JP6486080B2 - Distributed processing system, distributed processing method, and distributed processing program - Google Patents

Description

  One aspect of the present invention relates to a distributed processing system, a distributed processing method, and a distributed processing program that execute data processing.

  As a technique for distributing processing in a computer system among a plurality of computers, a technique for distributing processing among a plurality of servers has been often performed. With regard to this distributed processing, Patent Document 1 below describes a network computing system that dynamically sets the execution location of an external program for processing according to the current operating status of each computer. This system transfers the program held by the first computer to the second computer, processes the predetermined data by the second computer, and takes the time required for the first procedure to obtain the result. The time that is determined to be short is compared with the time required for the second procedure for transferring predetermined data from the second computer, processing by the first computer, and transferring the result to the second computer. Process the data according to the procedure.

JP 2002-99521 A

  The system described in Patent Document 1 finally causes any one computer to process data, but it may be more efficient to distribute similar data processing to a plurality of computers. A mechanism for distributing processing among a plurality of servers is common, but a mechanism for distributing similar processing between a client terminal that outputs a final result of an operation and a server is not known. A mechanism for efficiently distributing the load of the entire system by distributing similar data processing between the client terminal and the server with improved performance is desired.

  A distributed processing system according to an aspect of the present invention is a distributed processing system including a client terminal and a server, and acquires an acquisition unit that acquires client information indicating an environment or an operation result of the client terminal from the client terminal, and an acquisition unit A determination unit that estimates the processing capability of the client terminal by executing machine learning using the client information acquired by the step, and determines a processing ratio between the client terminal and the server based on the estimated processing capability. The processing ratio is a ratio of executing data processing using unprocessed data, and an allocating section that allocates unprocessed data to the client terminal and the server based on the processing ratio determined by the determining section; , Processed data obtained by data processing at the client terminal and data at the server The binding between the obtained processed data by sense, and an output unit for outputting the client terminal.

  A distributed processing method according to an aspect of the present invention is a distributed processing method executed by a distributed processing system including a client terminal and a server, and acquires client information indicating an environment or an operation result of the client terminal from the client terminal. And the processing capability of the client terminal is estimated by executing machine learning using the client information acquired in the acquisition step, and the processing ratio between the client terminal and the server is calculated based on the estimated processing capability. A determination step for determining, wherein the processing rate is a rate at which data processing using unprocessed data is executed, and based on the determination rate and the processing rate determined in the determination step, An allocation step for assigning process data and a client terminal That the processed data obtained by the data processing, the bond between the obtained processed data by the data processing in the server, and an output step of outputting from the client terminal.

  A distributed processing program according to one aspect of the present invention is a distributed processing program that causes a computer system including at least two computers to execute a distributed processing method executed by a distributed processing system including a client terminal and a server. The distributed processing method obtains client information indicating an environment or an operation result of the client terminal from the client terminal, and performs machine learning using the client information obtained in the obtaining step, thereby increasing the processing capability of the client terminal. Estimating and determining a processing rate between the client terminal and the server based on the estimated processing capability, wherein the processing rate is a rate of executing data processing using unprocessed data, The decision step and the decision step Based on the processing ratio, the allocation step of allocating unprocessed data to the client terminal and the server, the combination of the processed data obtained by the data processing at the client terminal and the processed data obtained by the data processing at the server, An output step of outputting from the client terminal.

  In such an aspect, the processing capability of the client terminal is estimated by machine learning based on the environment or operation result of the client terminal, and the processing ratio between the client terminal and the server is determined based on the estimation result. Therefore, the same data processing can be distributed between the client terminal and the server in consideration of the environment or operation result of the client terminal. As a result, the entire system load can be efficiently distributed.

  According to one aspect of the present invention, it is possible to efficiently distribute the load of the entire system by distributing similar data processing between the client terminal and the server.

It is a block diagram which shows the function structure of the distributed processing system which concerns on embodiment. It is a figure which shows the hardware constitutions of the computer used with the distributed processing system which concerns on embodiment. It is a flowchart which shows the process in the determination part shown in FIG. It is a figure which shows the example (a) and (b) of visualization of the data after interpolation. It is a sequence diagram which shows operation | movement of the distributed processing system which concerns on embodiment. It is a sequence diagram which shows operation | movement of the distributed processing system which concerns on embodiment. It is a figure which shows the structure of the distributed processing program which concerns on embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are denoted by the same reference numerals, and redundant description is omitted.

  In the present embodiment, the present invention is applied to distributed processing of data interpolation. Data interpolation is a process for obtaining a numerical value that fills each section of a data string based on a known numerical data string, and is used for various purposes such as spatial interpolation for augmented reality. Since this interpolation processing may take time depending on factors such as the execution environment of the computer and the amount of data, it may be preferable to execute the data interpolation by distributed processing using a plurality of computers.

  The function and configuration of the distributed processing system 1 according to the embodiment will be described with reference to FIGS. The distributed processing system 1 according to the present embodiment interpolates measurement values related to the environment (also referred to as “environmental values” in the present specification) obtained from a plurality of sensors arranged indoors or outdoors. It is a computer system that estimates a typical environment and visualizes the estimation result by computer graphics. Since the sensors are discretely arranged in the space, the distributed processing system 1 performs data interpolation in order to comprehensively display the environmental values in the space. The distributed processing system 1 can provide the user with augmented reality by showing the estimation result intuitively and easily for the user. Examples of environmental values include temperature, humidity, and intensity of radio waves received by communication devices, but the types of measurement values are not limited to these. In this connection, the type and specific configuration of the sensor are not limited. Note that the process of interpolating discrete values in space is also referred to as “spatial interpolation”, but the present invention is not limited to spatial interpolation but can be applied to other data processes. In this embodiment, the term “data interpolation” is used.

  As shown in FIG. 1, the distributed processing system 1 includes at least a server 10 and one or more client terminals 20. The feature of this distributed processing system 1 is that data interpolation, which tends to increase the calculation load, is distributed between the server 10 and the client terminal 20 instead of being distributed among a plurality of servers. . The distributed processing system 1 can access a database 30 that stores measurement values obtained from a plurality of sensors S arranged in a space R that is an environment measurement target. The database 30 and the plurality of sensors S may be prepared as a part of the distributed processing system 1 or may exist in a system different from the distributed processing system 1.

  The server 10 is a computer capable of data communication with the database 30 and each client terminal 20 through a communication line such as the Internet or an intranet. The server 10 may be constructed by a single computer or a plurality of computers. When the server 10 is composed of a plurality of computers, each functional element of the server 10 is distributedly processed by these computers. However, as described above, the feature of the present embodiment lies in the distributed processing between the client and the server, and not in the distributed processing between a plurality of computers constituting the server 10. The type of computer that plays the role of the server 10 is not limited. For example, a stationary or portable personal computer (PC) may be used, a computer having higher performance than the PC, or a workstation may be used. It may be used.

  The client terminal 20 is a computer that displays the final processing result of the distributed processing system 1, that is, the estimation result of the spatial environment, and may be said to be a computer that provides the user with augmented reality based on the estimation result. it can. In the present embodiment, the client terminal 20 includes a web browser, and the estimation result is visually displayed on the web browser. The user can intuitively grasp the state of the space environment via the client terminal 20. The type of computer used as the client terminal 20 is not limited. For example, a stationary or portable PC may be used, a high-function mobile phone (smart phone), a mobile phone, a personal digital assistant (PDA), a tablet, or the like. The portable terminal may be used, or a wearable terminal such as a head mounted display (HMD) may be used. In general, the processing capability of the client terminal 20 (performance of hardware resources (CPU, memory, etc.) mounted on the client terminal 20) is lower than that of the server 10.

  A general hardware configuration of each computer 100 used as the server 10 or the client terminal 20 is shown in FIG. A computer 100 includes a CPU (processor) 101 that executes an operating system, application programs, and the like, a main storage unit 102 that includes a ROM and a RAM, an auxiliary storage unit 103 that includes a hard disk and a flash memory, and a network. The communication control unit 104 includes a card or a wireless communication module, an input device 105 such as a keyboard, a mouse, and a touch pad, and an output device 106 such as a display and a touch panel. Of course, the hardware modules to be mounted differ depending on the type of the computer 100. For example, a stationary PC and a workstation often include a keyboard, a mouse, and a monitor as an input device and an output device, but a touch panel often functions as an input device and an output device in a smartphone and a tablet.

  Each functional element of the server 10 and the client terminal 20, which will be described later, reads predetermined software on the CPU 101 or the main storage unit 102, and controls the communication control unit 104, the input device 105, the output device 106, and the like under the control of the CPU 101. The operation is realized by reading and writing data in the main storage unit 102 or the auxiliary storage unit 103. Data or a database necessary for processing is stored in the main storage unit 102 or the auxiliary storage unit 103.

  The database 30 is a device or functional element that stores a plurality of measured values measured by a plurality of sensors S at predetermined time intervals (for example, every 10 minutes or every hour). For example, the database 30 stores the measurement values of the plurality of sensors S at each time of 8 o'clock, 9 o'clock, 10 o'clock,. The implementation method of the database 30 is not limited, and may be a device different from the server 10 as in the present embodiment, or may be provided on an auxiliary storage device in the server 10. The database 30 may be a relational database or a CSV file.

  The method for accumulating the measured values in the database 30 is not limited. For example, the sensor S may directly communicate with the database 30 and store the measurement value in the database 30, or another computer may collect the measurement value from the sensor S and store the value in the database 30. The measured values stored in the database 30 are uninterpolated data used as input values for data interpolation. Uninterpolated data is raw data that has not yet undergone data interpolation.

  Since the characteristic of the distributed processing system 1 is the processing method of data stored in the database 30, the following description is based on the assumption that uninterpolated data to be processed already exists in the database 30.

  Next, the functional configuration of the server 10 will be described. As illustrated in FIG. 1, the server 10 includes a page providing unit 11, a client information acquisition unit (acquisition unit) 12, a determination unit 13, an allocation unit 14, and an interpolation unit 15 as functional components.

  The page providing unit 11 is a functional element that provides the client terminal 20 with a web page for visually displaying the state of the environment in the space. When the web page request (hereinafter referred to as “page request”) is received from the client terminal 20, the page providing unit 11 transmits a script of the designated web page (hereinafter referred to as “page data”) to the client terminal 20. To do. In this embodiment, the web page is described by, for example, HTML (HyperText Markup Language) and Java Script (trademark or registered trademark), but the language for describing the web page is not limited to these.

  In the page data, a script or program for acquiring client information indicating the environment of the client terminal 20 or the operation result is embedded. The distributed processing system 1 performs distributed processing of data interpolation between the server 10 and the client terminal 20, but in order to realize this, the processing capability of the client terminal 20 must be grasped, and thus client information is acquired. Need to do. The “client terminal environment” in this specification refers to an operating environment that can affect the processing efficiency of the client terminal, and is, for example, an operating environment related to hardware, software, or a communication network. The “operation result of the client terminal” in the present specification is the processing capability of the client terminal measured when the client terminal executes data interpolation. For example, the time required for data interpolation (processing time) or hardware For example, consumption of hardware resources. In this specification, the client information includes client environment information indicating the environment of the client terminal and client result information indicating the operation result of the client terminal.

  In this embodiment, the client environment information indicates an operating system (OS) and a web browser installed in the client terminal 20 and a communication speed between the client terminal 20 and the server 10. However, the contents of the client environment information are not limited to these.

  The OS and web browser information of the client terminal 20 can be acquired from the web browser itself executed on the client terminal 20. Specifically, a web browser can be specified by analyzing user agent information, and an OS can be specified by analyzing information on a platform on which the web browser is operating. The client terminal 20 can be roughly classified from these two pieces of information to determine approximate performance. For example, it can be determined whether the client terminal 20 is a desktop PC or a portable terminal. The pseudo code of the client_environment_acquire function for acquiring OS and web browser information is shown below.

function client_environment_acquire () {
Get the user agent (window.navigator.userAgent) Analyze the user agent to determine the web browser, store the result of the determination in a variable Get the browser platform (window.navigator.platform) Analyze the browser platform and run the OS Judgment and storing the determination result in a variable Returns an array showing the determination result of the OS and web browser
}

  The communication speed between the client terminal 20 and the server 10 is obtained by obtaining the difference between the data transmission time and the reception time as the communication time and dividing the size of the transmitted data by the communication time. Pseudo code of a client side function (client_network_speedcheck) and a server side function (server_network_speedcheck) used to determine the communication speed is shown below. The client_network_speedcheck function is embedded in the page data and provided to the client terminal 20. On the other hand, the server_network_speedcheck function is executed by the server 10 without being embedded in the page data. Since the accuracy of the time that can be measured is different for each OS, the measurement accuracy is matched between the client terminal 20 and the server 10 before obtaining the communication time. In the present embodiment, the distributed processing system 1 measures the communication speed every time the client terminal 20 and the server 10 communicate with each other, but the timing for measuring the communication speed is not limited to this. The communication speed may be measured at time intervals.

function client_network_speedcheck () {
Store client current time in variable ct1 Request server current server time
if (There was a normal response from the server)
Matching the measurement accuracy of server time and client time Upstream speed = Transmission data size / (Server current time-ct1)
Re-acquire the current time of the client, and store that time in the variable ct2. Downstream speed = received data size / (ct2-server current time)
else
Show errors
end if
}

function server_network_speedcheck () {
Measure the current server time Match the measured time format to the client Return the current server time
}

  In the present embodiment, the client performance information includes data interpolation processing time (client interpolation time) in which uninterpolated data assigned to the client terminal 20 is input, and the client terminal 20 requests the server 10 for environmental value data. And the total processing time until all the interpolated data are obtained. However, the content of the client performance information is not limited to these.

  The pseudo code of the interpolate_benchmark function for obtaining the processing time of data interpolation executed at the client terminal is shown below. This function acquires the current time before and after the data interpolation, and sets the difference between these times as the processing time (client interpolation time).

function interpolate_benchmark (uninterpolated data) {
Stores the current time of the client in time1 Executes data interpolation using uninterpolated data as input Stores the current time of the client in time2 Processing time = time2-time1
Return the interpolated data and calculation time obtained
}

  The interpolate_benchmark function has not only data interpolation for visualization, but also a benchmark test for the processing. Thus, by using actual data interpolation as a benchmark test, it is not necessary to execute a benchmark test separately from the data interpolation for visualization.

  The client information acquisition unit 12 is a functional element that acquires client information. The client information acquisition unit 12 receives client environment information or client performance information from the client terminal 20 and outputs the information to the determination unit 13.

  The determination unit 13 is a functional element that determines a processing ratio between the client terminal 20 and the server 10. The “processing ratio” in this specification is a ratio at which data interpolation is performed between the client terminal 20 and the server 10, and it can be said that this is a ratio of uninterpolated data to be processed. If the entire data interpolation or the entire uninterpolated data is 1, and the processing rates of the client terminal 20 and the server 10 are 0.3 and 0.7, respectively, the client terminal 20 performs data interpolation for 30% of uninterpolated data. And the server 10 performs data interpolation on the remaining 70% of the uninterpolated data. The determination unit 13 outputs the determined processing ratio to the allocation unit 14.

  The determination unit 13 estimates the processing capability of the client terminal 20 by executing machine learning, and determines the processing ratio based on the estimated processing capability. “Machine learning” in this specification is a process of predicting an unknown value by learning training data that is a set of known values.

  First, the determination unit 13 estimates the processing capability of the client terminal 20 by executing machine learning using the client environment information as an input parameter and the client result information obtained by the previous data interpolation as correct data. And the determination part 13 determines which group the client terminal 20 belongs to among the several groups preset according to the processing capability based on the estimation result. For example, assuming that the processing capability is represented by processing time, there are a group Ga whose processing time is less than 20 ms (milliseconds), a group Gb whose processing time is 20 ms or more and less than 50 ms, and a group Gc whose processing time is 50 ms or more. Suppose that it is set. In this case, the determination unit 13 determines that the client terminal 20 belongs to the group Gb when the processing capability (processing time) of the client terminal 20 is estimated to be 40 ms by machine learning. Subsequently, the determination unit 13 determines the next processing ratio according to the flowchart of FIG. 3 prepared for each group.

  Thereafter, the determination unit 13 estimates the processing capacity again by executing machine learning using the client performance information obtained by the next data interpolation executed according to the processing ratio as the next correct answer data, and the client terminal 20 belongs. Determine the group to perform. Then, the determination unit 13 further determines the next processing ratio according to the flowchart of FIG. 3 corresponding to the group. By repeating such processing, the processing ratio is gradually optimized. However, since there is no client result information in the first machine learning, the determination unit 13 executes machine learning using the client environment information as an input parameter without using correct data. Since it is sufficient to acquire the client environment information first, the determination unit 13 may reuse the acquired client environment information in the second and subsequent machine learning.

  With reference to FIG. 3, the process for determining the processing ratio in a certain group will be described in detail. The method for determining the processing rate varies depending on the progress of optimization within one group. Roughly speaking, (a) until a certain amount of client performance information is accumulated as a sample (until the initial sample group is completed), (b) after obtaining the initial sample group, the optimal solution is obtained. Until the determination, (c) the method for determining the processing ratio is different in the three scenes after determining that the optimum solution has been obtained (step S11).

  Until the first n pieces of client result information are prepared (that is, until the initial sample group is completed), the determination unit 13 determines the processing ratio at random (step S12). When the client terminal 20 and the server 10 perform data interpolation based on this processing ratio, the determination unit 13 stores the client performance information obtained by the data interpolation as a sample (step S12). The determination unit 13 repeats this series of processes until n pieces of client performance information are obtained as an initial sample group (step S12).

  When n samples (client performance information) are obtained, the determination unit 13 searches for an optimal solution of the processing ratio in the group using a genetic algorithm. Specifically, the determination unit 13 evaluates the sample group by ranking these samples based on the client performance information (Step S13), and executes the following processing while the optimal solution changes (Step S14). ; YES). That is, the determination unit 13 leaves a sample with a processing ratio of the highest h% as a higher group, and discards the remaining samples (step S15). Here, “the result is higher” means that the evaluation value based on at least one of the processing time and the total processing time is higher. For example, the determination unit 13 may select a plurality of samples as an upper group in the order of short processing time or total processing time, and the evaluation value calculated by an arbitrary method based on the processing time and the total processing time is high. A plurality of samples may be set as the upper group in order.

  Subsequently, the determination unit 13 selects a parent (two processing ratios) using ranking selection in the parent selection method of the genetic algorithm (step S16), and may be closer to a more appropriate processing ratio from the parent. A new processing ratio having the property is generated (step S17). When the client terminal 20 and the server 10 execute data interpolation based on the new processing ratio, the determination unit 13 stores the client performance information obtained by the data interpolation as a sample (step S18). The determination unit 13 repeats the processes of steps S16 to S18 until the number of samples reaches the threshold value n (see step S19).

The processing of steps S16 to S18 will be described in further detail. To each sample of the decision unit 13 first-level group, allocating the occurrence probability p _f depending on the rank r based on the results of operations. Rank r is sampled as appearance probability p _f is set large close to 1, it is selected with high probability as an element used for generating the processing percentages. The determination unit 13 obtains the appearance probability _{pf according} to the equation (1).

Next, select the two treatment rates randomly in accordance with the determined unit 13 has obtained occurrence probability p _f. If the two processing ratios are the same, the determination unit 13 performs the selection process again so that different processing ratios are obtained. In addition, even when the combination of processing ratios is the same as that selected in the past, the determination unit 13 performs the selection process again. When a new combination that is not the same value is selected, the determination unit 13 generates a new processing ratio by obtaining the median value of the combination. In this series of processes, the combination of the sample, the local optimal solutions for the optimization only a limited range of input values so results in optimal problems may occur, the determination unit 13 using the mutation probabilities p _m processing The ratio is varied randomly. By this processing, it becomes possible to bring the optimum solution closer to the global optimum solution that is the optimum solution as a whole, not the local optimum solution.

  When the number of samples reaches the threshold value n (step S19; YES), the determination unit 13 evaluates the sample group again (step S13), and repeats the processing of steps S15 to S19 while the optimal solution changes, thereby determining the optimal solution. I will explore. The determination unit 13 searches for an optimal solution by repeating the process of generating a new processing rate from the processing rate with good results. If the optimal solution does not change even after trying steps S15 to S19 X times (step S14; NO), the determination unit 13 determines that the optimal solution for the processing ratio has been acquired (step S20).

  After acquiring the optimal solution, the determination unit 13 sets the optimal solution as the processing ratio for the client terminals 20 belonging to the group (step S21).

It should be noted that the threshold value X, the ratio h (%), the mutation probability p _m , and the threshold value X determination method used in such a genetic algorithm are not limited.

  In this way, the determination unit 13 searches for an optimal processing ratio while executing data interpolation between the server 10 and the client terminal 20. Therefore, the processing rate determined by the determination unit 13 in response to a request from the client terminal 20 is not always optimal.

  The allocation unit 14 is a functional element that reads uninterpolated data to be processed from the database 30 and allocates the uninterpolated data to the client terminal 20 and the server 10 based on the processing ratio determined by the determination unit 13. The allocating unit 14 transmits the uninterpolated data allocated to the client terminal 20 to the client terminal 20 and outputs the uninterpolated data allocated to the server 10 to the interpolating unit 15.

  The allocation unit 14 allocates the non-interpolated data to the client terminal 20 and the server 10 so that the allocation ratio of the non-interpolated data matches the processing rate. For example, if the processing rates of the client terminal 20 and the server 10 are determined to be 0.3 and 0.7, respectively, the allocation unit 14 allocates 30% of the uninterpolated data to the client terminal 20 and the remaining 70% Allocate minutes to the server 10. Of course, depending on the number of values constituting the uninterpolated data, the allocation ratio of the uninterpolated data may not completely match the processing ratio. For example, even if the processing ratio between the client terminal 20 and the server 10 is determined to be 3: 7, the allocation ratio of uninterpolated data may be 2.8: 7.2.

  The allocating unit 14 may simply distribute the non-interpolated data to the client terminal 20 and the server 10. Alternatively, the assigning unit 14 may assign the non-interpolated data obtained by the sensor S installed in a specific region in the space R to the client terminal 20 while following the processing rate. For example, the allocating unit 14 identifies one or more sensors S corresponding to the area requested from the client terminal 20 (for example, the line-of-sight direction measured by the HMD or the area based on the viewing area), and obtained by the sensor S. Uninterpolated data may be assigned to the client terminal 20. In this case, it becomes possible for the client terminal 20 to first perform data interpolation in the requested region and visualize the environment value in the region first (interpolated data obtained from the server 10 is Then visualized).

  The interpolation unit 15 is a functional element that generates interpolated data by performing data interpolation on uninterpolated data assigned to the server 10. The data interpolation algorithm is not limited. For example, an inverse distance weighting method that is one of spatial interpolation algorithms may be used. By this processing, an estimated value of the environment (for example, temperature, humidity, radio wave intensity) in an arbitrary region between adjacent sensors S is obtained. The interpolation unit 15 transmits the obtained interpolated data to the client terminal 20. The interpolated data is data including both the environmental value indicated by the non-interpolated data and the estimated environmental value at a virtual measurement point between adjacent sensors.

  Next, the functional configuration of the client terminal 20 will be described. As illustrated in FIG. 1, the client terminal 20 includes a page acquisition unit 21, a client information providing unit (acquisition unit) 22, an interpolation unit 23, and a display control unit (output unit) 24 as functional components.

  The page acquisition unit 21 is a functional element that acquires a web page for visually displaying the state of the environment in the space from the server 10. The page acquisition unit 21 transmits a page request (a request for a web page for visually displaying the state of the environment in the space) to the server 10 based on a user operation, and in response to the request, the server 10 The page data sent from the (page providing unit 11) is received. When the page data is executed by the CPU 101, the web page is displayed on the web browser, and a request for uninterpolated data is transmitted from the client terminal 20 to the server 10.

  The client information providing unit 22 is a functional element that acquires client information (client environment information and client results information), and is realized by the CPU 101 executing the client_environment_acquire function, the client_network_speedcheck function, and the interpolate_benchmark function embedded in the page data. Is done. The client information providing unit 22 acquires the OS and web browser information and the communication speed using the client_environment_acquire function and the client_network_speedcheck function, and transmits these information to the server 10. Further, the client information providing unit 22 transmits the processing time obtained by the interpolate_benchmark function and the total processing time to the server 10 as client result information.

  The interpolation unit 23 is a functional element that generates interpolated data by performing data interpolation on uninterpolated data (uninterpolated data received from the server 10) assigned to the client terminal 20, and is embedded in page data This is realized by the CPU 101 executing the interpolate_benchmark function. Similar to the interpolation unit 15 of the server 10, the data interpolation algorithm is not limited, and an arbitrary method such as a reverse distance weighting method can be used. By this processing, an estimated value of the environment (for example, temperature, humidity, radio wave intensity) in an arbitrary region between adjacent sensors S is obtained. The interpolation unit 23 outputs the obtained interpolated data to the display control unit 24.

  The display control unit 24 is a functional element that outputs a combination of the interpolated data obtained by data interpolation in the client terminal 20 and the interpolated data obtained by data interpolation in the server 10. Here, the term “join” means a set of interpolated data obtained by the client terminal 20 and the server 10 that is output as a set of data. When the interpolated data is input from the interpolating unit 23 and the interpolated data is received from the server 10, the display control unit 24 visually displays the interpolated data on the web page by computer graphics. By this processing, the interpolated data is visualized.

  FIG. 4 schematically shows an example of a three-dimensional image provided by the display control unit 24 and showing the environment in the space. In FIG. 4, the magnitude of the environmental value in the space is indicated by the shaded shade, but the three-dimensional image is a color image (for example, a place where the environmental value is high is indicated by a color close to red and a place where the environmental value is low is indicated by blue The image may be a color close to or a gray scale image. The example (a) in FIG. 4 is a three-dimensional image in which the environment of the entire space R is visualized, and the example (b) is a screen in which the visualized environment value is superimposed on the video obtained from the camera C existing in the space R. (Screen providing augmented reality).

  Next, the operation of the distributed processing system 1 will be described with reference to FIGS. 5 and 6 and the distributed processing method according to the present embodiment will be described.

  First, the page acquisition unit 21 of the client terminal 20 transmits a page request to the server 10 (step S111), and the page providing unit 11 of the server 10 transmits page data to the client terminal 20 in response to the request (step S112). . This page data includes a script or a program (for example, a client_environment_acquire function, a client_network_speedcheck function, and an interpolate_benchmark function) for acquiring client information.

  In the client terminal 20, the page acquisition unit 21 receives the page data. By executing this page data by the CPU 101, a web page is displayed on the screen. Further, by executing the script in the page data, the client information providing unit 22 acquires the OS and web browser information and the communication speed as client environment information (step S113, acquisition step), and the client environment information is acquired. It transmits to the server 10 (step S114). Further, the page acquisition unit 21 requests uninterpolated data from the server 10 (data request, step S115).

  In the server 10, the client information acquisition unit 12 receives client environment information (acquisition step). Subsequently, the determination unit 13 determines the processing rate of data interpolation between the client terminal 20 and the server 10 using an algorithm such as machine learning (step S116, determination step). Then, the assigning unit 14 extracts uninterpolated data to be processed from the database 30, and assigns the uninterpolated data to the client terminal 20 and the server 10 based on the processing ratio (step S117, assignment step). The allocating unit 14 transmits uninterpolated data (uninterpolated data for the client terminal) allocated to the client terminal 20 to the client terminal 20 (step S118).

  In the client terminal 20, the interpolation unit 23 receives the uninterpolated data and performs data interpolation on this data (step S119). At this time, the client information providing unit 22 acquires client result information obtained by executing the data interpolation. On the other hand, in the server 10, the interpolation unit 15 performs data interpolation on the non-interpolated data (server non-interpolated data) input from the allocating unit 14 (step S120), and the interpolated data generated by this processing is used as the client. It transmits to the terminal 20 (step S121). In the client terminal 20, the interpolation unit 23 transmits the client result information to the server 10, and the client information acquisition unit 12 receives the information (step S122, acquisition step). Further, the display control unit 24 visualizes the combination of the interpolated data generated by the interpolation unit 23 and the interpolated data received from the server 10 (step S123, output step).

  In the present embodiment, when executing the next data interpolation (processing for interpolating the next uninterpolated data), the determination unit 13 uses an algorithm such as machine learning with the client environment information and the client performance information as inputs. Then, the processing rate of data interpolation between the client terminal 20 and the server 10 is determined (reset) (step S124, determination step). When the client terminal 20 requests the next data (step S125), the allocating unit 14 extracts the next uninterpolated data from the database 30, and the uninterpolated data based on the latest processing ratio (reset processing ratio). Are allocated to the client terminal 20 and the server 10 (step S126, allocation step). The allocating unit 14 transmits uninterpolated data (uninterpolated data for the client terminal) allocated to the client terminal 20 to the client terminal 20 (step S127).

  In the client terminal 20, the interpolation unit 23 receives the uninterpolated data, performs data interpolation on this data (step S128), and the client information providing unit 22 acquires client performance information obtained by executing the data interpolation. To do. On the other hand, in the server 10, the interpolation unit 15 performs data interpolation on the remaining uninterpolated data (server non-interpolated data) (step S129), and transmits the interpolated data generated by this processing to the client terminal 20. (Step S130). In the client terminal 20, the interpolation unit 23 transmits the next client result information to the server 10, and the client information acquisition unit 12 receives the information (step S131, acquisition step). Further, the display control unit 24 visualizes the combination of the next interpolated data (step S132, output step). On the other hand, in the server 10, the processing ratio of data interpolation between the client terminal 20 and the server 10 using an algorithm such as machine learning in which the determination unit 13 receives the client environment information and the newly obtained client performance information. Is determined (reset) (step S133, determination step).

  The distributed processing system 1 repeats the processing of steps S124 to S132 in this manner, thereby executing a series of data interpolation while dynamically adjusting the processing ratio between the client terminal 20 and the server 10, and the processing ratio is set. It can be close to the appropriate value. As a result, distributed processing can be executed efficiently. Further, by repeating the data interpolation and visualization, the user can grasp a change in the environment in the space R (for example, a change in the environmental value with the passage of time).

  Note that the execution order of the steps is not limited to that shown in FIGS. For example, the execution timing of the processes in steps S114 and S115 along the time axis may be changed, or the execution timing of the processes in steps S121 and S122 (or steps S130 and 131) may be changed.

  Next, a distributed processing program P for realizing the distributed processing system 1 will be described with reference to FIG. The distributed processing program P includes a server program P1 for causing the computer to function as the server 10 and a client program P2 for causing the computer to function as the client terminal 20. The language in which the program is described may be arbitrarily determined. As an example, it is conceivable that the server program P1 is described in C ++ and the client program P2 is described in JavaScript (trademark or registered trademark).

  The server program P1 includes a main module P10, a page providing module P11, a client information acquisition module P12, a determination module P13, an allocation module P14, and an interpolation module P15. The main module P10 is a part that comprehensively controls functions on the server side. The functions realized by executing the page providing module P11, the client information acquiring module P12, the determining module P13, the assigning module P14, and the interpolation module P15 are respectively the page providing unit 11, the client information acquiring unit 12, and the determining unit. 13, the functions of the assigning unit 14 and the interpolation unit 15 are the same.

  The client program P2 includes a main module P20, a page acquisition module P21, a client information provision module P22, an interpolation module P23, and a display control module P24. The main module P20 is a part that comprehensively controls the functions on the client side. The functions realized by executing the page acquisition module P21, the client information provision module P22, the interpolation module P23, and the display control module P24 are respectively the page acquisition unit 21, the client information provision unit 22, the interpolation unit 23, and The function is the same as that of the display control unit 24.

  The server program P1 and the client program P2 may be provided after being fixedly recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, these programs P1 and P2 may be provided via a communication network as data signals superimposed on a carrier wave.

  As described above, the distributed processing system according to one aspect of the present invention is a distributed processing system including a client terminal and a server, and acquires client information indicating an environment or an operation result of the client terminal from the client terminal. The processing capability of the client terminal is estimated by executing the machine learning using the acquisition unit and the client information acquired by the acquisition unit, and the processing ratio between the client terminal and the server is determined based on the estimated processing capability The processing unit is a rate at which data processing using unprocessed data is performed. Based on the determination unit and the processing rate determined by the determination unit, the client terminal and the server are not processed. An assigning unit for assigning data, processed data obtained by data processing at the client terminal, The binding between the obtained processed data by the data processing in over server, and an output unit for outputting the client terminal.

  A distributed processing method according to an aspect of the present invention is a distributed processing method executed by a distributed processing system including a client terminal and a server, and acquires client information indicating an environment or an operation result of the client terminal from the client terminal. And the processing capability of the client terminal is estimated by executing machine learning using the client information acquired in the acquisition step, and the processing ratio between the client terminal and the server is calculated based on the estimated processing capability. A determination step for determining, wherein the processing rate is a rate at which data processing using unprocessed data is executed, and based on the determination rate and the processing rate determined in the determination step, An allocation step for assigning process data and a client terminal That the processed data obtained by the data processing, the bond between the obtained processed data by the data processing in the server, and an output step of outputting from the client terminal.

  A distributed processing program according to one aspect of the present invention is a distributed processing program that causes a computer system including at least two computers to execute a distributed processing method executed by a distributed processing system including a client terminal and a server. The distributed processing method obtains client information indicating an environment or an operation result of the client terminal from the client terminal, and performs machine learning using the client information obtained in the obtaining step, thereby increasing the processing capability of the client terminal. Estimating and determining a processing rate between the client terminal and the server based on the estimated processing capability, wherein the processing rate is a rate of executing data processing using unprocessed data, The decision step and the decision step Based on the processing ratio, the allocation step of allocating unprocessed data to the client terminal and the server, the combination of the processed data obtained by the data processing at the client terminal and the processed data obtained by the data processing at the server, An output step of outputting from the client terminal.

  In such an aspect, the processing capability of the client terminal is estimated by machine learning based on the environment or operation result of the client terminal, and the processing ratio between the client terminal and the server is determined based on the estimation result. Therefore, the same data processing can be distributed between the client terminal and the server in consideration of the environment or operation result of the client terminal. As a result, the entire system load can be efficiently distributed.

  If the client environment or operation results cannot be grasped, there is a possibility that a large load is applied to a client terminal having a lower processing capacity than that of the server, and the processing efficiency of the entire system is lowered. On the other hand, since the operating environment and specifications of each client terminal vary, if the processing ratio is uniformly determined, processing results can be provided to a wide variety of client terminals within an acceptable response time. It is very difficult. On the other hand, in the above aspect of the present invention, similar data processing can be distributed between the client terminal and the server by considering the environment or operation result of the client terminal. For example, data interpolation that is generally computationally expensive can be processed by both the client terminal and the server within the time allowed by the user.

  In the distributed processing system according to another aspect, the client information includes client environment information indicating at least the operating system and the web browser of the client terminal, and client result information indicating an operation result of data processing in the client terminal, and a determination unit However, the client environment information may be used as an input parameter for machine learning, and the client performance information may be used as correct data for machine learning.

  By acquiring information on the operating system and the web browser, the general type or specification of the client terminal can be known, and as a result, the processing capability of the client terminal can be estimated. For example, by specifying whether the client terminal is a desktop machine or a portable terminal, it is possible to roughly estimate whether the processing capability is high or low. On the other hand, the actual operation result of the data processing at the client terminal is given as correct data. As a result, in machine learning, it is possible to determine an appropriate processing ratio for the client terminal in consideration of the approximate processing capability of the client terminal and the actual operation result. For example, since a mobile terminal generally has a lower processing capacity than a desktop machine, it is desirable that the processing rate of the mobile terminal be lower than the processing rate of the desktop machine.

  In the distributed processing system according to another aspect, the client environment information may further indicate a communication speed between the client terminal and the server.

  The communication speed between the client terminal and the server is helpful in determining how much data should be passed to the client terminal. For example, in data interpolation, interpolated data in a certain space may swell many times as much as uninterpolated data in the space. Therefore, when the communication speed is low, the processing rate of the client terminal is increased compared to when the communication speed is high, and a large amount of uninterpolated data is transmitted to the client terminal. It is conceivable to reduce the cost of data communication while preventing it. By considering not only the environment of the client terminal itself but also the communication speed, it is possible to determine an appropriate processing ratio for the client terminal.

  In the distributed processing system according to another aspect, the determination unit randomly determines the processing rate until the number of samples reaches the threshold value, and when the number of samples reaches the threshold value, the data processing result at the client terminal is The processing ratio may be determined using an upper group consisting of upper sample groups. Here, the sample is client result information indicating an operation result of data processing at the client terminal.

  By randomly determining the processing ratio until a large number of samples (client performance information) are gathered, it is possible to verify whether or not the various processing ratios are appropriate values as if performing a benchmark test. When a certain number of samples are collected, the processing rate is determined using a sample group having a higher data processing result at the client terminal, so that a processing rate appropriate for the client terminal can be determined.

  In the distributed processing system according to another aspect, when the determination unit determines that the optimal solution of the processing rate has been acquired, the optimal solution may be determined as the processing rate.

  If an optimal solution is found, it is efficient to continue using the optimal solution as a processing ratio. This is because the processing time required to continue searching for the optimal solution can be omitted, and the data processing can be continuously distributed between the client terminal and the server at a processing rate that seems to be most appropriate.

  In the distributed processing system according to another aspect, the data processing may be data interpolation. In general, data interpolation has a high processing load, and the amount of data greatly increases by interpolation processing. Therefore, when a client terminal having a processing capability lower than that of the server performs the processing, it is necessary to adjust the amount of unprocessed data (uninterpolated data) transmitted to the client. By applying the present invention to data interpolation, efficient processing can be expected.

  In the distributed processing system according to another aspect, the unprocessed data may be data indicating an environmental value in the space, and the output unit may display the combination of the processed data as a three-dimensional image.

  In this case, the environmental value in the space can be interpolated to present the state of the environment in the space to the user in an easy-to-understand manner. Advanced image processing that provides augmented reality requires appropriate computation and data communication. However, by interpolating data interpolation between the client terminal and the server by the above processing, environment values are interpolated. Thus, it becomes possible to execute a series of processes visualized by a three-dimensional image in almost real time.

  The present invention has been described in detail based on the embodiments. However, the present invention is not limited to the above embodiment. The present invention can be variously modified without departing from the gist thereof.

  The specific algorithm of machine learning for estimating the processing capability of the client terminal is not limited. Moreover, although the genetic algorithm was used in the said embodiment, utilization of this algorithm is not essential. Alternatively, another algorithm may be used instead of the genetic algorithm.

  In the above embodiment, the OS and browser information and the communication speed are used as the client environment information. However, the client environment information may be only a part of these, or information different from these (for example, CPU and Information on hardware such as memory and disk I / O) may be used.

  In the above embodiment, the data processing time and the total processing time using the uninterpolated data assigned to the client terminal 20 as input are used as the client result information, but only one of these two types of processing is used. Time may be used as client performance information. Alternatively, a CPU operating rate or a memory usage statistical value (for example, an average value or a median value) while data interpolation is being performed may be used as client performance information.

  In the above embodiment, the distributed processing system 1 adjusts the processing ratio every time a data request is received from the client terminal 20, but the timing for dynamically adjusting the processing ratio is not limited to this. For example, the determination unit of the distributed processing system may determine the processing rate at regular time intervals (that is, periodically), or may determine the processing rate in response to a user's explicit instruction.

  In the above embodiment, the display control unit 24 visualizes the combination of the interpolated data on the web browser with a three-dimensional image, but the output method of the interpolated data is not limited. For example, the distributed processing system may display the combination of the interpolated data in the form of a two-dimensional image, or may display it in the form of a two-dimensional image or a three-dimensional image on an application other than the web browser. The distributed processing system may output a combined three-dimensional image of interpolated data to a printer or store it in a storage device such as a memory or a database. Alternatively, the distributed processing system may output the interpolated data combination to another computer system via a communication network.

  In the above embodiment, the distributed processing system 1 interpolates the environment values in the space, but the content indicated by the non-interpolated data processed by the distributed processing system of the present invention is not limited at all. The distributed processing system can be applied to arbitrary data interpolation processing such as interpolation of image data or audio data. In the above embodiment, the present invention is applied to data interpolation. However, the present invention may be applied to other arbitrary data processing such as data collection and calculation.

DESCRIPTION OF SYMBOLS 1 ... Distributed processing system, 10 ... Server, 11 ... Page provision part, 12 ... Client information acquisition part, 13 ... Determination part, 14 ... Assignment part, 15 ... Interpolation part, 20 ... Client terminal, 21 ... Page acquisition part, 22 ... Client information providing unit, 23 ... Interpolating unit, 24 ... Display control unit, 30 ... Database, P ... Distributed processing program, P1 ... Server program, P10 ... Main module, P11 ... Page providing module, P12 ... Client information obtaining module, P13 ... determination module, P14 ... allocation module, P15 ... interpolation module, P2 ... client program, P20 ... main module, P21 ... page acquisition module, P22 ... client information provision module, P23 ... interpolation module, P24 ... display control module.

Claims (7)

A distributed processing system comprising a client terminal and a server , wherein each of the client terminal and the server executes data processing using unprocessed data ,
Client including client environment information indicating a communication speed between the operating system and web browser of the client terminal and the client terminal and the server, and client result information indicating an operation result of previous data processing in the client terminal An acquisition unit for acquiring information from the client terminal;
The processing capability of the client terminal is estimated by executing machine learning using the client environment information as an input parameter and the client performance information as correct data. Based on the estimated processing capability, the client terminal and the server a determination unit that determines the processing percentage between the treat rate is the rate at which to execute the next data processing using the following raw data, and said determination unit,
An allocating unit that allocates the next unprocessed data to the client terminal and the server, based on the processing rate determined by the determining unit;
Dispersion comprising the processed data obtained by the next data processing in the client terminal, the bond between the obtained processed data by the next data processing in the server, and an output unit for outputting from said client terminal Processing system. The determination unit randomly determines the processing ratio until the number of samples reaches a threshold value, and when the number of samples reaches the threshold value, the data processing result at the client terminal is a high-order sample group using a higher population of determining the processing rate, wherein said sample is the client performance information,
The distributed processing system according to claim 1 . When the determination unit determines that the optimal solution of the processing rate has been acquired, the optimal solution is determined as the processing rate.
The distributed processing system according to claim 2 . The data processing is data interpolation;
The distributed processing system as described in any one of Claims 1-3 . The raw data is data indicating an environmental value in space;
The output unit displays a combination of the processed data as a three-dimensional image;
The distributed processing system according to claim 4 . A distributed processing method executed by a distributed processing system , comprising a client terminal and a server , wherein each of the client terminal and the server executes data processing using unprocessed data ,
Client including client environment information indicating a communication speed between the operating system and web browser of the client terminal and the client terminal and the server, and client result information indicating an operation result of previous data processing in the client terminal An acquisition step of acquiring information from the client terminal;
The processing capability of the client terminal is estimated by executing machine learning using the client environment information as an input parameter and the client performance information as correct data. Based on the estimated processing capability, the client terminal and the server a determining step of determining a processing percentage between the treat rate is the rate at which to execute the next data processing using the following raw data, and said determining step,
An allocating step of allocating the next unprocessed data to the client terminal and the server based on the processing rate determined in the determining step;
Dispersion containing the processed data obtained by the next data processing in the client terminal, the bond between the obtained processed data by the next data processing in the server, and an output step of outputting from said client terminal Processing method. A computer comprising a client terminal and a server , wherein each of the client terminal and the server executes data processing using unprocessed data, and a distributed processing method executed by a distributed processing system is provided with at least two computers A distributed processing program to be executed by a system,
The distributed processing method includes:
Client including client environment information indicating a communication speed between the operating system and web browser of the client terminal and the client terminal and the server, and client result information indicating an operation result of previous data processing in the client terminal An acquisition step of acquiring information from the client terminal;
The processing capability of the client terminal is estimated by executing machine learning using the client environment information as an input parameter and the client performance information as correct data. Based on the estimated processing capability, the client terminal and the server a determining step of determining a processing percentage between the treat rate is the rate at which to execute the next data processing using the following raw data, and said determining step,
An allocating step of allocating the next unprocessed data to the client terminal and the server based on the processing rate determined in the determining step;
Wherein comprising a processed data obtained by the next data processing in the client terminal, the bond between the obtained processed data by the next data processing in the server, and an output step of outputting from said client terminal,
Distributed processing program.

Images