JPH09265459A - Method for controlling data processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To execute scheduling capable of suppressing the reversion between the generation order of respective processing requests and the end order of processing and shortening the processing time of the whole information processing system provided with plural data processors having respectively different processing speeds. SOLUTION: In schedular processing, the estimated processing time of each database processor 5 (5a or 5b) for a generated processing request, the processing time of the whole queue for the processing request in each processor 5 and the remaining time of the processing request being processed by each processor 5 are respectively estimated. Then a processing request generated in the queue of the processor 5 having the shortest estimated processing end time obtained by totalizing the remaining time, the processing time of the whole queue and the estimated processing time is added.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to scheduling when a plurality of processing requests are processed by different data processing devices in an information processing system having a plurality of data processing devices having different processing performances.

[0002]

2. Description of the Related Art A conventional method for distributing a plurality of processing requests to a plurality of data processing devices is disclosed in, for example, Japanese Patent Laid-Open No.
It is disclosed in Japanese Patent Laid-Open No. 237568, Japanese Patent Laid-Open No. 4-3706303, and the like. The former is to predict the current and future loads for multiple data processing devices and let the computers with relatively low load output process the output, paying attention to the load states of each computer and balancing the loads. By doing so, an attempt was made to realize an efficient system. The latter, as shown in the flow chart of FIG. 10, allocates processing requests to the data processing apparatus in order each time the data processing apparatus becomes empty. Each processing request is stored in one queue and the processing order is stored in that order. By using a vacant data processing device, processing is performed in the order of generation of requests that continue to occur in chronological order.

[0003]

In the past, when creating a new system, data processing devices with different performances were not used intentionally, but higher performance data processing devices were added while inheriting past assets. In practice, there are many opportunities to use data processing devices with different performances at the same time due to problems such as connection with other systems and purchase prices. Since the conventional method that distributes processing requests to multiple data processing devices mainly performs the above processing, if there is a difference in performance among the installed data processing devices, which data processing device will process Depending on whether it is performed, a difference between a request processed faster and a request processed later occurs, and the generation order of the processing request and the end order are reversed. If different users make processing requests, individual users may be dissatisfied with subsequent requests being processed before themselves.
Further, in the conventional method, since a data processing device with good performance and a data processing device with inferiority are treated equally, if a data processing device with inferior performance is used carelessly because the load is small,
As a whole, the processing efficiency often becomes poor. It is not always necessary to use all the data processing devices evenly, and there is a problem of which data processing device is used in which order with respect to the order of processing requests.

The present invention is intended to solve the above-mentioned problems. A first object of the present invention is to determine the order of generation of requests and the order of completion of processing for each processing request that is generated sequentially in a time series. It is to suppress the reverse. A second purpose is to perform scheduling that shortens the processing time of the entire information processing system.

[0005]

A method of controlling a data processing device according to the present invention comprises an information processing system having a plurality of data processing devices having different processing speeds and scheduling a plurality of processing requests in the order in which the requests are generated. At
For the generated processing request, the expected processing time in each of the data processing devices, the processing time of the entire queue of processing requests for each of the data processing devices, and the remaining time of the processing request being processed in each of the data processing devices are described. Each of them comprises a step of estimating, a step of adding the generated processing request to the queue of the data processing device having the smallest processing end expected time, which is the sum of the remaining time, the processing time of the entire queue, and the expected processing time. is there.

In the step of estimating the expected processing time in each of the data processing devices and the processing time of the entire queue of processing requests in each of the data processing devices for the processing request, the processing time of each processing request is estimated in order to estimate the processing time. The data amount and load information for each data processing device are used.

Further, in the step of estimating the processing time of the entire queue of processing requests in each of the data processing devices, the elapsed time of each processing request being processed is measured, and the processing request being processed is processed from the elapsed time. The expected end time is corrected.

In the step of estimating the expected processing time in each of the data processing devices and the processing time of the entire queue of processing requests in each of the data processing devices with respect to the processing request, the occurrence distribution of the processing requests and the processing time of the processing request are performed. The statistical information of each is used as a reference for the above estimation.

Furthermore, the average time of the generation intervals of the processing requests and the average time of the processing times in each of the data processing devices are sampled, and when the average time of the generation intervals is larger than the average time of the processing times, the plurality of the plurality of data processing units are selected. The data obtained by multiplying the ratio of the performance ratio of the processing speed between the data processing devices by the ratio of the average time of the occurrence intervals to the average time of the processing time is used as the ratio of the performance ratio of the processing speed between the plurality of data processing devices. The processing time is estimated for each processing device.

When the processing time for each data processing device has a large error compared with the estimated time, the queue of processing requests for each data processing device is changed as a whole.

[0011]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1. An embodiment of the present invention will be described below by taking a database processing device as an example of a data processing device. FIG. 1 is a configuration diagram of the entire information system that distributes a plurality of processing requests 3 to a plurality of database processing devices 5 having different performances. The processing request 3 generated by the processing request program 1 independently in the queue 2 of the queue
Throw. One processing request program 1 may send a plurality of requests. The queue 2 is often a buffer on the shared memory 8 in the system. Scheduler 4
Makes a prediction of the processing time for the processing request 3 on the queue at an appropriate timing such as when the processing request 3 occurs or when the database processing device becomes empty, and based on that, 5a and 5b are estimated. The database processing device of the above is executed for distribution. The actual processing request is sent to the database processing device 5
The control program 6 that controls the database processing device 5 is passed to the. The scheduler 4 activates the control program 6 and requests the processing of the processing request 3. The database processing device 5, the computer on which the control program 6 is executed, and the storage device 9 in which data is actually stored are connected via a bus (not shown). Reference numeral 5a is a database processing device having a high processing speed, and 5b is a database processing device having a low processing speed. Although FIG. 1 shows a case in which there are two database processing devices for simplification, a system equipped with three or more database processing devices may be used.

Next, scheduling inside the scheduler 4 will be described with reference to the process flow chart of FIG. The series of processes shown in FIG. 2 are performed at an appropriate timing such as when the processing request 3 is generated or when the database processing device 5 becomes empty. With respect to the unprocessed processing request 3 on the queue 2, the processes from S1 to S5 are performed in order from the beginning until the database processing device 5 is no longer in an empty state. First, one processing request 3 for making a prediction of the processing end time is taken out (step S1). Next, with respect to the extracted processing request 3, the processing time estimate when processing is performed in all the installed database processing devices 5 is calculated (step S2). At this time, the processing time is estimated based on the performance ratio of the database processing device 5 (step S2b). The performance ratio is directly calculated as the ratio of processing time. That is, it is assumed that the processing time of a database processing device having a double performance ratio is half. What is important here is that the expected processing time is obtained as a relative ratio, not as an absolute time. Then, the estimates for the respective database processing devices 5 are compared with each other, and one database processing device that is likely to complete the processing is determined (step S3). If the determined database processing device 5 is free, that database processing device 5
Is used to process the processing request 3 (step S5). If the processing request 3 is not available, the processing request is pending and the next processing request 3 is estimated (step S1). As for the processing itself, a control program 6 different from the scheduler 4 is activated to request the processing. The pending processing request 3 is added to the estimate on the assumption that it is processed by the assigned database processing apparatus 5, but it is undecided at this point whether or not it is actually processed by the database processing apparatus 5. That is,
The processing request 3 generated earlier is characterized by waiting for the high-speed database processing device 5 instead of immediately processing by the vacant database processing device 5 due to the prediction of the end time. Further, when the time of the subsequent processing request 3 is predicted, it is expected that the end time will be faster if the database processing device 5 is intentionally used later than waiting, and the end time is expected to be earlier than the waiting processing request 3. Start processing. Further, the process request program 1 itself, which is different from the scheduling, puts the process request 3 in the queue 2 or deletes it from the queue 2.

According to the above procedure, the slow database processing device 5b is shunned and the fast database processing device 5a is avoided by performing estimation based on the performance ratio corresponding to 5a and 5b in FIG.
It is possible to perform scheduling that effectively utilizes the. The effect of performing this schedule will be described with reference to FIG. In the case of scheduling in which the performance ratio is not taken into consideration, the processing request 11a is allocated to the fast database processing device 5a, and then the processing request 11b is allocated to the empty slow database processing device 5b. Processing request 1
When the processing of 1a is completed, the processing request 11c is next assigned to 5a and executed. In such scheduling, the end time of the processing request 11b becomes later than that of the processing request 11c.

On the other hand, in the new scheduling, after allocating the processing request 12a to the database processing device 5a, it is judged that the processing request 12b should be executed by 5a (step S3 in FIG. 2). wait. The processing request 12c is also assigned to 5a for the same reason. When the processing request 12d is reached, the use of the database processing device 5b which is slow is admitted, and it is assigned to 5b and executed (step S5 in FIG. 2). Further, when the processing of the processing requests 12a to 12d is completed, the remaining processing requests are scheduled again by the same method. By such scheduling, the processing requests 12a, 12b, 12c, 12d, 1
The reversal of the end times of 2e and 12f will not occur. Further, the end time of the processing request 12f, which is the last processing request, is shorter than that of the conventional schedule 11f.

However, when the processing request to be preprocessed by the slow database processing device 5a is much smaller than the processing request to be waited for, the processing may be reversed even by the above method due to an estimation error. . However, the reversal in this case is not due to improper scheduling but due to the processing content of the processing request itself. Because it is small, it ends first. In order to strictly follow the processing order, extremely speaking, all processing requests can be processed only by the fast database processing device, but this does not lead to effective use of resources and is inefficient as a whole. is there.
Even if the processing order is reversed, the entire processing time can be shortened according to the above method. The second object is achieved while keeping the first object of the invention.

The process request 3 is the process request program 1, and can be regarded as the user itself. The above method, from the standpoint of each user,
It's a schedule that follows the process of deciding which processing units should be lined up so that they can be processed as quickly as possible. The user may be dissatisfied that another user who is lined up after himself finishes the process before himself. According to the above method, there is no inversion of the end time due to being assigned to the involuntary database processing device 5.

After all, in order to prevent the ending order of a plurality of processing requests 3 from being reversed depending on how the database processing device 5 is assigned, the processing requests 3 that have occurred are subject to the above database processing devices 5. The estimated processing time for each of the database processing devices 5, the processing time of the entire queue of the processing request 3 for each of the database processing devices 5, and the remaining time of the processing request 3 being processed in each of the database processing devices 5 are estimated. This is achieved by adding the generated processing request 3 to the queue of the database processing device 5 having the smallest processing completion expected time, which is the sum of the remaining time, the processing time of the entire queue, and the expected processing time.

Further, even if the order of occurrence of a plurality of processing requests 3 can be reversed depending on how the database processing device 5 is allocated, the processing time of the entire information processing system can be minimized by For the generated processing request 3, the expected processing time in each of the database processing devices 5 and the processing request 3 for each of the database processing devices 5
Of the entire queue and the remaining time until the end of the processing request being processed in each of the database processing devices 5 are estimated, and the remaining time, the processing time of the entire queue and the expected processing time are summed up. This is achieved by adding the generated processing request 3 to the queue of the database processing device 5 that is closest to the average value of the expected processing completion time.

Embodiment 2. In the first embodiment, only the performance difference of the database processing device is used as the estimation condition at the time of scheduling, but the processing time of the processing request 3 greatly differs depending on the processing content of each processing request 3. Therefore, scheduling in which the processing content is added to the estimate can be considered. The configuration diagram of the second embodiment is similar to that of FIG.

Scheduling taking the processing content into consideration will be described with reference to the processing flow chart of FIG. Similar to the processing flow of FIG. 2, for each unprocessed processing request 3, estimates for all database processing devices 5 are calculated,
The database processing device 5 to be assigned is determined. 2 is different from FIG. 2 in that in the step of making a processing time prediction (step S12b), the processing content of processing request 3 is used as a judgment factor in addition to the performance difference of the database processing device. The data amount and the load information on the database processing device 5 are used to represent the processing content. The data amount may be a file size of data to be processed, a block size, the number of records, or the like. The load information is a value unique to the data, and may be the number of selection conditions or the performance ratio of the storage medium in which the data is stored. The processing time is estimated by multiplying these by the performance ratio of the database processing device 5.

By following the above procedure, the processing time of the processing request 3 can be estimated more accurately, and the reversal of the processing end order due to the size of the processing request 3 can be considerably suppressed. Further, the effect of the first embodiment can be enhanced such that the processing time in the entire information processing system can be more accurately minimized.

Embodiment 3 FIG. In order to make a more accurate estimate, it can be considered that the processing time of the processing request 3 that is being processed by the database processing device 5 is also subject to the estimation. In addition, depending on how to estimate the processing request 3 during this processing, exception processing can be performed when a large error occurs in the estimation so far. The processing elapsed time is used to estimate the processing request during processing.

Scheduling for adding the elapsed time of the processing request 3 being processed to the estimation will be described with reference to the processing flow chart of FIG. One processing request 3 shown in FIG. 1 for predicting the processing end time is taken out (step S21a). If the fetched processing request 3 is being processed, an estimate is made based on the processing elapsed time from the start of the processing to the present (step S21c). Then, the next processing request 3 on the queue 2 is taken out (step S21a). If the request is not being processed, the same steps as those in the first embodiment are continued (step S22a). The subsequent steps are the same as in the first embodiment.

Different effects can be obtained depending on how the processing elapsed time is added to the estimate. For example, if the estimated time of the remaining processing time is corrected in proportion to the elapsed time of the processing request 3 being processed, more accurate processing time can be predicted. Also, when an abnormal state in which the actual processing elapsed time greatly exceeds the initial estimated value set at the time of scheduling, the remaining processing estimated time is set to be very large, so that the subsequent processing request 3
Can be easily assigned to another database processing device 5. In other words, it is possible to actually realize that the subsequent processing request 3 waiting for the database processing device 5 that is in an abnormal state due to an unknown cause is lined up in another database processing device 5 with numbness. You can do it. Of course, as the elapsed time decreases, the prediction of the remaining processing time is reduced, and when it exceeds a certain threshold value, it is considered as an abnormal state and the expected time is extremely increased and the subsequent processing request 3 is shut down. It is also possible to combine the effects of. Eventually, when the processing time becomes abnormal with a large error compared with the estimated time, the queue of the processing request for each database processing device 5 is changed in the information system as a whole to change the situation. Can respond in real time.

Embodiment 4 In the first embodiment, only the processing requests 3 stored in the queue 2 by the processing request program 1 are targeted for scheduling. However, the processing request 3 is not prepared in advance and is generated intermittently. Therefore, it is possible to consider a schedule that takes into account processing requests that have not yet occurred.

The processing flow of the fourth embodiment will be described with reference to FIG. If the processing request 3 for which an estimate is to be made is the request that occurred last at that point (the last processing request on the queue), the expected distribution of future occurrences of the processing request 3 is added to the estimate (step S32c). ). If the occurrence distribution is dense to some extent, the occurrence interval until the subsequent processing request is generated is short, and therefore the estimation of the last processing request 3 on the queue 2 does not need to be changed so much. However, when the occurrence distribution is sparse, the occurrence interval until the subsequent processing request is generated is long, so that time is wasted. Therefore, by performing the estimation while suppressing the performance ratio of the database processing device 5 with respect to the last processing request 3 on the queue, the slow database processing device 5 is easily used and the processing request generation interval is reduced. A time-saving schedule can be done. The subsequent steps are the same as those in the first embodiment (steps S32d to S35).

By the method as described above, it is possible to perform the scheduling in consideration of the processing request that will occur from now on, and it is possible to effectively use the database processing apparatus.

Embodiment 5 A configuration in which the first embodiment, the second embodiment, the third embodiment, and the fourth embodiment are arbitrarily combined can be considered. FIG. 7 is a configuration diagram that simultaneously includes the first to fourth embodiments described above. In a plurality of database processing devices having different performances, a plurality of processing requests that occur intermittently and have different processing contents are distributed. Processing request program 1 is queue 2 independently
Process request 3 is intermittently thrown. The scheduler 4 makes a prediction of the processing time of the processing request 3 being processed and the processing request 3 waiting for processing on the queue 2 at an appropriate timing, and while considering the distribution of the processing requests that have not yet occurred, 5a,
The database processing device 5b is distributed and executed.

Next, the scheduling inside the scheduler 4 will be described with reference to the process flow chart of FIG. At an appropriate timing such as when the processing request 3 is generated or when the database processing device 5 becomes empty,
The process of FIG. 8 is performed. The processes from S41 to S45 are performed in order from the beginning of the processing request 3 on the queue 2 until the database processing device 5 is no longer in an empty state. First, one processing request 3 for predicting the processing end time is taken out (step S41a). If the fetched processing request is being processed, estimation is performed based on the processing elapsed time from the start of the processing to the present (step S41c), and then the next processing request 3 on the queue 2 is fetched. (Step S41a). If the request is not being processed, the process proceeds to the selection of the database processing device 5 (step S42a). If the processing request 3 that is the target of the prediction is the last request that occurred at that point (the last processing request on the queue),
Based on the occurrence distribution of the processing request 3, the performance ratio of the database processing device 5 is suppressed (step S42c). Next, the processing time is estimated based on the processing content of the processing request 3 and the performance ratio of the database processing device 5 (step S42d). The estimates for the respective database processing devices 5 are compared with each other, and one database processing device 5 that will finish processing the fastest is determined (step S43). If the determined database processing device 5 is free, the processing request 3 is processed using the database processing device 5 (step S4).
5) If there is no vacancy, the processing request 3 is pending and the next processing request is estimated (step S41).
a). As for the processing itself, a control program 6 different from the scheduler 4 is activated to request the processing. The pending processing request 3 is added to the estimate on the assumption that it is processed by the assigned database processing device 5, but whether or not it is actually processed is undecided at this point. Further, the process request program 1 itself, which is different from the scheduling, puts the process request 3 in the queue 2 or deletes it from the queue 2.

By following the above procedure, processing request 3
Processing contents of each and the database processing devices 5a and 5
It is possible to perform the scheduling based on the performance ratio of b, avoiding the slow database processing device 5b, and effectively utilizing the fast database processing device 5a. By this schedule, it is possible to prevent the request generation order and the processing end order from being reversed between processing requests. Further, based on the elapsed time of processing requests during processing and the occurrence distribution of unprocessed processing requests, it is possible to more accurately estimate the processing time and use the database processing device 5 without waste. It is also possible to combine only the required functions.

Embodiment 6 FIG. In the first and second embodiments, the current state for performing the schedule is used as the estimation material. In addition to this, in the third embodiment, the current processing status is used as a material for estimation. Further, in the fourth embodiment, the means for estimating the future processing request that has not yet occurred is used as a material for the estimation, and the estimation is performed in consideration of all in the fifth embodiment. Furthermore, it is conceivable to add past statistical information to the estimation material. FIG. 9 is a configuration diagram in which statistical information on the occurrence distribution of the processing request 3 and the actual processing time of the processing request 3 is collected and fed back to the scheduler 4. The statistical processing 20 checks the occurrence interval, and the statistical processing 21 checks the processing time.

The statistical processing 20 and the statistical processing 21 are independent of the scheduler 4 and the processing request program 1.
When the processing request program 1 sends the processing request 3 to the queue 2, the statistical processing 20 is always passed. The statistical processing 20 collects statistical information of the generation intervals of the respective processing requests 3 and passes the result to the scheduler 4. The scheduler 4 is the statistical processing 2
The result from 0 is used for the step of adding the occurrence distribution of the fifth embodiment to the estimation (S42c in FIG. 8). In addition, the statistical processing 21 collects statistical information on the processing time when the processing request 3 is actually processed by the database processing device, and passes the result to the scheduler 4. The scheduler 4 uses the result from the statistical processing 21 for the step (S41c in FIG. 8) of adding the elapsed time of the processing request 3 during processing of the fifth embodiment to the estimate.

For example, in the statistical processing 20, the average time t of the occurrence intervals of the past n processing requests 3 counted from the present time is sampled, and in the statistical processing 21, the average processing time s of the past n processing requests 3 is similarly taken. Shall be collected. In addition, the step of adding the occurrence distribution of the fifth embodiment to the estimation (FIG. 8, S42).
The value used for c) is T, and the elapsed time of the in-process request according to the fifth embodiment is added to the evaluation (FIG. 8, S41c).
Let S be the value used for. If t <s, T = 1
If t ≧ s, then T = s / t. Also, S is s
Give itself. In this case, in S42c, the performance ratio is reduced by multiplying T by the ratio of the performance ratio of the connected database processing device 5 to the database processing device 5 having the slowest processing speed in the system, thereby estimating the processing time. Can stand. In S41c, the estimated processing time is multiplied by S / (S + increase function of processing elapsed time) to reduce the estimated time as the processing time elapses, and it is possible to perform a schedule corresponding to the operating state of the system in real time. Becomes A linear function, for example, can be considered as the increasing function. Of course, in addition to freely changing the value of n, if the system operating status is determined by date, day of the week, or time zone, the statistical information to be collected is separated by date, day of the week, or time zone. It is also possible to pass a value to the scheduler 4.

According to the above method, the distribution of the processing requests 3 and the statistical information of the processing time are used to flexibly and optimally schedule according to the system-specific environment and the usage condition other than the performance ratio of the database processing device 5. It can be carried out.

The present invention can be applied regardless of whether the database processing device 5 is another processing device such as a CPU, a hardware board, a peripheral device such as a printer, a computer connected to a network, or a mixture thereof. Is possible.

[0036]

Since the present invention is constructed as described above, it has the following effects.

Since the processing request generated in the queue of the data processing device having the smallest processing completion expected time, which is the sum of the remaining time, the processing time of the entire queue, and the expected processing time, is added, it is arranged in time series. For each processing request sequentially generated along the line, it is possible to prevent the request generation order and the processing end order from being reversed, and it is possible to perform scheduling so as to shorten the processing time of the entire information processing system.

Further, since the data amount of each processing request and the load information for each data processing device are used to estimate the processing time, a more accurate estimation of the processing time can be obtained.

Further, since the elapsed time of each processing request being processed is measured and the processing completion expected time of each processing request being processed is corrected from the elapsed time, a more accurate processing time can be estimated. can get.

Further, since the statistical distributions of the processing request generation distribution and the processing time of the processing request are used as a reference for the above estimation, a more accurate estimation of the processing time can be obtained.

Furthermore, when the average time of the generation intervals is larger than the average time of the processing time, the processing time for each data processing device is estimated by reducing the ratio of the performance ratio of the processing speed among the plurality of data processing devices. With this configuration, a database processing device having a slow processing speed is easily used, and a schedule that does not waste the time corresponding to the processing request generation interval can be performed.

When the processing time of each data processing device has a large error compared with the estimated time, the queue of processing requests for each data processing device is reconfigured as a whole. It can respond to changes in real time.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a first embodiment, a second embodiment, a third embodiment and a fourth embodiment of the present invention.

FIG. 2 is a flowchart showing a processing flow according to the first embodiment of the present invention.

FIG. 3 is a timing chart showing the first embodiment of the present invention.

FIG. 4 is a flowchart showing a processing flow of embodiment 2 of the present invention.

FIG. 5 is a flowchart showing a processing flow of embodiment 3 of the present invention.

FIG. 6 is a flowchart showing a processing flow of embodiment 4 of the present invention.

FIG. 7 is a configuration diagram showing a fifth embodiment of the present invention.

FIG. 8 is a flowchart showing a processing flow of embodiment 5 of the present invention.

FIG. 9 is a configuration diagram showing a sixth embodiment of the present invention.

FIG. 10 is a flowchart showing a control method of a conventional data processing device.

[Explanation of symbols]

3 processing request, 5 database processing device (data processing device),

Claims (6)

[Claims] 1. An information processing system comprising a plurality of data processing devices having different processing speeds and scheduling a plurality of processing requests in the order in which the requests are generated. And a step of estimating the processing time of the entire queue of processing requests for each of the data processing devices and the remaining time of the processing request being processed in each of the data processing devices, the remaining time and the processing time of the entire queue And a method for controlling the data processing device, comprising the step of adding the generated processing request to the queue of the data processing device having the shortest expected processing end time, which is the sum of the expected processing time. 2. The step of estimating the expected processing time in each of the data processing devices and the processing time of the entire queue of processing requests in each of the data processing devices with respect to the processing request, in order to estimate the processing time, The method of controlling a data processing device according to claim 1, wherein a data amount and load information for each data processing device are used. 3. The step of estimating the processing time of the entire queue of processing requests in each of the data processing devices,
The method for controlling a data processing device according to claim 1, wherein the elapsed time of each processing request being processed is measured, and the expected processing end time of each processing request being processed is corrected from the elapsed time. 4. The generation distribution of processing requests and the processing time of the processing request in the step of estimating the expected processing time in each of the data processing apparatuses and the processing time of the entire queue of processing requests in each of the data processing apparatuses for the processing request. 2. The method for controlling a data processing device according to claim 1, wherein the statistical information of each is referred to in the estimation. 5. The average time of the generation intervals of the processing requests and the average time of the processing times in each of the data processing devices are sampled, and when the average time of the generation intervals is larger than the average time of the processing times, the plurality of data items are collected. Each of the above-mentioned data processings is a multiplication factor of the performance ratio of the processing speed between the processing devices and a multiplication factor of the ratio of the average time of the generation intervals to the average time of the processing time as the ratio of the performance ratio of the processing speed between the plurality of data processing devices. The method for controlling a data processing device according to claim 1, wherein a processing time for each device is estimated. 6. The processing request queue for each data processing device is recombined as a whole when the processing time for each data processing device has a large error compared with the estimated time. A method for controlling a data processing device according to any one of claims 1 to 5.