JPH0887473A - Data processor - Google Patents

Abstract

PURPOSE: To automate system management and tuning by using a system for data arrangement determination and a system which determines which process request executing computer respective process requests are allocated to on the basis of the history of the respective process requests at the same time, and making them cooperate with each other. CONSTITUTION: This data processor consists of (m) computers 110-1 110-m for process request execution, a computer 111 for front-end processing, a coupling path 114 which mutually connects the front-end computer 111 and respective process request executing computers 110-1 110-m, and data storage devices 118-1-118-m which are connected to the respective process request executing computers 110-1-110-m and store data. Those computers perform parallel data processing with high performance and the loads of the process requests are decentralized to cope with highly-loaded data processing. Then a process request allocation part 101 obtains information on the execution history of past process requests and selects the computers 110-1-110m as allocation destinations on the basis of the information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data processing device composed of a plurality of computers.

[0002]

2. Description of the Related Art A data processing apparatus that receives a processing request from a processing request generation unit such as a plurality of terminal devices, a computer, or an automatic cash dispenser connected by a communication path, and executes a predetermined processing based on the processing request is known. Widely used today. Conventionally, such a data processing device has been configured using a single general-purpose computer, but when higher processing performance is required, it has come to be configured using multiple computers. There is.

A data processing device using such a plurality of computers may have a non-shared data structure or a shared data structure. FIG. 38 is an example of a conventional data processing apparatus having a data non-shared type configuration, and FIG. 39 is an example of a conventional data processing apparatus having a data sharing type configuration. In any of the configurations, the application programs 1116-1 to 1116-m are installed in the plurality of processing request execution computers 1110-1 to 1110-m.
Respectively, and the process request allocating unit 1101 on the front-end computer 111 that receives the process requests from the process request generating units 1102-1 to 1102-n connected via the communication path 1115 is assigned to each process request executing computer. It is a method of assigning execution of processing requests.

The difference between the non-shared data type and the shared data type is
It is a combination of a data storage device and a processing request execution computer. In the case of the non-shared data type, each processing request execution computer locally has the data storage devices 1118-1 to 1118-m, and the data is divided and stored in each data storage device. Therefore, the data management units 1117-1 to 1117-
The m can only access the data on the data storage device connected to the own processing request execution computer. On the other hand, in the case of the data sharing type, since all the processing request execution computers share the data storage device 1118, any processing request execution computer can access any data.

As a method for the processing request allocating unit 1101 to allocate processing requests to each processing request execution computer,
For example, by looking at the correspondence table of a given processing request and the computer that executes that request, you can determine which processing request execution computer to assign in the case of processing, or determine the load status of each processing request execution computer. There is a method in which a means for investigating is added and the processing is assigned to the processing request execution computer determined to have a small load.

[0006] For example, "TUXEDO SYSTEM
Release 4.1 Transaction M
manager Administrator's Gu
"ide" (Prentice Hall) describes a method of determining the processing request execution computer by looking at the argument of the processing request.

When a plurality of data storage devices are used as in the configuration shown in FIG. 38, how to divide the data to be stored into the plurality of data storage devices and arrange the data is described in, for example, database benchmarking. (Jim Gray
Edited by Yu Kitsuregawa, Eiichi Watanabe, Nikkei BP Co., Ltd.) Data is randomly arranged in a data storage device by using a hash function, etc. In some cases, the range of data values is divided into the number of data storage areas, and these are assigned to the data storage device one by one.
In addition, as a method for changing the arrangement of data, "Dat
a Placement In Bubba, Geor
ge Copeland et. al. , SIGMO
D, Vol. 17, No. 3, 1988 ”, there is a method of calculating an optimum arrangement based on information for data access and performing the rearrangement.

By the way, in order to efficiently execute a job by using a plurality of processing request execution computers, it is necessary to allocate each processing request to a processing request execution computer whose processing cost is low. When using this data storage device, it is necessary to divide the data into a plurality of data storage devices and arrange them.

Conventionally, a system administrator examines an application program executed in a data processing device to analyze what data is accessed by the application program, and preferably executes individual processing requests in parallel. In addition, the data required to execute each processing request is stored in the processing request execution computer that executes the processing request as much as possible, and further, the data allocation and the frequency of occurrence of each processing request to the data processing device, etc. In consideration of the above, it is necessary to determine the distributed arrangement of data and the allocation method of the processing requests to the processing request execution computers so that the cost required to execute each processing request becomes as small as possible.

However, even if only one of the distributed arrangement of data and the allocation method of processing requests is determined,
It was very difficult to analyze the dynamic characteristics of the entire system in order to make full use of the system performance.

[0011]

As described above, in the conventional data management apparatus, the dynamic characteristics of the entire system can be analyzed by optimizing either one of the data distributed arrangement and the processing request allocation method. Since it is very difficult to do so, there is a problem that the system performance cannot be fully exhibited. In addition, there is a problem that it takes a lot of time to manage and tune the system.

The present invention has been made in view of the above circumstances, and creates / creates information necessary for determining data allocation.
Simultaneously using the method of automatically executing the analysis and determining the optimum data arrangement and the method of deciding which processing request execution computer to allocate each processing request based on the history of each processing request,
An object of the present invention is to provide a data processing device which can bring out the system capability sufficiently and can automate system management and tuning by coordinating these.

[0013]

The present invention provides a transaction processing system comprising at least one first computer that issues a processing request and a plurality of second computers that execute the issued processing request. Processing request allocating means for transferring a processing request generated by one computer to any of the plurality of second computers, and data storage means for storing data used for executing the processing request, which is connected to the second computer And means for determining the arrangement of the data stored in this means and rearrangement.

According to a first aspect of the present invention, the processing request allocating means acquires information on an execution history of past processing requests, and selects a second computer as an allocation destination based on the history information. It is characterized by

On the other hand, according to a second aspect of the present invention, the processing request allocating means acquires information on an arrangement of data stored in the storage means, and based on the arrangement information, a second computer as an allocation destination. Is selected.

In a preferred aspect of the first invention of the present invention, a plurality of data storage means for storing data and any one of the data storage means are connected, and a given processing request is made by using an application program. Data management for accessing the data in the data storage means necessary for the execution based on a request from the application program for executing the processing request and the processing request, and accumulating information on a history of access to the data A plurality of computers each having a means, a coupling means for interconnecting the plurality of computers, and a processing request sent from at least one processing request generation means connected to the coupling means and connected via a communication path. Accepting means, past processing request,
The computer that has executed the processing request, a means for accumulating an execution history including a cost required for the execution, and a new processing request sent from the processing request generation means, which should be executed by comparing with the execution history. Processing request allocating means having means for selecting the computer in which an application program exists and instructing execution of the processing request; and receiving information on the history of access to the data from the data managing means on the computer, The data allocation determining means determines the optimum data allocation based on the information, and instructs the data managing means to execute the data allocation changing process based on the determined data allocation. It is characterized by

In a preferred aspect of the second aspect of the present invention, a plurality of data storage means for storing data and any one of the data storage means are connected, and a given processing request is made by using an application program. Data management for accessing the data in the data storage means necessary for the execution based on a request from the application program for executing the processing request and the processing request, and accumulating information on a history of access to the data A plurality of computers each having a means, a coupling means for interconnecting the plurality of computers, and a processing request sent from at least one processing request generation means connected to the coupling means and connected via a communication path. Receiving means, indicating what kind of data is stored in the data storage means connected to each computer Means for acquiring data allocation information, and when receiving a new processing request sent from the processing request generating means, compares the data allocation information and selects the computer having an application program to be executed to execute the processing request. Processing request allocating means having means for instructing, and information regarding the history of access to the data from the data managing means on the computer, respectively, and based on the information, determining the optimum data arrangement, respectively, It is characterized by further comprising a data arrangement determining means for respectively instructing the data managing means to execute a data arrangement changing process based on the determined data arrangement.

Preferably, in the first and second inventions, the data management unit receives a data operation request accompanying the execution of an application program, and if the data to be operated is in a data storage device of its own computer, an operation request. Is executed, and when it is not in the own computer, a data operation is requested to the data management unit of the computer where the data exists. That is, the application program is provided with means for transparently accessing all data.

Preferably, in the first and second aspects of the invention, the data management section has means for changing the data arrangement according to the data arrangement determined by the data arrangement determining means.

Preferably, in the first and second aspects of the invention, the data arrangement determining means has means for instructing the start of processing for determining a new data arrangement.

[0021] Preferably, in the second invention, the data arrangement determining means includes means for notifying the processing request allocation section of the determined arrangement of the new data as data arrangement information.

Preferably, in the second invention, each of the data management units has means for notifying the processing request allocation unit of information of data to be managed as data arrangement information.

Preferably, in the first and second inventions, the same application program is arranged on each computer.

[0024]

According to the present invention, the processing request allocating section selects an optimum computer that executes the processing request, that is, a computer that can execute the processing request at the lowest cost, and allocates the processing. Usually, the cost is lowest when a process is assigned to a computer that has the data required to execute a processing request, but when assigned to a computer that does not have data, the cost is high because it requires access to remote data. Increase. At this time, all the application programs can transparently access all the data through the data management unit, so that it is not always guaranteed that the application programs are assigned to the optimum computers.

When the above processing requests are allocated,
The execution time of each processing request varies depending on the data arrangement,
In addition, the load on each computer may or may not be balanced. Therefore, based on the information indicating which data needs to be accessed to execute which processing request, the data allocation that optimizes the execution time of each processing request and the load on each computer is obtained, and the data is placed in that state. Reposition.

The processing request allocating section can balance the loads by allocating the processing requests according to the new data arrangement. Also, the execution time of each processing request is optimized. By such a feedback loop of processing request allocation and data rearrangement according to the data arrangement, the load balancing of the entire data processing apparatus is automated and the performance of the entire data processing apparatus is also improved.

Here, a feedback loop for processing request allocation and data relocation according to data placement is performed as follows in each invention.

In the first aspect of the present invention, the processing request allocating means accumulates the cost obtained as a result of allocating the processing request to a certain computer, and learns the optimum allocation method based on the accumulated cost. , The data arrangement determining means,
While observing the data access history, change the data layout to the optimum one. And the change of this data arrangement is
It is indirectly fed back to the processing request allocation result of the processing request allocation means.

In the second invention, the processing request allocating means allocates the processing request to each computer based on the data allocation information, while the data allocation deciding means appropriately allocates the data while observing the data access history. To the optimum one. Then, the change of the data arrangement is directly given to the processing request allocating means.

As described above, by using the data processing apparatus of the present invention, high-performance data processing can be realized by executing processing in parallel using a plurality of processing request execution computers. Further, even when the load of the processing request is high, by distributing the application program that executes the processing request to a plurality of processing request execution computers, it is possible to handle high-load data processing.

Further, according to the present invention, no matter which data storage device of the data sharing type or the data distribution type is used, each application program stores data in the data storage device connected to which processing request execution computer. Since it can also be accessed, it becomes possible to perform parallel processing without changing the application program developed on the conventional single computer. Also, due to this property, the processing request allocation device is not necessarily a deterministic algorithm that can select the optimal processing request execution computer every time, but uses a probabilistic algorithm that may exist when selecting a non-optimal processing request execution computer. Requests can be assigned.

Further, by the automatic optimum allocation of data by the data allocation determining unit, optimum data division allocation is performed for dynamic characteristics such as what processing request occurs at what frequency. The execution performance of each processing request is improved. In addition, according to the data allocation, the processing request allocation unit allocates each processing request to the processing request execution computer suitable for executing the processing request, so that automatic allocation is possible without determining the data distribution allocation and processing request allocation method. The processing load can be balanced, tuning of the entire data processing device can be automated, and system management becomes easy.

Further, even if the optimum allocation strategy in the processing request allocation unit is not clear or the optimum data distribution arrangement is not clear, there is a possibility that the processing load distribution will not be successful, but data processing It can operate as a device.

[0034]

Embodiments of the present invention will be described below with reference to the drawings.

Of the reference numbers given to the respective constituent elements (blocks), "i-j" and "i-j-k" (for example, 102-1)
The ones in the form of indicate that the same "i" have the same configuration. In addition, when it is not necessary to distinguish each of the blocks with the reference number i, “-
The description will be made by omitting "j" and "-j-k".

(First Embodiment) FIG. 1 shows the configuration of a first embodiment of the present invention.

The data processing apparatus shown in FIG. 1 includes m (m is 2 or more) computers for executing processing requests (processing request executing computers) 110-1 to 110-m and front-end processing computers (front end). Computer) 111, front-end computer 111, and each processing request execution computer 110-1
To 110-m, and data storage devices 118-1 to 118-m connected to the processing request execution computers 110-1 to 110-m and storing data.
It consists of and. In FIG. 1, the number of front-end computers is one, but a plurality of computers may be provided in parallel. Further, the data storage device may be provided locally in only a part of all the processing request execution computers (there may be only one data storage device).

By using a plurality of processing request execution computers in this way, high-performance data processing by parallel processing can be realized. Further, the load of processing requests can be distributed to a plurality of processing request execution computers to handle high-load data processing.

The front-end computer 111 includes n (n is 1 or more) processing request generating units 10 for generating processing requests.
2-1 to 102-n are connected via a communication path 115,
The processing request execution computers 110-1 to 110-m that receive the processing requests sent from the processing request generation units 102-1 to 102-n and have the application programs 116-1 to 116-m to execute the processing requests. And has a processing request allocation unit 101 for instructing execution of the processing request.

As the processing request generation unit 102, various types such as a terminal device, a calculator, an automatic cash dispenser, etc. can be considered.

Each processing request execution computer 110-1 to 110
-M operates the application programs 116-1 to 116-m that execute processing requests, and the data on the data storage devices 118-1 to 118-m that are necessary to execute the application programs 116-1 to 116-m. Each has data management units 117-1 to 117-m.

Processing request execution computers 110-1 to 110-
One of m has a data placement management unit 120. This data arrangement management unit 120 is provided for each processing request execution computer 1
10-1 to 110-m internal data management section 117-1 to
117-m operates in cooperation with each other, receives information regarding data access from each of the data management units 117-1 to 117-m, calculates an optimum data arrangement based on the information, and calculates each according to the result. Processing request execution computer 1
Instruct 10-1 to 110-m to change the data arrangement. A more specific internal configuration of the data arrangement management unit 120 and the data management unit 117 will be described later.

In this embodiment, the data arrangement determining unit 1
Although 20 is arranged in one of the processing request execution computers 110,
Instead, one unit may be provided on the front-end computer 111, or one unit may be provided on a computer provided in addition to the front-end computer 111 and the processing request execution computer 110, or one may be provided on any one of these computers. May be.

When a plurality of data storage devices are used, the data is divided into data storage devices 118-1 to 118.
Stored in -m. When dividing the data, the data can be sorted by a key and stored in each computer, for example. In this case, fault tolerance can also be improved by creating a copy of some or all of the data and storing the copy in a data storage device different from the data storage device in which the original data is stored. Is.

A hard disk, a semiconductor memory or the like can be used as the data storage device 118. In addition, magnetic tape, floppy disk, CD-RO
It is also possible to use M, MO and the like.

When the data is distributed and arranged in the data storage device, it is possible to arrange the data by simply dividing the data without overlapping, but by providing redundancy to some or all of the data. By arranging the original data and its copy in different data storage devices, it is possible to improve the fault tolerance of the data processing device.

Next, the operation of the data processing apparatus shown in FIG. 1 will be described.

The processing request output by any one of the processing request generation units 102-1 to 102-n is sent to the processing request allocation unit 101 on the front end computer 111 through the communication path 115. The processing request allocation unit 101
From among the processing request execution computers 110-1 to 110-m, the optimum one for executing the processing request is selected and the processing is assigned.

The optimum process request execution computer for executing the process request is the process request execution computer that can execute the process request at the lowest cost. The cost can be represented, for example, by the time required to execute the processing request. Alternatively, the cost can be expressed by the number of communications required to execute the process.

As will be described later, the method of selecting the processing request execution computer for executing the processing request is "how much cost in the past, which processing request execution computer was executed by which processing request execution computer?" By providing a means for accumulating an execution history that represents the execution request, when a new processing request arrives, the processing request is compared with the execution history to determine which processing request execution computer should be executed at the lowest cost. be able to. In this method, the optimum processing request execution computer is selected based on the past execution history, so there is no guarantee that the optimum one will be selected, and it is probabilistic whether or not the optimum one will be selected. Become. The reason why such a probabilistic method can be used is that all the data can be transparently accessed from an application program on any processing request execution computer, as will be described later.

When the optimum processing request executing computer can be selected by a deterministic algorithm depending on which data is stored in the data storage device of which processing request executing computer, the algorithm for the processing request allocating device is used. There is also a method of incorporating as a program.

A more specific internal structure of the processing request allocation unit 101 will be described later.

Now, the processing request execution computers 110-1 to 110-1
The processing request assigned to 10-m corresponds to the application program 1 arranged on the processing request execution computer.
16 is executed. As will be described later, since all the data are transparently accessible from the application programs 116-1 to 116-m on any of the processing request execution computers 110-1 to 110-m, the application program 116- 1-116-
As m, which processing request execution computer 110-1 to 11
The same thing may be arranged in 0-m. Therefore, it is possible to perform parallel processing without changing the application program developed on the conventional single computer.

The data management units 117-1 to 117-m are
It receives a data operation request necessary for its execution from the application programs 116-1 to 116-m and executes it. At this time, if the data to be operated is not in the data storage device 118 of the own processing request execution computer 110, the data management unit 117 of the processing request execution computer 110 having the data is requested to perform the operation. If it is in the local computer 110, it is operated locally. In such a data operation, data to be operated is usually designated by a key. Whether a data operation is performed locally by having a table that records "what range of key value is in which processing request execution computer has the substance of the data" or data management unit of another processing request execution computer It is possible to determine whether to request 117. The processing method when operating locally is, for example, "Technical introduction to database systems,
Makoto Takizawa, Soft Research Center, ISBN4-9
15778-07-x ".

The data arrangement determination unit 120 is a data management unit 117 of each of the processing request execution computers 110-1 to 110-m.
From -1 to 117-m, statistical information including information "which data needs to be accessed to execute which processing request" and frequency information of the processing request is collected.
Therefore, the data allocation determination unit 120 may exist on any computer as long as it is a computer that is coupled to each processing request execution computer 110 through the coupling path 114. Although it exists in one of the processing request execution computers in the example of FIG. 1, it may be on the front-end computer 111, or a new computer may be connected to the connection path 114 and placed on it.

As will be described later, the data arrangement determining unit 120 determines how to arrange the data based on the statistical information collected from the respective data managing units 117-1 to 117-m. Optimal data allocation is determined by considering whether the load of 110-m is balanced and the cost required to execute each processing request can be reduced. By moving the data managed by each of the data management units 117-1 to 117-m based on the result, the optimum data arrangement is achieved. This data movement can be performed by the data allocation determination unit 120 instructing the data management unit 117 on each processing request execution computer 110. Further, an implementation method of manually moving the data according to the new data layout determined by the data layout determination unit 120 is also possible.

The timing at which the data arrangement determining unit 120 determines the optimum arrangement of data and rearranges the data is determined by a method in which the operator instructs the data arrangement determining unit 120 and by the data arrangement determining unit 120 in advance. From the statistical information, the method of performing at a predetermined time interval, the method of performing when the data allocation determining unit 120 collects a predetermined amount of statistical information, and the statistical information show a bias in the number of data access of each processing request execution computer 110. How to do it when
A method to perform when there is a bias in the load information of each processing request execution computer 110, a method to instruct the data allocation determination unit 120 to perform relocation when the processing request allocation computer 101 allocates a processing request to a processing request execution computer 110, etc. Various methods are possible.

By the automatic optimal allocation of data by the data allocation determining unit 120, the dynamic characteristics of the entire system such as the frequency of each processing request and the access data of each processing request can be tracked. It is possible to optimize the data layout. Further, according to the data allocation, the processing request allocation unit 101 allocates each processing request to the processing request execution computer 110 suitable for executing the processing request, so that new data allocation and its allocation are performed each time the data allocation is changed. The load of processing can be automatically distributed without giving a method for allocating processing requests based on it. Furthermore, the tuning of the entire data processing apparatus can be automated, and system management becomes easy.

Further, even if the optimum allocation strategy in the processing request allocation unit 101 is not clear, or even if the optimum distributed arrangement of data is not clear, it is possible to distribute the processing load by allocating the data in a distributed manner. Although there is a possibility that the operation will not be successful, it is guaranteed to operate as a data processing device.

In the example of the data processor of FIG. 1, the number of front-end computers 111 is one, but as shown in FIG.
A plurality of front-end computers 111-1 to 111-, each having processing request allocation units 101-1 to 101-l.
It is also possible to use l. A plurality of front end computers 111-1 to 111-l are connected to the communication path 115 and the connection path 11
If one front-end computer 111 fails, another front-end computer 1
By replacing with 11, the fault tolerance is improved.

Further, when the performance required for the processing of the processing request allocating unit cannot be obtained with one front-end computer,
The load can be distributed by using multiple units. In addition, the communication path accommodating a plurality of processing request generation units is divided into a plurality of
A method of connecting a plurality of front-end computers to different communication paths is also possible.

The data processing apparatus of this embodiment can be applied to online transaction processing and other online processing.

Here, after fixing the data arrangement, each processing request is executed by using a method of dynamically determining only to which processing request executing computer the processing request is allocated. Considering the case of dynamically deciding whether to allocate to a computer, it is not possible to allocate processing requests so that the data required by each processing request is in the processing request execution computer, so system performance Can not be fully exerted. In order to maximize the performance of the system, it is necessary for the system administrator to reallocate the data by deciding how to allocate the data based on the newly determined allocation method of processing requests. is there. Therefore, it takes a lot of time to manage the system.

On the other hand, when the business computer that can execute the processing request has been determined in advance, a method for determining the data arrangement based on the dynamic characteristics, for example, "Data Plac"
element In Bubba, George Cop
eland et. al. , SIGMOD, Vol. 1
7, No. 3, 1988 ”, the access to the file is monitored, and even if the arrangement of the data is changed based on the dynamic characteristic of the data access by the method of manually moving the file, the initial processing request is required. Since the data allocation used when deciding the allocation method is different, the data required by each processing request is not necessarily in the processing request execution computer that executes the processing request, so the system performance is sufficient. Can not be demonstrated to. In order to make full use of the performance of the system, it is necessary to decide which processing request execution computer to newly allocate each processing request in consideration of the new data arrangement and the occurrence frequency of each processing request. Therefore, system management is extremely troublesome.

In the present embodiment (or each of the embodiments described later), a processing request allocating means for selecting an optimum computer for executing the processing request, that is, a computer capable of executing the processing request at the lowest cost and allocating the processing is provided. By providing a data allocation determining unit that performs automatic optimum allocation of data and forming a feedback loop of processing request allocation and data relocation according to data allocation, both data allocation and processing request allocation can be performed. It makes it possible to optimize dynamically.

(Second Embodiment) FIG. 3 shows the configuration of the second embodiment of the present invention.

The difference between the data processing device shown in FIG. 3 and the data processing device of FIG. 1 lies in the connection system of the data storage device 118. In the data processing device of FIG. 3, all the processing request execution computers 110-1 to 110-m share one data storage device 118. In this method, each of the processing request execution computers 110-1 to 110-m can access any data in the data storage device 118.

Each processing request execution computer 110-1 to 110-1 in FIG.
Data management units 117-1 to 117-m in 110-m
First, it checks whether the other data management unit 117 is accessing the data that the user wants to access. If the other data management unit 117 is not accessing, the data is first read into the area prepared in the own data management unit 117, and the necessary operation is performed on it. The data management unit 117 can return the data to the data storage device 118 after performing the necessary processing. However, if there is a high possibility that the same data will be accessed again, the data management unit 117 does not write back the data. It is also possible to keep it held in the area within the data management unit 117. In this case, the other data management unit 11
7 cannot directly access the data, and requests the data management unit 117 for processing.

Therefore, in the case of the present embodiment, the data allocation determining unit 120 does not ask "where in the data storage device 118 the data should be allocated" but "the data management unit 117 of which processing request executing computer". However, which data may be retained in the area within the data management unit 117? ”Is determined.

In this embodiment, the system of the present invention is not limited to the data non-shared data processing device as shown in FIG.
It is shown that it is also applicable to a data sharing type or disk sharing type data processing device.

(Third Embodiment) FIG. 4 shows the configuration of the third embodiment of the present invention.

In the data processing apparatus shown in FIG. 4, the processing request allocating unit 101, which was on the front-end computer 111 in the data processing apparatus of FIG. 1, is provided on each of the processing request execution computers 110-1 to 110-m. It was done like this.

Each processing request execution computer 110-1 to 110
-M processing request allocation units 101-1 to 101-m
The plurality of processing request generation units 102 are connected via the communication paths 115-1 to 115-m, respectively. As in the example of FIG. 4, each processing request execution computer 110-1 to 110-m is connected to different communication paths 115-1 to 115-m, but instead of all processing request execution computers or some of them. May be connected to the same communication path (there may be only one communication path).

As in the case of FIG. 3, the data storage device 118 may be shared by all the processing request execution computers 110-1 to 110-m.

(Specific Examples of Data Placement Determining Unit 120 and Data Managing Unit 117 of First to Third Embodiments) Next, the first
~ Specific examples of the data arrangement determination unit 120 and the data management unit 117 of the third embodiment will be shown. FIG. 5 shows the data management unit 1.
20 is a schematic configuration diagram of 20 and a data arrangement determination unit 117. FIG. As described above, the data allocation determination unit 120 is composed of the correlation information storage table 202, the correlation information management unit 203, and the data allocation determination unit 204, and the data management unit 117 includes the access request reception unit 201 and the data storage information storage unit 205. , A data arrangement changing unit 206 and a data access unit 207. In addition, in FIG. 5, the coupling path 114 and the communication path 1
15, processing request allocation unit 101, processing request generation unit 10
2 is omitted. In addition, application program 1
Although 16 may be one or plural, only one is shown in the drawing. The data storage device 118 may be locally connected to the processing request execution computer 110 or may be shared by a plurality of processing request execution computers 110.

In the following, for convenience of explanation, the data management unit 1
20 and the data arrangement determination unit 117 will be referred to as a data management device.

In general, this data management apparatus first receives a processing request from a computer or the like that requires access to data stored in a distributed manner in a plurality of different data storage areas. Is received, and the mutual correlation information of the data indicating that the specific plurality of data have been accessed together in a series of processing is updated.

Here, the series of processes means processes performed from a predetermined start time to a predetermined end time. For example, it is possible to treat a process as a series of processes from its creation to its termination, or when that process performs different processes according to input, each process can be defined as a series of processes. . Alternatively, it is also possible to divide the operation of the process into a plurality of parts and define each as a series of processing. Furthermore, it is possible to define a process executed as a transaction as a series of processes. Further, it is possible to define the same except for specific processing. Alternatively, all the processes executed between the predetermined start time and end time can be set as a series of processes.

Each time the above processing request is received, the operation of updating the mutual correlation information of data is repeated, and after sufficiently acquiring the information on the correlation information in the system, the correlation information and the data storage area in which the data is stored are changed. Based on the data storage information shown, a new data arrangement in the data storage area that matches the characteristics of the system is determined.

Therefore, according to such a data management method, when the data stored in the plurality of data storage areas are frequently referred / updated together in a series of processes, the correlation is automatically calculated. It is possible to make a determination, and the data arrangement can be automatically changed so that such data is stored in the same data storage area.

Further, since the information necessary for arranging the data is automatically acquired, it is not necessary for the system administrator to analyze the processing on the system to determine the data arrangement, and the execution time of each process is shortened. The performance of the system can be brought out easily.

This data management device receives a data access request from the application program 116,
Correlation information is created and updated when the history of access requests meets a predetermined condition. After acquiring a sufficient amount of correlation information in this way, new data arrangement information is created based on this and the data storage information.

The data management device also stores data on one or more data storage devices 118. Here, the data management device manages a plurality of data storage areas, but it is possible to allocate them to each data storage device one by one, or to allocate them to each data storage device. Is. It is also possible to collectively handle a plurality of data storage devices as one data storage area. The number of data storage areas in the data storage device and the size of each area are arbitrary, and the number of data storage areas between the data storage devices and the size of the areas can be different from each other. .

The data storage device 118 managed by the data management device has a function of storing data, and data at an arbitrary position in the data storage device 118 can be read or rewritten. Any material can be used, and a semiconductor memory, a hard disk, a floppy disk, a magnetic tape, or the like can be used.

The data management device can be configured as an independent device, or can be realized as software or hardware as a function of the computer.

The operation of the data management apparatus shown in FIG. 5 will be described below.

The data access request generated from the processing request executed in the application program 116 is received by the access request receiving unit 201. The access request is made up of data required to perform data access and data required to generate correlation information. The access request reception unit 201 passes the received data access request to the correlation information management unit 203 and the data access unit 207.

The data access unit 207 reads or updates data from the data storage device 118 based on the received access request, and returns the result to the application program 116 which is the source of the data access request.

The correlation information management unit 203 creates the correlation information from the newly received access request and compares it with the correlation information stored in the correlation information storage table 202. If the correlation information is already registered, the correlation information is stored. Information is updated, and if it is unregistered correlation information, the correlation information is registered in the correlation information storage table 202. At this time, those that have not been updated for a certain period of time or those that do not meet the predetermined conditions are deleted from the correlation information storage table 2.
The correlation information management unit 203 repeats such updating of the correlation information every time there is an access request, and passes the correlation information to the data arrangement determination unit 204 when sufficient correlation information is obtained.

The data arrangement determination unit 204 determines a new data arrangement based on the received correlation information and the data storage information stored in the data storage information storage unit 205.
After deciding a new data arrangement, the data arrangement determining unit 204 transfers it to the data arrangement changing unit 206 and the data storage information storage unit 205.

When changing the arrangement, the data arrangement changing unit 205 first notifies the data access unit 207 of which data location and how. Upon receiving this notification, the data access unit 207 temporarily prohibits update requests for those data. Next, the data arrangement changing unit 206 exchanges the data based on the received data arrangement and, when the exchange is completed, notifies the data storage information storage unit 205 and the data access unit 207 of that fact.

When the data storage information storage unit 205 receives from the data layout update unit 206 that the data replacement has been completed, the data storage information storage unit 205 updates the data storage information based on the information about the data layout previously received from the data layout determination unit 204. . Further, when the data access unit 207 receives the notification that the data exchange is completed, the data access unit 207 changes the management information for the data access due to the change of the data arrangement as necessary, and then, Enables update requests.

(Specific Example of Processing Request Allocating Unit 101 of First to Third Embodiments) Next, a specific example of the processing request allocating unit 101 of the first to third embodiments will be described.

FIG. 6 shows a structural example of the processing request allocation unit 101 of this embodiment. Further, another configuration example is shown in FIG. In order to distinguish the configurations, the processing request allocation unit in FIG. 6 is represented by 1 and the processing request allocation unit in FIG. 7 is represented by 11.

The processing request allocating units 1 and 11 receive the processing requests generated by the one or more processing request generating units 2-1 to 2-n, and one or more processing request execution computers capable of executing the processing requests. A computer optimal for execution is selected from 10-1 to 10-m, and the received processing request is transferred to the selected computer.

First, FIG. 6 shows an example of the structure of the processing request allocation unit.
Follow the instructions below. The processing request received from any one of the processing request generation units 2 is received by the processing request receiving unit 3. The processing request receiving unit 3 passes the received processing request to the processing request transmitting unit 6 and the characteristic parameter extracting unit 4. The characteristic parameter extraction unit 4 extracts the characteristic parameter from the passed processing request, passes it to the execution computer selection unit 5, and passes the data to the execution history information storage unit 8 to record it as the characteristic parameter of the corresponding processing request. Keep it. Execution computer selection section 5
Selects the optimum computer that executes the corresponding processing request by comparing the value of the characteristic parameter passed from the characteristic parameter extraction unit 4 with the past execution history information recorded in the execution history information storage unit 8. To do. The selected computer is transmitted to the processing request transmission unit 6 and the execution history information storage unit 8 is notified so that it is recorded as the computer that executed the corresponding processing request. The processing request transmitting unit 6 completes the processing request allocation processing by transmitting the processing request passed from the processing request receiving unit 3 to the computer notified from the executing computer selecting unit 5. The execution history information management unit 7 manages the execution history information recorded in the execution history information storage unit 8,
It deletes old execution history information that exceeds a predetermined time or the number of processing times. The execution computer 10 receives the processing request, executes the processing request, and transmits the cost required for the execution to the execution cost detection unit 9 of the processing request allocation unit. The execution cost detecting unit 9 records the execution cost in the processing request execution history information of the corresponding processing request by transmitting the received execution cost to the execution history information storage unit 8. When it is necessary for the processing request execution computer 10 to transmit the execution result of the processing request to the processing request generation unit 2, the processing request execution computer 10 directly transmits the processing result to the processing request generation unit using an appropriate communication means.

Another example of the configuration of the processing request allocation section will be described with reference to FIG. The configuration example of FIG. 7 is the same as the configuration example of FIG. 6 until the processing request is transmitted to the processing request execution computer.
The processing request execution computer 10 that has received the processing request executes the processing, but unlike the configuration example of FIG. 6, does not explicitly inform the processing request allocating unit of the cost required for the processing. The processing request execution computer 10 sends back the processing result of executing the processing request to the processing request allocation unit 11. This processing result is the processing result receiving unit 12 of the processing request allocating unit.
Is received and transmitted to the processing result transmission unit 13, and also notifies the execution cost detection unit 9 that the processing result has been received. The processing result transmission unit 13 that has received the processing result sends the processing result back to the processing request generation unit 2 that has generated the corresponding processing request. The execution cost detection unit 9 that has been notified that the processing result has been received obtains the cost required to execute the corresponding processing request and informs the execution cost to the execution history information storage unit 8 of the corresponding processing request. The execution cost is stored in the processing request execution history information. As a means for obtaining the execution cost, a method in which the cost required for execution is the difference between the time when a processing request is received and the time when the processing result corresponding to the processing request is received, or the execution of the corresponding processing request is overtaken It is possible to use a method in which the execution cost is the number of processing requests completed.

In the configuration example shown in FIGS. 6 and 7, FIG.
As shown in (a), the characteristic parameters extracted from the received processing request are recorded in the execution history information storage unit 8, and the execution computer selection unit 5 stores the characteristic parameters of the past execution history recorded in the execution history information storage unit. The execution computer is selected. On the other hand, the processing request itself may be recorded in the execution history information storage unit 8 instead of the characteristic parameter. In this case, as shown in FIG. 8B, the processing request received by the processing request receiving unit 3 is directly recorded in the execution history information storage unit 8, and the execution computer selecting unit 5 records the characteristic parameter of the past processing request execution history information. When looking at, the characteristic parameter is extracted again through the characteristic parameter extraction unit 4. In the configuration example of the processing request allocation unit shown in FIG. 9, the configuration of the processing request allocation unit of FIG. 6 is modified so that the processing request is recorded in the execution history information storage unit 8 as it is. FIG.
In the configuration example of the processing request allocating unit shown in FIG. 7, the configuration of the processing request allocating unit in FIG. 7 is modified so that the processing request is recorded in the execution history information storage unit 8 as it is.

The operation principle of an embodiment of the processing request allocation unit will be described below.

In the processing request allocating unit of the present invention, when the processing request allocating unit receives a new processing request, first, the characteristic parameter thereof is extracted. The characteristic parameter of the processing request is not one for one processing request, but is generally a set of a plurality of values.

Next, the characteristic parameter of the new processing request,
It compares the characteristic parameters of the stored execution history information of past processing requests, and for each computer and for each characteristic parameter, obtains the execution history information having the value of the characteristic parameter closest to the characteristic parameter of the new processing request. Pick out.

That is, when the number of computers capable of executing the corresponding processing request is M and the characteristic parameter consists of a set of N values, the characteristic parameter of the new processing request is X [n].
(0 ≦ n <N), M × N pieces of execution history information H
[M, n] (0 ≦ m <M, 0 ≦ n <N) is selected so as to satisfy the following condition.

[0103]

[Equation 1] Here, A is a set of all currently stored execution history information, and H [m, n] [p] is execution history information H [m, n].
The p-th value of the set of characteristic parameter values possessed by
(F) means the number of the computer on which the processing request recorded in the execution history information F was executed.

From among the M × N pieces of execution history information H [m, n] thus selected, first, a computer P [n] (0 ≦ n <is a candidate for executing a processing request for each characteristic parameter.
N) is selected so as to satisfy the following condition. Where cos
t (F) means the cost required to execute the processing request recorded in the execution history information F.

[0105]

[Equation 2] That is, from the execution history information H [m, n], the execution history information with the smallest cost is selected for each characteristic parameter, and the computer that has executed the processing request corresponding to that execution history information executes the characteristic parameters. Let it be the candidate computer P [n].
Here, the execution history information H [P selected as the execution candidate
If the execution cost of [n], n] is particularly smaller than the execution cost of other execution history information having the same characteristic parameter, the execution history information is selected as an execution candidate. There is no. However, H [P [n],
[n] has no significant difference in the execution cost from the costs of other execution history information, in other words, if there is no execution history information with a particularly low cost, simply set the minimum cost. Even if you select the execution history information you have,
It is unlikely that it is actually the best candidate for execution. Therefore, in order to remove the execution history information having such a low possibility from the execution candidate computer, for example, the execution candidate computer P [n] that does not satisfy the following condition is removed from the execution candidate computer.

[0106]

(Equation 3) That is, the execution history information H [m, which has the same characteristic parameter
There is a significant difference when the average value of the costs of [n] is larger than the cost of the execution history information H [P [n], n] that is the execution candidate by a predetermined value T times or more. Thinking,
If not, remove it from the execution candidate computer. T is 1.1, for example
Has a value such as. According to the algorithm, if there is no execution candidate computer, the value of P [n] such as -1 is not valid as a computer number.

The N execution candidate computers P selected in this way
From [N], a computer to be actually executed is selected. At this time, the weight information W [N] of each characteristic parameter is used. W
Each element W [k] of [N] represents the strength of correlation between the value of the kth characteristic parameter of the processing request and the computer number executing the processing request. Let PP [N] and WW [N] be the corresponding elements of P [N] and W [N] arranged in descending order of W [N] value.
The procedure for selecting a computer to actually execute from W [N] is C
The description in language is as follows.

[0108]

[Equation 4] Here, it is assumed that random () is a function that generates a random number that takes a real value between 0 and 1. A variable for substituting the number of the computer to be executed is pickednode,
The initial value is -1, which is not any computer number. Random numbers are generated in descending order of weight, and if the value is smaller than the weight, the computer is selected. Otherwise, the candidate with the next highest weight is similarly selected. If the candidates are not decided until the end, select a computer to execute by an appropriate method. In the example of the procedure shown here, one of M computers is randomly selected.

An embodiment in which the present invention is applied to a processing request allocating section by a remote procedure call (RPC) will be described below.

FIG. 11 shows the structure of the processing request allocation unit. The configuration of FIG. 11 is obtained by adding a process request table (storage unit) 26 and a process request table management unit 25 to the configuration of FIG. The processing request allocating unit of FIG. 10 has a configuration for allocating a single type of processing request, and the configuration of FIG. 11 is capable of handling a plurality of types of processing requests.

The processing request receiving unit 3 receives the processing request packet transmitted from the processing request generating unit 2 through the coupling path 14. For example, when Ethernet is used as the connection path, a packet that arrives at a specific port of the UDP protocol is received. The packet is data having a byte string structure as shown in FIG. 12A, but the RPC processing request packet interprets this byte string as data having a structure as shown in FIG. 12B, for example. That is, the head of the packet contains data indicating the type of processing request. The example in FIG. 12B indicates that the type of processing is called proc1. Actually, the unique identification number corresponding to proc1 is recorded as a numerical value occupying the first 4 bytes of the packet. After that, the required number of arguments accompanying the processing request are arranged. In the example of FIG. 12 (b),
Three 4-byte length data represented by x, y, and z are arranged. The RPC packet structure has many structures other than those shown here, but the method of the present invention can be applied to packets of any structure.

Since which part of the RPC packet contains the request type is fixed, the processing request receiving unit 3 which has received the packet can extract the request type from the packet, and the processing request table Tell 26. Here, the processing request receiving unit can also be configured to receive processing request packets that arrive at a plurality of ports.
In this case, for each port, a processing request table that records information about processing requests that arrive at that port is created, and when the processing request receiving unit receives a packet, the processing request table is created in the processing request table corresponding to the port that received the packet. A configuration that conveys the type of request is also possible.

As shown in FIG. 13, the processing request table 26 is a table for recording the correspondence relationship of five types of information of request type, argument information, weight information, execution computer information, and execution history information pointer.

The request type is a value indicating the type of request at the head of each processing request packet. In the example of FIG. 13, pro
This corresponds to c1 and proc2, but actually proc
Numerical values corresponding to 1 and proc2 are recorded. The argument information records the data type of the argument for each request type. For example, in the example of FIG. 13, there are three arguments of proc1, and their data types are int (integer type) and strin in order.
g (character string type) and float (floating point type) are shown. Actually, int and string respectively
It is realized as an array of numerical values assigned so as to correspond to g and float.

The argument information shown in the example of FIG. 13 is in
Only a type that does not have a structure such as t or float, but in the case of the structure shown in FIG.
As shown in FIG. 4B, a numerical value indicating a structure (described as “struct” in the figure), the number of members of the structure (2 in the example of the figure), and a numerical value corresponding to the data type of the member ( In the example of the figure, it is expressed as int and char). In the case of the array as shown in FIG. 14C, a numerical value indicating the array as shown in FIG. 14D (described as array in the drawing) and a numerical value indicating the number of elements (example of the drawing) In 10), it is expressed as a set of numerical values (described as int in the figure) corresponding to the data type of the element. The structure or array may have the constituent elements of the structure or array.

In many RPC systems, FIG.
The format of the packet actually used for communication is determined from the interface description as shown in. This processing is usually performed by a program called an RPC compiler. As shown in FIG. 15, the RPC compiler receives an RPC interface description as an input, a client stub for transmitting a processing request in a packet format corresponding to the interface description, a server stub for receiving the request, and a client and a server. Outputs a header file that describes information commonly used in. Rpcg for ONC RPC
A program called idl in en and DCERPC corresponds to the RPC compiler here.

In this embodiment, as shown in FIG.
In addition to the above, the compiler also outputs packet information. As shown in FIG. 16, the packet information is composed of data to be registered in the request type argument information of the processing request table. That is, when the interface description is given as shown in FIG. 14E, a set of proc1 is generated as the request type and a set of int, string, and float is generated as the argument information. As mentioned earlier, actually proc1,
Numerical values that can be uniquely identified are assigned to int, string, and float. FIG.
In the configuration diagram of FIG. 1, the processing request table management unit 25 reads the packet information and registers the request type and argument information in the packet information in the processing request table 26.

The weight information in the processing request table of FIG. 13 is a numerical value indicating the weight of each argument of the request type. When predicting the processing request execution computer to which the processing request should be allocated,
It shows which characteristic parameter should be noted. The weight information is an array having elements that have a one-to-one correspondence with each element of the characteristic parameter. In the example of proc1 of FIG. 13, the number of arguments is three, but the weight information is a vector of four elements. The feature parameter is a numerical value obtained by changing the bit positions of some of the arguments in addition to all the arguments. This is because it also includes. Here, with respect to the integer type argument, the one whose bit position is changed is also added to the characteristic parameter, so that the weight information is a vector of four elements. The method of extracting the characteristic parameter from the RPC packet, the method of changing the bit position, and which characteristic parameter the characteristic parameter is generated by performing the bit position conversion will be described later.

The execution computer information of the processing request table of FIG.
It is a list of processing request execution computers capable of executing each request type, and actually has information such as the network address of the computer and the port number on the computer necessary for transferring the processing request. For example, the first execution computer information of proc1 in FIG. 13 indicates that the network address of the computer is 62.31 and the port number is 70.

The execution history information pointer of the processing request table 26 in FIG. 13 points to a group of execution history information stored in the execution history information storage unit 8 that belongs to each request type. As shown in FIG. 17, the execution history information is necessary to send back the contents of the processing request, the number of the computer that executed the processing request, the cost required for the execution (for example, the time required for the execution here), and the processing result. The information (eg, network address and port number here) of the processing request generation unit. Instead of recording the processing request as it is, the characteristic parameter extracted from the processing request or only the characteristic parameter having a large weight used for the prediction of the execution computer may be extracted.

FIG. 18 shows the flow of information centered on the processing request table. FIG. 18 shows information extracted from FIG. 11 that is necessary for determining the processing request execution computer and is transferred between the modules therein. The operation of this embodiment will be described below with reference to FIGS. 11 and 18.

The information recorded in the processing request table 26 is
As shown in FIG. 15, the RPC compiler generates the packet information. This packet information is generated in the form of a file or the like, and the processing request table management unit 25 of the processing request allocation unit 25
Is read by and registered in the processing request table 26. At this time, the initial value of the weight information in the processing request table 26 may be given by the programmer or the system administrator, or all elements may be set to constant initial values (for example, 0.9). The execution computer information may be given by a programmer or a system administrator, or there is an implementation method in which a computer capable of executing the request is checked and registered by a name server or the like. An implementation method is also possible in which each processing request execution computer 10 informs the processing request assigning section of the type of request that can be executed by itself, and fills in the processing request table 26 based on that information. The initial value of the execution history information pointer of the process request table 26 indicates the execution history information of the corresponding request type in the execution history information storage unit 8 when it remains, and the empty execution when it does not remain. Point to a set of history information.

When the processing request receiving unit 3 receives the processing request RPC packet from the processing request generation unit 2, the processing request packet is passed to the first characteristic parameter extraction unit 28 and the processing request transmission unit 6 in the next stage. Further, the processing request and the processing request generation unit that outputs the processing request are registered in the execution history information storage unit 8 as new execution history information. Further, the information indicating the type of the processing request at the head of the received processing request packet is extracted and passed to the processing request table 26.

The processing request table 26 is obtained by pulling the processing request table using the type of request received from the processing request receiving unit 3 as a key.
The resulting argument information is stored in the characteristic parameter extraction unit having two pieces, the weight information is stored in the execution computer selection unit 5, the execution computer information is stored in the processing request transmission unit 6, and the execution history information pointer is stored in the execution history information storage unit 8. hand over.

The first characteristic parameter extracting unit 28 which has received the processing request packet and the argument information from the processing request receiving unit 3 and the processing request table 26 respectively extracts the characteristic parameter of the received processing request, and the execution computer selecting unit 5 Pass to.

The processing request packet is a byte string as shown in FIG. At the beginning thereof is data indicating the type of request. For example, if the first word (4 bytes) indicates the type of request, in the example of FIG.
2 (174 in hexadecimal) is a value representing the type of request.
Here, 372 represents proc1. The data indicating the request type is followed by a plurality of arguments associated with the request. In the case of proc1, the first argument is an integer type (in
t), the second argument is a character string type (string), and the third argument is a real number type (float). There are various representation methods on the network packet of each data type, and the present invention can be applied to any representation method. Here, an integer and a real number are represented as 4-byte data, and a character string is composed of a character string length represented by a 4-byte integer, and a character sequence represented by 1 byte following the character string length. For example, in the example of FIG. 19, the first argument is an integer value 67.
00, the second argument is a character string "abcdefg", and the third argument is a real value 12.5.

Argument information, which is another input data of the characteristic parameter extraction unit, is an array of information indicating the type of each argument for each request type. This argument information is an example of the information of the data structure of the processing request, the basic data unit is extracted from the processing request by referring to this information, and each basic data unit is used as the characteristic parameter. For example, as shown in FIG. 20, if the integer type is 1, the real number type is 2, and the character string type is 3, the argument information of proc1 is a vector having three values 1, 3, and 2 ( Or array)
become.

The characteristic parameter extraction unit that has received the above two types of information operates as shown in FIG. 21, extracts characteristic parameters, and passes them to the execution computer selection unit 5. That is, the characteristic parameter extraction unit inputs the argument portion (P in FIG. 21) and argument information (T in FIG. 21) of the processing request packet, cuts out the argument portion of the processing request packet corresponding to each element of the argument information in order, and The feature parameter values for are sequentially output. For example, when the data type of the argument described in the argument information is an integer type (1 in the example of FIG. 21), 4-byte data is cut out from the argument part of the processing request packet, and it is set to 1
It is output as the value of the characteristic parameter as it is as the integer value of the word. Although not shown in the example of FIG. 21, data of the required number of bytes is similarly cut out for a character type and an enumeration type, read as an integer value, and the value is output as a characteristic parameter. In the case of the real number type (2 in the example of FIG. 21), similarly, 4-byte data is cut out, interpreted as a single precision real value, and the value is output as a characteristic parameter. The same applies to double precision.

In the case of the character string type (3 in the example of FIG. 21), several implementation methods can be used. In the example of FIG. 21, the first 4 bytes of the character string (when it is less than 4 bytes, 0 or blank characters are supplemented) are read as an integer value of 1 word, and the value is output as a characteristic parameter. In this way, the character string can be trimmed into a character string of a predetermined length and read as a numerical value to be a characteristic parameter. For trimming the character strings, a method of cutting out the fixed length character string at the beginning, a method of cutting out the fixed length character string at the end,
There is a method in which a fixed length at the beginning and a fixed length at the end are cut out and connected. There is also a method in which not only the character string cut out in this way, but also the length of the character string is added as a characteristic parameter. There is also a method in which the character string is not digitized, but is passed as it is to the execution computer selection unit 5 as a characteristic parameter, and the execution computer selection unit 5 selects the execution computer by comparing the character strings in lexicographic order.

The data type described in the argument information is shown in FIG.
Although it does not appear in the example of 1, there is also complex type data such as an array or a structure. In the case of a fixed-length array, the simplest implementation method is to take out the characteristic parameters of all the elements and make them the characteristic parameters of the array. In the case of a fixed-length or variable-length array, as in the case of the character string type, a method that uses a fixed number at the beginning, a fixed number at the end, or a combination of a fixed number at the beginning and a fixed number at the end as a characteristic parameter There is also. In the case of a variable length array, the size of the array can be added to the feature parameter.

In the case of a structure, the characteristic parameters of the respective constituent elements are sequentially taken out and used as the characteristic parameters of the structure. The characteristic parameters can be extracted by the same procedure when combined like an array of structures or an array of character strings.

In this embodiment, for some of the characteristic parameters, the values obtained by exchanging the bit positions of the values are also added to the characteristic parameters. By adding the bit positions interchanged to the feature parameter, for example, if there is a correlation between the value of the higher-order bit of the argument and the computer suitable for executing the processing request, it is possible to execute it on the same computer. The value of the characteristic parameter of the desired processing request can be made close to each other, and the judgment ability of the execution computer selection unit 5 can be enhanced. In addition, the bit positions of the characteristic parameters of the arguments having a correlation with the execution computer are replaced with each other to become a random characteristic parameter having no correlation with the execution computer. On the contrary, when the bit positions are exchanged and the execution computer has a correlation, the feature parameter before the exchange of the bit positions is a random feature parameter having no correlation with the execution computer. Therefore, a random component that serves as a reference when comparing the correlation between the value of each characteristic parameter and the execution computer always exists in the characteristic parameter. Due to the existence of such random components, all the characteristic parameters and the execution computer have a similar correlation, and whether all the characteristic parameters have a uniform correlation with the execution computer, or have a uniform correlation. It is possible to eliminate the situation where it is impossible to judge whether or not there is.

There are various methods for exchanging bit positions, and the present invention can be applied to any exchanging method.
An example of a method that is useful as a method for exchanging bit positions is shown in FIG.
As shown in 2, the bit position is inverted (bit reverse). That is, the upper bits and the lower bits of the characteristic parameter are inverted, and the bit string obtained as a result is read as a numerical value to be the value of the characteristic parameter.

The method of deciding the argument in which the bit position is converted is added to the characteristic parameter is a method of adding the bit position of all the characteristic parameters obtained from the argument to the characteristic parameter, and a predetermined specific data. There is a method of adding the bit position reversed with respect to the type to the characteristic parameter, a method of adding the bit position reversed with respect to the argument specified by the programmer or the system administrator to the characteristic parameter, and the like. In the example of FIG. 21, a method is used in which the bit position is inverted with respect to the integer type argument and is added to the feature parameter.

In order to obtain the effect of adding a random component as a characteristic parameter, an implementation method in which a random number is generated and used as a characteristic parameter is also possible.

The processing request table 26 stores the argument information and the execution history information.
The second characteristic parameter extraction unit 29 receiving from the execution history information storage unit 8 extracts the characteristic parameters in the same procedure as the first characteristic parameter extraction unit described above. The execution history information storage unit 8 sequentially sends the plurality of execution history information items belonging to the specific request type indicated by the execution history information pointer received from the processing request table 26 to the second characteristic parameter extraction unit 29. The second characteristic parameter extraction unit 29, which has received the execution history information, extracts the characteristic parameter from the processing request recorded therein in the same manner as the first characteristic parameter extraction unit 28, and extracts it from the execution computer selection unit. Pass to 5. At this time, the execution computer and the execution cost recorded in the execution history information of the processing request are also passed to the execution computer selection unit 5.

The execution computer selection unit 5 has a structure as shown in FIG. The execution computer selection unit 5 receives the characteristic parameter of the newly arrived processing request from the first characteristic parameter extraction unit, and from the execution history information having the same request type executed in the past, from the second characteristic parameter extraction unit. The extracted characteristic parameter, execution computer, and execution cost are sequentially received and passed to the approximate cost selection unit 30.

The approximate cost selection unit 30 uses the execution history information of the past execution by the execution computer for one of the plurality of characteristic parameters and one of the plurality of computers that executed the same type of processing request in the past. The execution cost of the one having the closest characteristic parameter value is selected. That is, the approximate cost selection unit 30 in charge of the i-th feature parameter and the j-th execution computer performs the operation as shown in FIG. The value of the i-th characteristic parameter of the characteristic parameters of the new processing request is P, and the set of execution history information of the processing requests of the same type executed in the past by the j-th execution computer is H.
For each execution history information h in H, the value of the i-th characteristic parameter of h extracted by the second characteristic parameter extraction unit is compared with P, and the difference of execution history information having the smallest difference is calculated. The execution cost is output as the approximate cost C.
In the method shown in FIG. 24, when there is no processing request executed in the past by the corresponding execution computer, 0 is output as the execution cost. When the method shown in FIG. 25 is used, ∞ (largest execution cost) is output when there is no processing request executed in the past. When the execution cost is 0, the processing request is preferentially assigned to the execution computer, and when the execution cost is ∞, the processing request cannot be assigned to the execution computer. In the method shown in FIGS. 24 and 25, the comparison of the characteristic parameter values is performed arithmetically. A comparison method such as Hamming distance can be used for this comparison. It is also possible to use a lexicographical order-based comparison for character strings. A method using a plurality of such comparison methods in combination is also possible.

The execution computer selection unit 5 uses not only the implementation method in which all the characteristic parameters are used for execution computer selection, but also the implementation method in which only a fixed number of characteristic parameters at the head are used, and only the fixed number of characteristic parameters having large weight information. There is also an implementation method, or an implementation method in which only the characteristic parameters whose weights are larger than a certain value are used.

The approximate costs obtained by the approximate cost selecting section 30 are collected in the minimum cost calculator selecting section 31 for each characteristic parameter. The minimum cost computer selection unit 31 selects an execution computer that gives the lowest approximate cost from these approximate costs, and outputs it as the minimum cost computer number N regarding the characteristic parameter. Minimum cost calculator selection unit 31
26, as shown in FIG. 26, a part (S4 to S7) for selecting an execution computer number that gives the minimum approximate cost and a part (S8) for determining whether or not the minimum property is meaningful.
~ S10). Assuming that the approximate cost passed from the approximate cost selecting unit 30 in charge of the kth execution computer to the minimum cost computer selecting unit 31 in charge of the jth feature parameter is Ik, steps S4 to S7 in FIG.
Are approximate costs I1, I2, ...
The one with the lowest cost is selected from Im (m is the number of execution computers), and the number of the execution computer that gives the cost is output as the execution candidate computer number N. Then, it is judged whether or not it has a meaningful minimum property by whether or not the minimum approximation cost is particularly smaller than the other approximation costs. Therefore, first, step S4
From S7 to S7, the total S from I1 to Im is calculated, and the average value S / m obtained by dividing it by m and the minimum approximate cost C of the selected execution computer N are multiplied by a predetermined constant (for example, 1.1. If the S / m is larger, N is output as the minimum cost calculator for the jth feature parameter. If not, it is determined that the approximate cost C of the execution computer N does not have significant minimality, and the computer number -1 which means that no execution computer is selected is output.

Based on the execution candidate computers selected for each characteristic parameter by the minimum cost computer selecting unit 31 and the weight information of each characteristic parameter passed from the processing request table 26, the computer number selecting unit 32 makes a processing request. Select the execution computer number to be assigned. As shown in FIG. 27, the computer number selection unit 32 first sorts the sets of execution candidate computers and the weights of the feature parameters for each feature parameter in order from the item with the largest weight by the weight order sort unit 33. As shown in FIG. 27, the output is put into the probabilistic selection unit 34 in order from the heaviest one, and the execution computer number is selected.
As shown in FIG. 28, the probabilistic selection unit 34 sets the execution candidate computer number C, the weight W of the characteristic parameter that has selected the computer number, and the value between 0.0 and 1.0 generated by the random number generation unit 35. The random number R to be taken and the execution computer number P selected by the stochastic selection unit 34 in the previous stage are input, and the execution candidate computer number N determined in this stage is output. The most upstream stochastic selection unit 34
In P of 1, -1 indicating that no execution computer is selected is input. As shown in FIG. 28, the operation of the probabilistic selection unit 34 is the computer number when the execution computer is determined in the previous stage, and the candidate computer number C when the random number R is smaller than the weight W otherwise. If none of the above, the output N is -1, which indicates that no computer is selected. A stochastic selection unit 34 having a weight of 1.0 and a candidate computer number which is the selection result of the arbitrary computer selection unit 36 is provided in the most downstream, in case no execution computer is selected for any feature parameter.

The arbitrary computer selection unit 36 has a function of selecting an arbitrary computer when the probabilistic selection unit 34 for any feature parameter does not select the execution computer. Figure 2
As shown in FIG. 7, when the arbitrary computer selection unit 36 randomly determines an execution computer by using a random number, random allocation is performed. The arbitrary computer selection unit 36 predicts the load of each processing request execution computer based on the bias allocated to each execution computer so far, and the execution computer whose load is predicted to be small, that is, the processing request allocated in the past. Of the execution computers with less processing requests and processing requests that are currently allocated and processed, and know the load information of each execution computer,
There is also an implementation method based on which an execution computer with a small load is preferentially selected.

The execution computer number thus selected is sent to the processing request transmission unit 6. The processing request transmitting unit 6 uses this execution computer number to subtract the execution computer information passed from the processing request table 26, and transfers the processing request passed from the processing request receiving unit 3 to the corresponding computer. At this time, the execution computer that transferred the processing request is registered in the execution history information of the corresponding processing request stored in the execution history information storage unit 8.

The processing request thus allocated and transferred is executed by the processing request execution computer 10, and the execution result is sent back to the processing request allocation unit 11. The processing result receiving section 12 of the processing request allocating section receives the processing result and sends the processing result to the execution cost detecting section 9 and the processing result transmitting section 13. The execution cost detection unit 9 extracts the execution history information corresponding to the processing result from the execution history information storage unit and records the execution cost. In this case, the time taken for execution is used as the execution cost. Therefore, the processing request transmitting unit 6 records the time at that point in the execution cost column of the execution history information corresponding to the transfer of the processing request. The execution cost detector 9 re-records the difference between the time recorded in the execution cost column and the current time as the execution cost in the execution cost column. Although the execution time is used as the execution cost in this embodiment, there is also a method in which the processing request execution computer 10 sends back the execution cost together with the processing result. In this case, the processing request execution computer 10 can give arbitrary data as the execution cost. As an example of the execution cost, it is possible to use the number of disk references required during the execution, the number of times data is accessed by another node, the number of times waiting for a lock, and the like.

In addition to the method in which the processing request allocating section relays the processing result and sends it back to the processing request generating section as in the present embodiment, the processing request execution computer directly sends the processing result back to the processing request generating section, and the processing request allocating section It is also possible to adopt an implementation method in which the communication path between them is monitored, and if a packet of the processing result is found, it is taken in and the execution cost is detected.

In the present embodiment, the correspondence between the processing request sent to the processing request execution computer and the processing result is determined by referring to the execution history information. However, in this method, it is impossible to simultaneously allocate a plurality of packets to the same processing request execution computer and perform multiple processing. In order to enable this, the processing request packet is given a unique identification number as shown in FIG. The processing request execution computer is shown in FIG.
As shown in (b), the processing result is returned with the identification number attached to the processing request. By using this unique identification number, even when a plurality of processing requests are simultaneously assigned to the same execution computer, it becomes possible to associate the processing request recorded in the execution history information with the processing result. In addition to the method of identifying the processing request to the same computer using such an identification number, the communication channel connecting the processing request assigning unit and the processing request executing computer is logically multiplexed, and at the same time, the processing channel to which the processing request to the same computer is different. Can be identified by using.

The weight information of each characteristic parameter used for selecting the processing request execution computer to which the processing request is assigned may be given by a programmer or a system administrator, but it is desirable that such a value can be automatically determined. For that purpose, a weight calculator (27 in FIG. 11) as shown in FIG. 30 is used. The weight calculation unit 27 determines from the past execution history information stored in the execution history information storage unit 8 which characteristic parameter has a strong correlation with the optimum processing request execution computer, and the correlation is strong. The weight of the characteristic parameter is set to be larger. The weight calculation unit 27 roughly includes a cluster calculation unit 37 that obtains the number of clusters for each characteristic parameter,
The weight normalization unit 38 normalizes the number of clusters and converts it into weight information. The cluster calculation unit 37 corresponding to the jth feature parameter has a configuration as shown in FIG. 31 and receives as input all execution history information belonging to the type of request for which weight information is to be calculated. The execution history information is input to the characteristic parameter order sorting unit 41, rearranged in the order of the jth characteristic parameter, and sequentially sent to the computer number change counting unit 42. The computer number change counting unit 42 sequentially receives the execution history information, and outputs the number of times the execution computer recorded therein changes as the number of clusters.
If there is a correlation between the j-th characteristic parameter and the optimum execution computer, it is highly likely that the execution history information having the same execution computer will be lined up if sorted by the j-th characteristic parameter, so the number of clusters is small. Become.

The number of clusters of each characteristic parameter thus obtained is sent to the weight normalization section 38, and also to the maximum value detection section 40 and the minimum value detection section 39. The maximum value detector 40 calculates the maximum value of the number of clusters for each characteristic parameter, and the minimum value detector 39 calculates the minimum value of the number of clusters for each characteristic parameter. The weight normalization unit 38 for each characteristic parameter receives the number of clusters C of the characteristic parameter, the output min of the minimum value detection unit, and the output max of the maximum value detection unit as input, and normalizes the number of clusters to obtain the characteristic parameter. Is output as the weight W. For this normalized output, the following formula can be used as an example.

W = 0.9 * (max-C) / (max-
min) +0.05 When this expression is used, the weighting of the feature parameter with the smallest number of clusters is 0.95, and the weighting of the feature parameter with the largest number of clusters is 0.05. A similar normalization can be used to perform linear normalization with values between 0.0 and 1.0. There is also an implementation method in which normalization is performed using a non-linear function so that little weight is applied while the number of clusters is large.

Of the execution history information stored in the execution history information storage unit 8, the old one is deleted by the execution history information management unit 7. The execution history information to be deleted is an object recorded before a certain period of time or an old object exceeding the number when the number of pieces of execution history information to be stored in the execution history information storage unit is determined. In this embodiment, the number of pieces of execution history information that can be stored in the execution history information storage unit 8 (the number of saved histories) is constant. The number of saved histories may be given by a programmer or system administrator, but it is desirable that such a value can be automatically determined. for that reason,
A save history number determination unit that has a configuration as shown in FIG. 32 and automatically adjusts the save history number is used.

The save history number determination unit manages the current save history number W and the save history number w smaller than the current save history number W, and selects the execution computer using the past W execution history information for a new processing request. The estimated cost of the case and the estimated cost when the execution computer is selected by using the past w pieces of execution history information, and saved based on the statistic of the difference between the execution cost actually allocated and executed. Change the number of histories. This operation will be described below with reference to FIG.

The storage history number determination unit has two execution computer selection units, an execution computer selection unit 1 and an execution computer selection unit 5.
These execution computer selection units have the same configurations as those shown in FIG. 23 and FIG. 27, but the output is not the execution computer number, but the predicted execution required for execution of the processing request by the selected execution computer. The cost. That is, in FIG. 23, the minimum cost computer selection unit of the execution computer selection unit 5 is
It outputs not only the execution candidate computer number that gives the minimum cost, but also the minimum execution cost. Further, in FIG. 27, not only the number of the execution candidate computer relating to each characteristic parameter but also the execution cost associated therewith are sorted by the weight order sort unit, and the probabilistic selection unit also handles the execution candidate computer number and the execution cost as a set. The execution computer selection unit 1 can also serve as the execution computer selection unit 8 used in the processing request allocation unit as shown in FIG.

The execution computer allocating unit, which has received the new processing request, uses the W pieces of execution history information stored in the execution history information storage unit to execute the prediction required for the execution of the processing request by the execution computer selecting unit 1. Calculate the cost 1. At this time, the execution computer number estimated to be executable at the predicted cost 1 is sent to the processing request transmission unit to allocate the processing request. At the same time, the execution computer selection unit 5 uses the w pieces of execution history information to obtain the predicted cost 2 required to execute the processing request. Predicted cost 1 and predicted cost 2 thus obtained
Then, the deviation calculation unit 1 and the deviation calculation unit 2 respectively obtain the difference between the execution cost and the execution cost required to execute the processing request by the execution computer selected by the execution computer selection unit 1. The deviation calculator outputs the absolute value of the arithmetic difference between the predicted cost and the actual cost.

The outputs and actual costs of the deviation calculator 1 and the deviation calculator 2 for each processing request are passed to the average calculator 1, the average calculator 2, and the average calculator 3, respectively. The average calculation unit obtains and outputs an average value of a fixed number in the past with respect to the input value. That is, the outputs of the average calculation unit 1 and the average calculation unit 2 are averages of the absolute values of the differences between the predicted cost 1 and the predicted cost 2, respectively. The output of the average calculation unit 3 is the average value of the actual costs. The outputs of the average calculator 1 and the average calculator 2 are passed to the error rate calculator 1 and the error rate calculator 2. The error rate calculation unit 1 and the error rate calculation unit 2 average the difference between the predicted cost and the actual cost passed from the average calculation unit 1 and the average calculation unit 2, respectively, and calculate the average of the actual costs passed from the average calculation unit 3. Output the value divided by the value. These values are called error rate 1 and error rate 2, respectively. The error rate is an index showing how much the predicted value is equal to the actual cost.

As shown in FIG. 33, the storage history number updating unit
Input the current save history number W, the lower save history number w, the error rate 1 and the error rate 2, calculate the desired save history number WW and the lower save history number ww, and calculate the desired save history number WW and the lower save history number ww. Set as the number of saved histories. Assuming that the error rate 1 and the error rate 2 are Eu and El, respectively, if El is smaller than the error rate T that must be satisfied in advance (S12), the current number of saved histories is too large and expected. It means giving a better predicted value than it does. At this time, the lower storage history number w is set as the new storage history number WW, and a number smaller than w is set as the new lower storage history number ww (S13). Conversely, when Eu is larger than T (S1
In 4), the current number of saved histories is too small to give the expected quality predicted value. At this time, the current storage history number is set as a new lower storage history number ww, and a number larger than W is set as a new storage history number WW (S15).

(Effects of the Invention) (As described above, by using the processing request allocating unit of the present invention, in an information processing system composed of a plurality of computers, processing requests and the optimum computer for executing the processing requests can be obtained. If there is a correlation between
The correlation can be automatically detected and the processing request can be assigned to the optimum computer. Since the information necessary for allocating the processing request is automatically detected, the programmer or system administrator does not need to give an allocation procedure, and the performance of the entire information processing system can be easily obtained. (Fourth Embodiment) Next, a fourth embodiment of the present invention will be described.

The overall system diagram of this embodiment is as follows:
This embodiment is the same as that shown in the above embodiment, and the difference between this embodiment and the above-mentioned embodiment is that the method of cooperation between the processing request allocation unit, the data arrangement determination unit 120, and the data management unit 117 is different. It is in. Hereinafter, differences between the present embodiment and the above-described embodiments will be mainly described. It should be noted that, in order to distinguish it from the processing request allocating unit of the above-described embodiment, the processing request allocating unit of this embodiment is represented by 301.

In FIG. 1, FIG. 2, and FIG. 3, when the processing request allocating unit 101 receives a new processing request, it compares the execution history with the application program 1 to be executed.
16 processing request execution computers 110, which exist, were selected to instruct execution of the processing request.

In this embodiment, the processing request allocation unit 301
In addition, the data storage device 11 of each processing request execution computer 110
8 is provided with a means for knowing data arrangement information indicating what kind of data is stored, and when a new processing request is received, the processing request execution computer 1 in which the application program 116 to be executed by comparing with the data arrangement information exists
10 is selected to instruct execution of the processing request.

FIG. 34 shows an example of the internal structure of the processing request allocation unit 301 of this embodiment. Processing request allocation unit 3
01 is a data allocation table 30 for recording data allocation information.
3 and a routing table 30 for recording routing information
4 and an allocation determination unit 302 that determines a processing request execution computer 110 to execute a newly arrived processing request based on these tables 303 and 304 and instructs execution of the processing request.

The data arrangement table 303 has a structure as shown in FIG. 35, for example. In the data arrangement table 303, for each database, information indicating which range of the key value is stored in which data storage device of which processing request execution computer is stored is recorded. For example, in the example of FIG.
It indicates that there is a database named COUNT, and the data whose primary key value is in the range of 0 to 999 is in the data storage device 118 of the processing request execution computer 110 having the number 0.

The data allocation table 303 may be manually created and given to the processing request allocating section 301. Further, when the data allocation determining section 120 determines new data allocation, the information is sent to the processing request allocating section 301. It is also possible to carry out and record the data allocation table 303.
Furthermore, an implementation method is also possible in which each data management unit 117 informs the processing request allocation unit 301 of what data it manages and records it in the data allocation table 303.

The routing table 304 has a structure as shown in FIG. 36, for example. The routing table 304 shows the application program 1 on the processing request execution computer 110.
16 records the types of process requests that can be executed, the names of arguments used when allocating process requests, and the names of databases that are also used when allocating process requests.

The processing request is, for example, as shown in FIG.
It consists of a header that indicates the type of processing request and some arguments required to execute the processing request. In the example of FIG. 37, the processing request called WITHDRAW for withdrawing a deposit is
ACCOUNT_ID showing account number, TELLER_ID showing window number, BRANCH_I showing branch number
It has four arguments, D and AMOUNT indicating the withdrawal amount.

In the routing table 304 of FIG. 36, the processing request WITHDRAW for withdrawing deposits is AC.
It indicates that the processing request is allocated by checking the value of the argument COUNT_ID and the database ACCOUNT. That is, when a WITHDRAW processing request arrives at the processing request allocating unit 301, the allocation determining unit 302 first looks at the routing table 304, and WITHDRAW indicates the value of ACCOUNT_ID and A.
Know to route using the CCOUNT database. Next, the ACCOUNT_ID of the processing request
To retrieve the value of the argument. After that, the allocation determination unit 3
02 is the data of the ACCOUNT database, looking at the data arrangement table 303, and the value of the key is the ACC of the processing request.
It is checked which computer number has the same value as the OUNT_ID argument in the processing request executing computer. The processing request execution computer thus checked is instructed to execute the processing request.

The routing table 304 can be manually created and recorded in the processing request allocation unit 301.

The processing request allocating unit 101 as shown in the first embodiment operates with a probabilistic algorithm that cannot always select the optimum processing request execution computer 110, while the processing request allocating unit 101 of the present embodiment operates. The processing request assigning unit 301 such as
If 3 is given correctly, it operates by a deterministic algorithm that can always select the optimum processing request execution computer 110. When the deterministic algorithm is used, it is possible to obtain an effect that the processing request can be assigned according to the new data arrangement immediately after the data arrangement is changed. On the other hand, in the case of the probabilistic algorithm, it is possible to obtain the effect that it is not necessary to give the processing request allocating unit information regarding allocation such as a routing table.

The present invention is not limited to the above-described embodiments, but can be implemented with various modifications without departing from the scope of the invention.

[0169]

According to the present invention, an optimum computer for executing a processing request, that is, a computer capable of executing the processing request at the lowest cost is selected, processing request allocating means for allocating the processing, and automatic data optimum allocation are provided. A data allocation determining means is provided to form a feedback loop for processing request allocation and data relocation according to the data allocation.

As a result, the optimum data division arrangement for the dynamic characteristics such as what kind of processing request occurs and how often by the automatic optimum arrangement of data by the data arrangement determining means. It is possible to improve the execution performance of each processing request.

Further, the processing request allocating means can allocate each processing request to a computer suitable for executing the processing request according to the new data allocation, so that the method of allocating the data distribution and allocating the processing request are not given. However, the processing load can be automatically distributed, and the execution time of each processing request is optimized.

Further, the tuning of the entire data processing device can be automated and system management becomes easy.

[Brief description of drawings]

FIG. 1 is a diagram schematically showing a configuration example of a data processing device according to a first embodiment.

FIG. 2 is a diagram schematically showing another configuration of the data processing device according to the first embodiment.

FIG. 3 is a diagram schematically showing a configuration example of a data processing device according to a second embodiment.

FIG. 4 is a diagram schematically showing a configuration example of a data processing device according to a third embodiment.

FIG. 5 is a schematic configuration diagram of a data management unit and a data arrangement determination unit according to the first to third embodiments.

FIG. 6 is a diagram showing a configuration example of a processing request allocation unit according to the embodiment.

FIG. 7 is a diagram showing another configuration example of a processing request allocation unit.

FIG. 8 is a diagram showing a modification of the flow of feature parameter extraction.

FIG. 9 is a diagram showing another configuration example of the processing request allocation unit.

FIG. 10 is a diagram showing another configuration example of the processing request allocation unit.

FIG. 11 is a diagram showing a detailed configuration example of a processing request allocation unit.

FIG. 12 is a diagram showing an example of a processing request packet.

FIG. 13 is a diagram showing the contents of a processing request table.

FIG. 14 is a diagram showing an example of argument information and an example of interface description of RPC.

FIG. 15 is an explanatory diagram of functions of the RPC compiler.

FIG. 16 is a diagram showing the contents of RPC packet information.

FIG. 17 is a diagram showing the contents of execution history information.

FIG. 18 is a diagram showing a data flow regarding a processing request table.

FIG. 19 is a diagram showing the contents of a processing request packet.

FIG. 20 is an explanatory diagram of an operation example of a characteristic parameter extraction unit.

FIG. 21 is a flowchart showing the operation of the characteristic parameter extraction unit.

FIG. 22 is a diagram showing an example of bit position inversion.

FIG. 23 is a diagram showing a configuration example of an execution computer selection unit.

FIG. 24 is a flowchart showing the operation of the approximate cost selection unit.

FIG. 25 is a flowchart showing another operation of the approximate cost selection unit.

FIG. 26 is a flowchart showing the operation of the minimum cost computer selection unit.

FIG. 27 is a diagram showing a configuration example of a computer number selection unit.

FIG. 28 is a flowchart showing the operation of the probabilistic selection unit.

FIG. 29 is a diagram showing an example of an RPC processing request packet and a processing result packet.

FIG. 30 is a diagram showing a configuration example of a weight calculation unit.

FIG. 31 is a diagram showing a configuration example of a cluster calculation unit.

FIG. 32 is a diagram showing a configuration example of a storage history number determination unit.

FIG. 33 is a flowchart showing the operation of the storage history number updating unit.

FIG. 34 is a diagram showing an example of an internal configuration of a processing request allocation unit according to the fourth embodiment.

FIG. 35 is a diagram showing a configuration example of a data allocation table.

FIG. 36 is a diagram showing a configuration example of a routing table.

FIG. 37 is a diagram showing a configuration example of a processing request.

FIG. 38 is a diagram showing a configuration of a conventional data processing device.

FIG. 39 is a diagram showing another configuration of the conventional data processing device.

[Explanation of symbols]

101, 101-1 to 101-m ... Processing request allocating section, 102-1 to 102-n, 102-1-1 to 102
-M-r ... processing request generation unit, 110-1 to 110-m,
111, 111-1 to 111-l ... Computer, 114 ... Coupling path, 115, 115-1 to 115-l ... Communication path, 11
6-1 to 116-m ... Application program, 1
17-1 to 117-m ... Data management unit, 118, 118
-1 to 118-m ... Data storage device, 120 ... Data allocation determination unit 201 ... Access request reception unit, 202 ... Correlation information storage table, 203 ... Correlation information management unit, 204 ... Data allocation determination unit, 205 ... Data storage information Storage unit, 206 ... Data allocation changing unit, 207 ... Data access unit 301 ... Processing request allocating unit, 302 ... Allocation determining unit, 303 ... Data allocation table, 304 ... Routing table 1, 11 ... Processing request allocating device, 2 ... Processing Request generation source computer, 3 ... Processing request receiving unit, 4 ... Characteristic parameter extracting unit, 5 ... Execution computer selecting unit, 6 ... Processing request transmitting unit, 7 ...
Execution history information management unit, 8 ... Execution history information storage unit, 9 ... Execution cost detection unit, 10 ... Processing request execution computer, 12 ... Processing result receiving unit, 13 ... Processing result transmitting unit, 25 ... Processing request table management unit, 26 ... Processing request table, 27 ... Weight calculation section, 28
... feature parameter extraction section 1, 29 ... feature parameter extraction section 2, 30 ... approximate cost selection section, 31 ... minimum cost computer selection section, 32 ... computer number selection section, 33 ... weight order sorting section, 34 ... stochastic selection section , 35 ... Random number generation section, 36 ... Arbitrary computer selection section, 37 ... Cluster calculation section, 38 ... Weight normalization section, 39 ... Minimum value detection section, 40 ... Maximum value detection section, 4
1 ... Sorting section for characteristic parameters, 42 ... Computer number change counting section

Claims (2)

[Claims] 1. A plurality of data storage units for storing data, a unit connected to any one of the data storage units, for executing a given processing request using an application program, and for executing the processing request. A plurality of computers each having data management means for accessing data in the data storage means necessary for the execution based on a request from the application program and accumulating information on a history of access to the data; Joining means for interconnecting a plurality of computers, means for accepting a processing request sent from at least one processing request generating means connected to this joining means, a past processing request, and the processing The computer that executed the request, a means for accumulating an execution history including the cost required for the execution, and the processing request generation A processing request allocating means having means for selecting a computer in which the application program to be executed is compared with the execution history when receiving a new processing request sent from the means, and instructing execution of the processing request; Information regarding the history of access to the data is received from the data management means on the computer, the optimum data arrangement is determined based on the information, and the data management means is based on the determined data arrangement. A data processing apparatus, comprising: a data layout determining means for instructing execution of data layout changing processing. 2. A plurality of data storage units for storing data, a unit connected to any one of the data storage units, for executing a given processing request by using an application program, and for executing the processing request. A plurality of computers each having data management means for accessing data in the data storage means necessary for the execution based on a request from the application program and accumulating information on a history of access to the data; Connecting means for interconnecting a plurality of computers; means for receiving a processing request sent from at least one processing request generating means connected to the connecting means and connected via a communication path; A means for acquiring data arrangement information indicating what kind of data is stored in the data storage means; When a new processing request sent from the request generating means is received, it is compared with the data allocation information, and the processing request allocating means having means for selecting the computer having the application program to be executed and instructing execution of the processing request Receiving information regarding the history of access to the data from the data managing means on the computer, determining the optimum data arrangement based on the information, and setting the determined data arrangement for the data managing means. A data processing apparatus, comprising: a data allocation determining unit for instructing execution of a data allocation changing process based on the data allocation determining unit.