adjustment algorism 453 may preliminarily give a value simulating a behavior of the partial system model before substitution, as the initial value for the parameter.
In
The model adjustment unit 204 may compare an output measured in the performance prediction target system 210 with a predicted output predicted by a system model, and may determine that adjustment is completed if the error (difference) is equal to or less than predetermined accuracy. In the exemplary embodiment, when the model adjustment condition 454 is satisfied, the model adjustment unit 204 determines that a system model including a black box is able to predict a behavior (action) of the real system within a predetermined accuracy and completes adjustment.
As timing to determine adjustment completion in the exemplary embodiment, following cases are exemplified;
The information processing device of the exemplary embodiment may store procedures of changing into substitution of other partial system model, changing a black box type, or the like, as measures which are carried out when predetermined conditions are not satisfied after the given number of adjustments are conducted and the adjustment becomes ineffective.
The information processing device of the exemplary embodiment exemplified in
The CPU 410 accesses the performance prediction target system 210 through the network 220 and measures input and output of the performance prediction target system 210 (step S501).
The CPU 410 acquires, in accordance with an operator's input operation, information of a partial system model which is substituted with a black box and adjusted, and is included in a system model stored in the system model DB 451 (step S503). The CPU 410 may acquire information of a partial system model which is substituted with a black box and adjusted, in accordance with instructions from a server through a communication network, e.g. Internet.
The CPU 410 substitutes the partial system model designated as an adjustment subject with a black box (step S505).
Suppose that when the module 313 in the system model 310 shown in
Next, by using a relation between input and output of the system model in which the partial system model is substituted with a black box, the CPU 410 calculates a predicted output predicted by the model with respect to the input (measured input value) which is acquired from the performance prediction target system 210 in step S501 (step S507).
Next, the CPU 410 determines whether or not adjustment of the black box is completed, based on the model adjustment condition 454 (step S509). When determining that the adjustment of the black box is completed (YES in step S509), the CPU 410 completes processing.
If the adjustment of the black box is not completed (NO in step S509), the CPU 410 carries out step S511. The model adjustment condition 454 for determination of adjustment completion of the black box is shown in
The CPU 410 adjusts the black box in accordance with the model adjustment algorism 453, based on the input and the output (measurement results) acquired from the performance prediction target system 210 in step S501 and based on the predicted output of the system model calculated in step S507 (step S511). After adjusting the black box in step S511, the CPU 410 calculates again a predicted output predicted by the model including the adjusted black box, in step S507.
In the exemplary embodiment, the CPU 410 adjusts the black box by using a method, e.g. learning or approximation such that a difference between the predicted output of the system model by calculation and the output (measured output value) acquired from the performance prediction target system 210 is made smaller. By using, as an evaluation function, the difference between the output (measured output value) acquired from the performance prediction target system 210 and the predicted output of the system model by calculation, the CPU 410 corrects a parameter of the black box by conducting reinforcement learning, genetic algorism, or Monte Carlo algorism. For example, if the black box is the neural network, the CPU 410 corrects each synapse weight. If the black box is the polynomial approximation, the CPU 410 corrects each coefficient.
In the exemplary embodiment, even though a partial element of a system model corresponding to input-output of a black box (input-output of the module 312 in the example of
If partial elements of a system model corresponding to input-output of the black box can be directly measured, supervised learning, like back propagation, or a least-square method may be used in order to bring a predicted output of the black box close to a measurement result of an output of the partial elements.
An information processing device of a third exemplary embodiment of the invention based on the above mentioned the first and the second exemplary embodiments is described below. The information processing device of the exemplary embodiment differs from the second exemplary embodiment in that a partial system model to be substituted with the black box is designate not from the outside, like an operator, but in its own device. In the exemplary embodiment, the information processing device preferentially substitutes a part of a system which is influenced from outer environment or a part of system including a lot of rounding in a partial system model, with the black box, depending on a type of the system model. In the exemplary embodiment, since operator's operation is simplified, the information processing device can quickly conduct performance prediction reflecting an actual behavior of a target system.
Regarding
A model adjustment part designation unit 607 sends information designating a partial system model to the adjustment part substitution unit 206 in order to preferentially substitute the part of a system which is influenced from outer environment or the part of a system including a lot of rounding in a partial system model, with the black box, depending on a type of the system model.
An information processing device of a fourth exemplary embodiment of the invention based on the above mentioned the first and the second exemplary embodiments is described below. The information processing device of the exemplary embodiment differs from the second exemplary embodiment in that validity of a system model is evaluated after adjustment of the black box is completed. When the system model in which adjustment of a black box is completed is inappropriate as the model of the real system, the information processing device of the exemplary embodiment can avoid performance prediction by the inappropriate system model.
Regarding
A model validity evaluation unit 708 evaluates whether or not it is appropriate to use the system model for performance prediction for the performance prediction target system 210, with respect to the system model after adjustment of a black box. If the model validity evaluation unit 708 determines that use of the model is appropriate as the result of the evaluation, the information processing device 700 uses the system model after the adjustment for performance prediction for the performance prediction target system 210. On the other hand, if the model validity evaluation unit 708 determines that use of the model is inappropriate as the result of the evaluation, the information processing device 700 informs an operator of the determination result.
The model validity evaluation unit 708 evaluates whether or not the model after the adjustment of the black box is appropriate, in a following way. For example, when a part of the designated module is substituted with a black box, the black box is adjusted under influence of the designated module. Since a black box is strongly influenced from the module including the black box, and adjusted, when the module to be really corrected is designated, variance (discrepancy) of a behavior (action) between the model and the actual system is confined within the module. In this case, the influence is not extended to the outside of the module. Therefore the black box in the case includes relatively simple configuration.
If the module which should not be really corrected is designated, variance of a behavior between the model and the actual system appears outside the module. In this case, since the variance is intended to be adjusted by the black box included in the module which should not be corrected, the black box has a complicated configuration.
The model validity evaluation unit 708 may determine the more simple a configuration of the black box is, the more appropriate the model after adjustment is. That the configuration of the black box is simple means, for example, that the number of terms is small if the black box is the polynomial function, and that the number of neurons or synapses is small in case of the neural network. This is applicable to a case in which a module configuring a system is substituted with a black box. Even in a situation in which the module is substituted with a whole of the system, or even in a situation in which a part of the module is substituted with a module, as mentioned above, simplicity of black box configuration is an evaluation criterion for a system model including the adjusted black box.
In order to evaluate validity of the adjusted model, such like Akaike Information Criterion (AIC) or Bayesian Information Criterion (BIC), which are general index for evaluation of a model quality, may be adopted as evaluation criteria.
The evaluation criterion for evaluation of model validity is not limited to the above examples. Any evaluation criterion may be adopted which is appropriate for an evaluation criterion of a system model including a black box.
An information processing device of a fifth exemplary embodiment of the invention based on the above mentioned the first and the second exemplary embodiments is described below.
The information processing device of the exemplary embodiment differs from the information processing unit of the above mentioned exemplary embodiment in that a partial system model is consequently substituted with a black box, and the system model is adjusted, then a system model which is appropriately substituted in the adjusted system models is evaluated, and a result of the evaluation is sent to an operator.
In the exemplary embodiment, even when it is not preliminarily understood which partial model has to be corrected, the information processing device of the exemplary embodiment presents an appropriately adjusted model. According to the information processing device of the exemplary embodiment, a model which can perform performance prediction reflecting an actual behavior of a system is obtained. Further, according to the exemplary embodiment, when a part of a partial system model is substituted with a black box, the black box is adjusted while strongly receiving influence of the partial system model. Depending on whether or not variance of a behavior between a model and an actual system is confined within the partial model, validity of the adjusted model is widely changed. By using the characteristics, the reason on whether or not adjustment of the designated partial model is appropriate is given.
Regarding
A model adjustment part designation unit 807 includes substitution order data 807a storing an order of partial system models to be substituted. The model adjustment part designation unit 807 designates a partial system model or a part thereof, which is substituted with a black box, to the adjustment part substitution unit 206, according to the order (information representing order) stored in the substitution order data 807a.
The model validity evaluation unit 708 determines, like the fourth exemplary embodiment, whether or not it is appropriate that respective system models in which black boxes are consecutively substituted and adjusted are used for performance prediction by the performance prediction target system 210.
As a result of validity evaluation by the model validity evaluation unit 708, an adjusted model presentation unit 809 presents a system model which exceeds the criterion of validity. As presentation method for the system model, for example, the adjusted model presentation unit 809 may display a list of system models which exceed the criterion of validity together with evaluation values of the validity. As another method for presenting the system model, for example, the adjusted model presentation unit 809 may present the system having evaluation of the highest validity. As presentation method for the system model, for example, the adjusted model presentation unit 809 may display on a screen, or record in a storage device (recording medium), e.g. a hard disc drive.
Following various pieces of information are associated and stored in the first configuration 807a-1 in a substitution order 901 as shown in
substitution part 902: information representing a part of a system to be substituted with a black box,
black box type 903: information representing a type of a black box,
data at adjustment completion 904: information representing a state of a black box whose adjustment is completed,
validity evaluation 905: information representing an valuation result of validity.
The left side of
In
A RAM 1140 is a random access memory which the CPU 1110 uses as a work area for temporary storage. The RAM 1140 includes an area which stores data required for achievement of the exemplary embodiment. In
A substitution partial system model 1145 and validity evaluation 1449 are temporarily stored inside the RAM 1140 in
A storage 1150 stores database, various parameters, or following data or programs required for achievement of the exemplary embodiment. The same data as that of
In
The system model DB 1151 of the exemplary embodiment stores a black box substitution order 1186 corresponding to a system model, in addition to data (481 to 485) of the system model DB 451 in
In the exemplary embodiment, after measuring input-output in step S 501, the CPU 1110 designates a partial system model in order (step S1203). The order of the partial system models designated by the CPU 1110 does not necessarily have to include all the partial system models. In an example shown in
In the exemplary embodiment, processing related to the step S1203 is not limited to the examples described above. In the exemplary embodiment, any method that is able to uniquely identify an order of modules may be adopted as processing method related to the step S1203 In step S1203, when a plurality of parts of a system are adjusted, the CPU 1110 may determine the order based on combinations. When it is preliminarily understood that there is no need to adjust the module 311, the CPU 1110 chooses combinations from the other modules 312 to 316.
As partially shown in
Next, the CPU 1110 substitutes a portion of the partial system model of the designated model with a black box (step S1205). As an example, when designating the module 313 in the system model 310 shown in
In the exemplary embodiment, the part of the partial system is not limited to the example described above. In the exemplary embodiment, the part of the partial system model may be “a part” which does not widely depart from a requirement of the module, i.e. “simulation of a partial element in a system”, by substitution with a black box.
The module 313, a part of which is substituted with a black box, receives an input from the module 311, and sends an output to the module 315, as with a time before the substitution. If a plurality of parts to be adjusted exist, a part of each module is substituted with a black box.
The CPU 1110 conducts processing similar to the exemplary embodiment in step S507, step S509, and step S511.
The CPU 1110 evaluates validity of the system model in which a black box is adjusted (step S1213). In the exemplary embodiment, the evaluation criterion explained in the fourth exemplary embodiment is used as an evaluation criterion for validity evaluation of the system model. Explanations on the criterion are omitted in the exemplary embodiment.
As a result of determination in step S1215, the CPU 1110 returns to step S1205 and repeats processing if a part to be next adjusted on a system model exist, and carries out step S1217 if the part to be next adjusted does not exist.
The CPU 1110 presents an appropriately adjusted model from among respective adjusted system models evaluated in step S1213 (step S1217). In the step, the CPU 1110 may present only system model which is evaluated as the most appropriate. The CPU 1110 may present the system model with evaluation values (the number of terms, a value of AIC, or the like) in order of validity.
An information processing system of a sixth exemplary embodiment of the invention is described below. In each exemplary embodiment described above, a configuration is described, in which a system targeted for performance prediction and an information processing device conducting the performance prediction of the system are included.
The information processing system of the exemplary embodiment differs from the above exemplary embodiment in a configuration in which a performance prediction system having a plurality of servers and carrying out performance prediction is included, and a performance prediction target system having a plurality of servers is included. In the exemplary embodiment, performance prediction for a system including a plurality of servers connected to a network can be conducted by a system including a plurality of servers that cooperates with each other. A configuration of each function and operations of the exemplary embodiment may adopt same or similar configurations of the above exemplary embodiment, and therefore detailed descriptions are omitted. In the exemplary embodiment, a configuration of the information processing system is described.
A performance prediction target system 1320 includes a Web server 1321, an AP server 1322, and a DB server each connecting to a network 1350. A performance prediction system 1310 include a performance prediction server 1311, a system model DB server 1312, and a system model execution server 1313 each connecting to a network 1350. The performance prediction server 1311 carries out substitution and evaluation of a partial system model of a black box. The system model DB server 1312 manages a system model DB. The system model execution server 1313 executes simulation based on a system model.
In the information processing system 1300 shown in
With respect to a system targeted for performance prediction, the invention is applicable to performance prediction reflecting an actual behavior (action) of the system. For example, when the invention is applied to an information processing system, accurate performance prediction can be achieved even though the system targeted for performance prediction includes a behavior which is unknown without actual operations.
In the information processing device of the invention, not only whole of system, but a module which is a part of the system can be substituted with a black box. Therefore, the information processing device of the invention can improve accuracy of the performance prediction for a model of a system similar to the above described system.
The information processing device of the invention can make a model of a system similar to the above described system by keeping a part which is substituted with a black box same as the above described system, and changing a parameter of a module (which is not substituted with a black box).
In this case, since a black box already adjusted based on an actual behavior of the system is included, therefore, it is expected that accuracy of performance prediction for the model of the system similar to the above described system may be correspondingly improved.
By the method for substituting the whole model of a system (310 in
Generally, it is not able to configure the model of the similar system if the whole model is substituted with a black box, since it is unclear how parameters of the information processing system (e.g. driving frequency of CPU, or the like) are reflected to the black box. Contrarily, in the information processing device of the invention, if only module which is a part of a system is substituted with a black box, parameters of the information processing system can be reflected to a module other than the substituted black box. Therefore, in the information processing device of the invention, a model of a similar system can be configured.
The exemplary embodiment of the invention is described above in detail. A system or devices in which separate characteristics included in each exemplary embodiment are combined in any manner are within the scope of the invention.
The invention may be applied to a system including a plurality of devices, or a single device. The invention may be applied to the case in which a control program which achieves functions of the exemplary embodiments is given to a system or a device directly or from remote places. Therefore, a control program which is installed in a computer in order to achieve the functions of the invention by the computer, a medium storing the control program, WWW (World Wide Web) server from which the control program can be downloaded are within the scope of the invention.
A part or all of the exemplary embodiments can be described as following supplemental notes, but are not limited to the following descriptions.
An information processing device, including:
I/O measurement means for measuring input and output of a performance prediction target system that is a subject of performance prediction;
adjustment part substitution means for, with respect to a system model of the performance prediction target system that is configured from a plurality of partial system models, substituting a designated partial system model with a black box that is connected to input and output of the designated partial system model;
predicted output calculation means for, on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box by the adjustment part substitution means, calculating predicted output of the system model for the input measured by the I/O measurement means; and
model adjustment means for adjusting a relation between the input and the output in the black box such that a difference between the output of the performance prediction target system measured by the I/O measurement means and the predicted output of the system model calculated by the predicted output calculation means is made smaller.
The information processing device of the supplemental note 1, further including,
a reception means for receiving input from a user who designates the partial system model substituted by the adjustment part substitution means.
The information processing device of supplemental note 1 or supplemental note 2, further including,
designation means for designating the partial system model substituted by the adjustment part substitution means.
The information processing device of any one of supplemental note 1 to supplemental note 3, wherein the part substitution means substitutes a part of the designated partial system model with the black box.
The information processing device of any one of supplemental note 1 to supplemental note 4, further including,
evaluation means for evaluating validity of the system model of the performance prediction target system after the model adjustment means adjusts the black box that is substituted with the partial system model.
The information processing device of any one of supplemental note 1 to supplemental note 5, wherein
the black box includes a configuration which is able to determine an appropriate output with respect to an input by at least one of learning and regression, and is one of a neural network, a hidden Markov model, a polynomial function, and a non-parametric regression function.
The information processing device of supplemental note 5 or supplemental note 6, wherein
the evaluation means determines validity of the system model of the performance prediction target system is higher, if the black box after adjustment by the model adjustment means includes a simpler configuration, if Akaike Information Criterion (AIC) is smaller, or if Bayesian Information Criterion (BIC) is lower.
The information processing device of any one of supplemental note 5 to supplemental note 7, wherein
the designation means designates the partial system model included in the system model in a predetermined order, and the evaluation means evaluates validity of the system model of the performance prediction target system after the model adjustment means adjusts the black box that is substituted with each partial system model,
the information processing device further including,
presentation means for presenting to a user the system model of the performance prediction target system after adjustment by the model adjustment means, the system model being evaluated as highly appropriate by the evaluation means.
The information processing device of any one of supplemental note 1 to supplemental note 8, wherein
the adjustment part substitution means substitutes a plurality of the partial system models with the one or more black boxes.
The information processing device of any one of supplemental note 1 to supplemental note 9, further including,
model accumulation means for accumulating the system model of the performance prediction target system on the basis of a plurality of the partial system models that compose the system model.
A system performance prediction method that is conducted by an information processing device, including,
measuring input and output of a performance prediction target system that is a subject of performance prediction;
for a system model of the performance prediction target system which is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model;
on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box, in the substitution, calculating predicted output of the system model for the input measured for the performance prediction target system; and
adjusting a relation between input and output in the black box such that a difference between the output measured for the performance prediction target system and the calculated predicted output of the system model is made smaller.
A control program causing a computer to execute the functions of:
an I/O measurement function for measuring input and output of a performance prediction target system that is a subject of performance prediction;
an adjustment part substitution function, for a system model of the performance prediction target system that is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model;
a predicted output calculation function, on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box by the adjustment part substitution function, calculating predicted output of the system model for the input measured by the I/O measurement function; and
a model adjustment function for adjusting a relation between the input and the output in the black box such that a difference between the output of the performance prediction target system measured by the I/O measurement function and the predicted output of the system model calculated by the predicted output calculation function is made smaller.
A performance prediction method that is conducted by an information processing device, including,
measuring input and output of a performance prediction target system that is a subject of performance prediction; for a system model of the performance prediction target system that is configured from a plurality of partial system models, substituting a designated partial system model with a black box which is connected to input and output of the partial system model;
on the basis of the system model of the performance prediction target system in which the designated partial system model is substituted with the black box, in the substitution, calculating predicted output of the system model for the input measured for the performance prediction target system;
adjusting a relation between input and output in the black box such that a difference between the output measured for the performance prediction target system and the calculated predicted output of the system model is made smaller; and
setting the predicted output measured on the system model as a result of performance prediction on the performance prediction target system, if the difference between the output measured for the performance prediction target system and the calculated predicted output of the system model becomes the smallest on the basis of the adjustment on the black box.
While having described an invention of the present application referring to the embodiments, the invention of the present application is not limited to the above mentioned embodiments. It is to be understood that to the configurations and details of the invention of the present application, various changes can be made within the scope of the invention of the present application by those skilled in the art.
This application claims priority from Japanese Patent Application No. 2011-134566 filed on Jun. 16, 2011, the contents of which are incorporation herein by reference in their entirety.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2011-134566 | Jun 2011 | JP | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/JP2012/065931 | 6/15/2012 | WO | 00 | 12/16/2013 |
