JP2005250823A - Multiple computer operation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a multiple computer operation system that selects a plurality of processing computers for executing a plurality of parallel computer arithmetic operations from a plurality of registered computers and assigns each processing computer as optimizing a power charge thereon. <P>SOLUTION: The multiple computer operation system comprises a registered computer database 11 for saving a power charge on and a current use state of every registered computer 3, a registration data update part 12 for updating the registration data in the registered computer database 11, an arithmetic operation input part 13 for receiving an input of a plurality of computer arithmetic operations, and an arithmetic operation assignment part 14 for selecting processing computers 3 equal in number to the plurality of input computer arithmetic operations in ascending order of power charge per unit computational complexity at the input time, and assigning the selected processing computers 3 to the plurality of input computer arithmetic operations. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention provides a plurality of computer operation systems for selecting and assigning a plurality of processing computers for executing a plurality of computer arithmetic processes in parallel from a plurality of registered computers, and information for the assignment to the user The present invention relates to a multiple computer operation system.

  Computer usage fees for providing computer processing services to external users are generally set separately for fixed basic fees and metered fees, and metered fees are charged for individual computer processing. Is done. In addition, the pay-per-use charge is generally calculated according to the unit price of time by the calculation time required for the computer calculation process and the used capacity of the storage device.

  In the case of parallel computing processing in which computer computing processing is divided and executed by multiple computers distributed on a network such as grid computing, the total usage fee for each computer is the computer usage fee for the computer computing processing. .

In the parallel processing described above, depending on how the computer processing is divided into a plurality of processing units and which processing unit is assigned to which computer, the degree of congestion, processing speed, and calculation time of each computer Depending on the unit price, there will be a difference in the end time and fee of the computer arithmetic processing. Therefore, when priority is given to the delivery date, the computer arithmetic processing is divided and assigned so that as many arithmetic processing units as possible are not congested or assigned to a computer with a high processing speed. Also, if you want to keep the computer usage fee low, you can assign the divided processing unit to an inexpensive computer with a pay-as-you-go fee, and optimize the computer processing including the communication cost for transferring the processing unit at that time to each computer A process division / allocation method has been proposed (for example, see Patent Document 1 below).
Special table 2001-503170 gazette

  The basic fee or metered fee for using the computer is determined by allocating the basic fee or metered fee, for example, referring to past performance, so that some or all of the cost required to operate the computer is borne by the user. . Therefore, as a cost required for the operation of the computer, a large rate of electricity for operating the computer accounts for a large proportion in a large computer or a supercomputer.

  By the way, in the case of the above parallel operation processing, since a plurality of computers used for the operation processing are distributed on the network, when the network is spread all over the world such as the Internet, the plurality of computers are also Will be distributed throughout the world. In addition, a plurality of arithmetic processing units processed at the same time are processed in the daytime in one computer and processed in the nighttime in another computer. Therefore, the power charge in the execution time zone of each arithmetic processing unit varies depending on the region and time zone for each computer. Further, even when the computer arithmetic processing is not divided into parallel arithmetic processing and processed by one computer, the power charge required for the arithmetic processing varies depending on the execution time zone.

  However, while the electricity charge varies depending on the region and time, the pay-per-use charge for each computer calculation process is a fair charge reflecting the electricity charge corresponding to the power consumption required for the calculation process. There is no current situation. Therefore, even if the sum of the metered charges for the individual computer calculation processes is simply optimized, it cannot be said that the power charge associated with the power consumption of each computer that has executed the computer calculation processes is optimized.

In addition, the use of computers has been increasing year by year due to the recent global interest in global environmental conservation, but how to reduce energy consumption and CO ₂ generation due to the power consumed by computers distributed on computer networks. At present, no attempt has been made to make this possible.

The present invention has been made in view of the above-mentioned problems, and its purpose is to select a plurality of processing computers for executing a plurality of computer arithmetic processes in parallel from a plurality of registered computers, Multiple computer operation system that optimizes and allocates power charges, power consumption, or CO ₂ generation amount related to power consumption of processing computers, and further, power charges and power consumption related to power consumption of each processing computer Alternatively, an object is to provide a multiple computer operation system that provides information to a user for optimizing and allocating the amount of CO ₂ generated.

In order to achieve the above object, a multiple computer operation system according to the present invention is a multiple computer operation system in which a plurality of processing computers for executing a plurality of computer arithmetic processes in parallel are selected and assigned from a plurality of registered computers. And receiving an input of the plurality of computer calculation processes and a registered computer database storing an amount to be optimized per predetermined unit calculation amount related to power consumption for each registered computer and a current use state An arithmetic processing input unit selects the same number of processing computers as the plurality of input computer arithmetic processing in ascending order of optimization amount per unit calculation amount at the time of input, and the selected processing computers are input An arithmetic processing assigning unit for assigning to the plurality of computer arithmetic processings. To. Furthermore, it is preferable that the amount to be optimized is any one of a power rate, a power consumption amount, and a CO ₂ generation amount.

According to the multiple computer operation system of the first feature, the optimization amount per unit calculation amount of the plurality of processing computers assigned by the arithmetic processing assignment unit to the plurality of computer arithmetic processings received by the arithmetic processing input unit is Since it is selected as the lowest combination at the time of input, it is related to the power consumption of each processing computer accompanying all execution of a plurality of computer arithmetic processing by putting each computer arithmetic processing into each assigned processing computer For example, a total such as a power rate, power consumption, or CO ₂ generation amount is optimized. As a result, if the amount to be optimized is a power charge, the difference in the power charge for each time zone of each computer executing the computer calculation process is reflected, and for example, a calculation process using an inexpensive power charge at night is performed. Therefore, the calculation cost can be reduced for the user, and the power peak can be reduced. Further, when the amount to be optimized is power consumption, CO ₂ generation amount, etc., it can contribute to power peak reduction and global environment conservation.

Note that the optimization amount per unit calculation amount related to the power consumption of each registered computer may be acquired from each registered computer, and the optimization amount per unit calculation time is divided by the calculation speed. You may calculate by. In the latter case, specifically, when the amount to be optimized is a power charge, a power charge per unit calculation time can be calculated from the power consumption of each computer and the power unit price. Also, the power consumption per unit calculation time can be calculated from the power consumption of each computer. As for the amount of produced CO ₂ with respect to power consumption, the difference in the power generation method and the like for each time period, since the relationship between the CO ₂ emissions are separately provided for the time period separately unit power consumption, the given CO ₂ generation Based on the correspondence between the amounts, the CO ₂ generation amount per unit calculation time can be calculated.

  Furthermore, in addition to the first feature, the multiple-computer operation system is configured such that the arithmetic processing allocation unit has the usage status in a free state or a waiting time within a predetermined time from the registered computers at the time of input. The second feature is to select a processing computer.

  According to the multiple-computer operation system of the second feature, since a plurality of computer arithmetic processes can be started immediately or within a predetermined time, the total amount of optimization in each computer can be minimized, In addition, it contributes to the early termination of a plurality of computer arithmetic processes.

  Furthermore, in addition to the first or second feature, the multiple computer operation system is configured to calculate the processing computer from the calculation time and amount of the computer arithmetic processing executed by the processing computer selected by the arithmetic processing allocation unit. A calculation speed calculation unit for calculating an optimization amount per unit calculation amount based on the calculated calculation speed and registering the calculation amount in the registered computer database; And

  According to the multi-computer operation system of the third feature, when the calculation speed is different from the catalog value of the computer or when there is no catalog value, the calculation speed according to the actual result can be obtained. Therefore, it is possible to use the optimization amount per unit calculation amount related to the power consumption of each registered computer, so that it is possible to select and assign processing computers that minimize the optimization amount with high accuracy.

Further, the multiple computer operation system selects information to select and assign a plurality of processing computers for executing a plurality of computer computing processes in parallel from a plurality of registered computers, and performs the plurality of computer computing processes. A multi-computer operating system provided to a requesting user, wherein the registration computer stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer A database, a registration data update unit that detects a change in at least one of the optimization amount per unit calculation amount, the calculation speed, and the usage state, and updates the registration computer database; Provide data registered in the registered computer database to the user terminal And recording data providing unit, to become comprise a fourth feature. Furthermore, it is preferable that the amount to be optimized is any one of a power rate, a power consumption amount, and a CO ₂ generation amount.

According to the multi-computer operating system of the fourth feature, a registration registered in a registered computer database for a user who wants to execute a plurality of computer arithmetic processes separately on a plurality of processing computers and execute them in parallel. Since it is possible to provide the latest data on the amount of optimization per unit calculation related to the power consumption of each computer, the calculation speed, and the current usage status, the user can select from among a plurality of registered computers. Necessary information on which computer should be selected and assigned as a processing computer can be obtained. As a result, for example, if the amount to be optimized is a power charge, the user can minimize the power charge required for a plurality of computer calculation processes. In addition, if the amount to be optimized is power consumption, the user can minimize the power consumption required for a plurality of computer calculation processes. Furthermore, if the amount to be optimized is CO ₂ generation amount, The amount of CO ₂ generated by multiple computer processing can be minimized, contributing to global environmental conservation.

  Further, in the computer operating system, in addition to the fourth feature, the registration data providing unit transmits registration data corresponding to a request time point to the user terminal in response to a data provision request from the user terminal. This is a fifth feature. Furthermore, in addition to the fourth or fifth feature, the plurality of computer operating system is configured such that the registration data providing unit sends registration data to the user terminal every time the registration data updating unit updates the registration computer database. Transmission is the sixth feature.

  Efficient registration data can be provided by the multiple computer operation system having the fifth or sixth characteristics.

Further, the multiple computer operation system selects information to select and assign a plurality of processing computers for executing a plurality of computer computing processes in parallel from a plurality of registered computers, and performs the plurality of computer computing processes. A multi-computer operating system provided to a requesting user, wherein the registration computer stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer A database, a registration data update unit that detects a change in at least one of the optimization amount per unit calculation amount, the calculation speed, and the usage state, and updates the registration computer database; From the user terminal, the characteristic information that can specify the amount of calculation of the plurality of computer arithmetic processes and the target end time A calculation processing information input unit for receiving a force, and a calculation amount of each computer calculation process is predicted from the characteristic information, and a total of the optimized amounts of the computer calculation processes is an expected end time of each computer calculation process Select the same number of the processing computers as the plurality of computer arithmetic processes so that the latest one of the plurality of computer arithmetic processes becomes the minimum within a range not exceeding the target end time, and the selected processing computers are used as the plurality of computer arithmetic processes. According to a seventh aspect, the present invention includes a first arithmetic processing allocation unit to be allocated and a processing result information providing unit that provides the processing result of the first arithmetic processing allocation unit to the user terminal. Furthermore, it is preferable that the amount to be optimized is any one of a power rate, a power consumption amount, and a CO ₂ generation amount.

  According to the multiple-computer operating system of the seventh feature, the selection and assignment of processing computers that minimize the total amount of optimization required for computer arithmetic processing within a range that does not exceed the input target end time is the first. Since it is performed by the arithmetic processing assigning unit, it is possible to provide a user who wants to execute a plurality of computer arithmetic processes separately in a plurality of processing computers by executing selection and assignment of the processing computer as information. As a result, according to the selection and assignment of the processing computer provided by the information, the user inputs the plurality of computer arithmetic processing into each processing computer, thereby performing the arithmetic processing with the minimum amount to be optimized by the target end time. Can be executed.

Further, the multiple computer operation system selects information to select and assign a plurality of processing computers for executing a plurality of computer computing processes in parallel from a plurality of registered computers, and performs the plurality of computer computing processes. A multi-computer operating system provided to a requesting user, wherein the registration computer stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer A database, a registration data update unit that detects a change in at least one of the optimization amount per unit calculation amount, the calculation speed, and the usage state, and updates the registration computer database; Characteristic information capable of specifying the amount of calculation of the plurality of computer arithmetic processes from the user terminal and the amount to be optimized An arithmetic processing information input unit that accepts an input of an upper limit value, a calculation amount of each computer calculation process is predicted from the characteristic information, and a range in which the total of the optimization target amounts of the computer calculation processes does not exceed the upper limit value And selecting the same number of processing computers as the plurality of computer arithmetic processes so that the expected end time of the plurality of computer arithmetic processes is the earliest, and assigning the selected processing computers to the plurality of computer arithmetic processes. An eighth feature is that it comprises a two-operation processing allocation unit and a processing result information providing unit that provides the processing result of the second arithmetic processing allocation unit to the user terminal. Furthermore, it is preferable that the amount to be optimized is any one of a power rate, a power consumption amount, and a CO ₂ generation amount.

  According to the multiple-computer operating system of the eighth feature, the selection and assignment of the processing computer with the earliest expected end time of the computer arithmetic processing within the range not exceeding the upper limit of the input optimization amount is the second. Since it is performed by the arithmetic processing assigning unit, it is possible to provide a user who wants to execute a plurality of computer arithmetic processes separately in a plurality of processing computers by executing selection and assignment of the processing computer as information. As a result, according to the selection and assignment of the processing computer provided by the information, the user inputs a plurality of computer arithmetic processing into each processing computer, so that the optimization amount accompanying the arithmetic processing exceeds a predetermined upper limit value. As long as there is no limit, the calculation process can be executed at the highest speed.

  Further, in the multiple computer operation system, in addition to any one of the fourth to eighth features, the registration data update unit inputs a calculation time and a calculation amount of a computer calculation process executed by the processing computer. The ninth feature is to accept and calculate the calculation speed.

  According to the multiple-computer operating system of the ninth feature, since the registered data update unit calculates a calculation speed according to the actual result, the optimization target per unit calculation amount related to the power consumption of each registered computer according to the actual result High-accuracy processing computers can be selected and assigned using the conversion amount.

  Further, in the multiple computer operation system, in addition to any one of the first to ninth features, the amount to be optimized is a power charge, and a power charge per unit calculation amount for each registered computer is as follows: A tenth power supply unit for receiving a bid power charge bid by each owner of the registered computer and sequentially registering the bid power charge in the registered computer database for each input; Features.

  According to the multiple-computer operating system of the tenth feature, the owner of the registered computer bids a cheaper or more expensive bid power fee instead of the actual power fee as the power fee per unit calculation amount. By inputting to the power charge input unit, the registered computer uses the bid power charge as the power charge per unit calculation amount, and is used for selection and allocation of the processing computer. As a result, if the owner of a registered computer wants to actively promote the use of his / her own computer, it is possible by cheap pricing, and conversely, the probability that his / her computer will be used from the usage status of other registered computers. When the price is high, it is possible to increase the revenue of the electricity bill by performing expensive pricing.

  An embodiment of a multiple computer operation system according to the present invention (hereinafter referred to as “the present invention system” as appropriate) will be described with reference to the drawings.

<First Embodiment>
As shown in FIG. 1, a plurality of computing nodes 3 that process a plurality of users (user terminals) 2 and a plurality of computing operations (hereinafter referred to as “jobs” for convenience) of a plurality of users 2 in a distributed manner. The present invention system 10 for assigning a computing node 3 (processing computer) for executing the processing to each job is connected to a computer network 4 such as the Internet so as to be able to transmit / receive data to / from each other.

  As shown in FIG. 1, in the grid computing in which a plurality of jobs of a user 2 are distributed and processed in a plurality of calculation nodes 3, the system 10 of the present invention is connected to all the calculation nodes 3 ( A process of selecting and assigning a specific computing node 3 (processing computer) that actually executes the job from among the registered computers is executed by program processing on the computer 5 constituting the system 10 of the present invention.

  As shown in FIG. 2, the system 10 of the present invention includes a registered computer database 11, a registered data update unit 12, a calculation processing input unit 13, a calculation processing allocation unit 14, and a power rate database 15.

  As shown in FIG. 3, the registered computer database 11 includes the latest power charge per unit calculation amount (unit price of power calculation amount) [yen / TFLOPS] for each calculation node 3 and the latest use status (free state or use state). Is stored in a predetermined storage area of the storage device of the computer 5. The power calculation unit price changes for each calculation node 3 and for each time zone according to the power unit price [yen / kWh] for each time zone of the power company to which each calculation node 3 receives power supply.

  As shown in FIG. 4, the power rate database 15 is formed by arranging the power unit price [yen / kWh] for each time zone by calculation node 3 and storing it in a predetermined storage area of the storage device of the computer 5. . As shown in FIG. 4, the time zone for defining the power unit price differs for each calculation node 3. For example, in a certain calculation node (node N1), the power unit for daytime from 10:00 to 18:00 is 13 yen / kWh. The power unit price at night in other time zones (18:00 to 10:00 on the next day) is 11 yen / kWh, which is cheaper than the daytime, and the other daytime time zone is 6 for other calculation nodes (node N2). The time is set to 20:00. This difference is because the charge system of the electric power company that supplies power to each calculation node 3 is different. In addition, each power unit price may vary depending on the season. In the example shown in FIG. 4, all the calculation nodes are located in the northern hemisphere, and the power unit price in summer is set high. The power rate database 15 further stores a power rate (unit price per hour) [yen / second] per unit usage time of each calculation node 3 calculated based on the power unit price and power consumption for each time zone. . The power calculation unit price [yen / TFLOPS] is simply calculated by dividing the power hour unit price [yen / second] by the calculation speed [TFLOPS / second] of each calculation node 3. In the present embodiment, it is assumed that the calculation speed of each calculation node 3 uses a pre-registered calculation speed.

  When the power unit price for each time zone of each calculation node 3 changes, or when the usage status changes as a result of a job being submitted to a specific calculation node 3 by the arithmetic processing assignment unit 14 as described later, the registered computer database 11 needs to be updated. For this reason, the registration data updating unit 12 accesses the power rate database 15 and stores in advance the change time of the power unit price for each time zone of each calculation node 3, and each time the change time comes, The power calculation amount unit price is calculated as described above, and the power calculation amount unit price of the calculation node 3 in the registered computer database 11 is updated. The registered data update unit 12 also checks the usage status of the corresponding computing node 3 in the registered computer database 11 every time the arithmetic processing allocation unit 14 submits a new job and every time the submitted job ends. Update.

  The arithmetic processing input unit 13 receives a plurality of jobs that each user 2 wants to execute in parallel via the computer network 4. In the present embodiment, it is assumed that jobs to be processed in parallel are divided into a plurality of jobs in advance for each user 2.

  The arithmetic processing allocation unit 14 is activated when the arithmetic processing input unit 13 receives a new job, selects the same number of calculation nodes 3 as a plurality of jobs input in accordance with an algorithm described later, and selects the calculation selected for each job. Node 3 is assigned separately. In this case, selection and allocation of the calculation nodes 3 are executed by the above algorithm so that the total power charge related to power consumption associated with execution of each job in each selected calculation node is minimized.

  Hereinafter, the operations of the system 10 of the present invention, the user 2, and the calculation node 3 will be described with reference to FIG.

  First, the user 2 (user A in FIG. 5) transmits a plurality of jobs (JA1 to JA3) to the system 10 of the present invention (# 1). Subsequently, the arithmetic processing input unit 13 receives input of a plurality of jobs (JA1 to JA3) (# 2). Upon receiving the input, the arithmetic processing allocation unit 14 is activated, and the arithmetic processing allocation unit 14 accesses the registered computer database 11 and extracts the latest registered data (power calculation unit price, usage status) (# 3). . It is assumed that the extraction result is in the state illustrated in FIG.

  Next, the calculation nodes 3 are rearranged in ascending order of the extracted power calculation unit price, and the same number of calculation nodes 3 in which the usage status is empty (“3” in the example of FIG. 5) are selected (# 4). ). Here, in the illustration of FIG. 3, since the power calculation unit price is small in the order of the calculation nodes N2, N3, N5, the calculation nodes N2, N3, N5 are selected. Next, the calculation amount [TFLOPS] of each job (JA1 to JA3) is predicted based on the type and calculation size of the job, and the calculation node 3 with the smaller unit price of power calculation amount is allocated in order from the job with the larger calculation amount. (# 5). If the calculation amount of job JA1 is the largest, calculation node N2 is assigned. Here, in a situation where the calculation amount of each job is determined to be substantially equal, the calculation nodes may be allocated to each job in an appropriate order without predicting the calculation amount.

  Next, the system 10 of the present invention inputs the jobs JA1 to JA3 to the calculation nodes N2, N3, and N5 assigned by the arithmetic processing assignment unit 14 (# 6). When each job is submitted to each of the calculation nodes N2, N3, and N5, the usage status of each calculation node changes from the empty status to the usage status. The usage status of the computation nodes N2, N3, N5 in the registered computer database 11 is changed to the usage status (in use), and the registered computer database 11 is updated (# 7).

  Subsequently, each of the calculation nodes N2, N3, and N5 executes the job that has been input, and when each job is completed, the result is transmitted to the user A who is the source of the jobs JA1 to JA3 (# 8). At the same time, data indicating the end of the job is transmitted to the system 10 of the present invention (# 9). Since the end timing of each job at each computation node N2, N3, N5 varies depending on the computation speed of each computation node and the amount of computation of each job, the registered data update unit 12 is configured to use the computation nodes N2, N3, N3. Each time the job end data is received from N5, the usage status of the received calculation node is sequentially changed to an empty state, and the registered computer database 11 is updated (# 10).

Second Embodiment
Next, a second embodiment of the system of the present invention will be described. As shown in FIG. 1, the inventive system 20 according to the second embodiment is connected to a plurality of users (user terminals) 2 and a plurality of computing nodes 3 via a computer network 4 as in the first embodiment. doing. The system 20 of the present invention is the latest registration of all the computing nodes 3 (registered computers) connected to the computer network 4 in grid computing in which a plurality of jobs of a user 2 are distributed and processed in a plurality of computing nodes 3. Processing for providing data to the user 2 is executed by program processing on the computer 5 constituting the system 20 of the present invention.

  As shown in FIG. 7, the system 20 of the present invention includes a registration computer database 21, a registration data update unit 22, a calculation processing information input unit 23, a registration data provision unit 24, and a power rate database 25. The power rate database 25 has the same configuration as the power rate database 15 of the first embodiment.

  As shown in FIG. 8, the registered computer database 21 includes the latest power charge per unit calculation amount (unit price of power calculation amount) [yen / TFLOPS], the calculation speed [TFLOPS / second], and the latest Is used in a predetermined storage area of the storage device of the computer 5. The power calculation unit price changes for each calculation node 3 and for each time zone according to the power unit price [yen / kWh] for each time zone of the power company to which each calculation node 3 receives power supply. Similarly to the first embodiment, the power calculation unit price is obtained by dividing the power hour unit price [yen / second] stored in the power rate database 25 by the calculation speed [TFLOPS / second] of each calculation node 3. Is easily calculated. In the second embodiment, the calculation speed of each calculation node 3 is registered in the registered computer database 21, and the registered data update unit 22 performs the calculation time and the calculation amount of the job executed by each calculation node 3. The calculation speed calculated based on the value is used. Therefore, the registered data updating unit 22 receives the input of the actual value from each calculation node 3 every time the calculation is completed or periodically, and updates the calculation speed of the corresponding calculation node 3 in the registered computer database 21.

  Further, similarly to the registration data update unit 12 of the first embodiment, the registration data update unit 22 accesses the power rate database 25 and stores in advance the change time of the power unit price for each time zone of each calculation node 3. In addition, every time the change period comes, the power calculation unit price is calculated as described above, and the power calculation unit price of the calculation node 3 in the registered computer database 21 is updated. In addition, the registration data update unit 22 causes the arithmetic processing information input unit 23 to receive information from each user 2 or each calculation node 3 every time each user 2 submits a new job and every time the entered job ends. Based on the characteristic information relating to the received input or completed job, the usage status of the corresponding calculation node 3 in the registered computer database 11 is updated. The waiting time after the job is changed to the use state from the empty state is predicted by the registered data update unit 22 from the calculation amount of the input job and the calculation speed of the input calculation node. The calculation amount of the input job is estimated by a predetermined estimation algorithm based on the characteristic information (job type, program size, etc.) received by the arithmetic processing information input unit 23.

  The calculation processing information input unit 23 receives from the user 2 or each calculation node 3 the characteristic information used for estimating the calculation amount of the job submitted by each user 2 to each calculation node 3, and sends it to the registered data update unit 22. Transfer the information.

  Upon receiving a data provision request from each user 2, the registration data providing unit 24 extracts the latest registration data corresponding to the request time in response to the request, extracts the registered computer database 21, and requests the requested user. 2 to send. The registration data providing unit 24 transmits the registration data to each user 2 or the user 2 who desires automatic distribution of registration data every time the registration data update unit 22 updates the registration computer database 21.

  As described above, since the user 2 can obtain the latest registration data in the registered computer database 21, each user can determine which calculation node 3 the job should be submitted to based on the registration data.

<Third Embodiment>
Next, a third embodiment of the system of the present invention will be described. As shown in FIG. 1, the present invention system 30 according to the third embodiment is similar to the first and second embodiments in that a plurality of users (user terminals) 2 and a plurality of computing nodes are connected via a computer network 4. 3 is connected. In the grid computing in which a plurality of jobs of a user 2 are distributed and processed in a plurality of calculation nodes 3, the system 30 of the present invention specifies that the job is actually executed from all the calculation nodes 3 (registered computers). A computer constituting the system 30 of the present invention in response to a processing request from each user 2 for two types of processing for providing the user 2 with information for selecting and assigning the computing node 3 (processing computer) 5 is executed by program processing.

  As shown in FIG. 9, the present system 30 includes a registered computer database 31, a registered data update unit 32, an operation processing information input unit 33, a first operation processing allocation unit 34, a second operation processing allocation unit 35, and processing result information. A providing unit 36 and a power rate database 37 are provided. The registered computer database 31, registered data update unit 32, and power charge database 37 have the same configuration as the registered computer database 21, registered data update unit 22, and power charge database 25 of the second embodiment. Duplicate explanations are omitted.

  First, processing operations of the arithmetic processing information input unit 33, the first arithmetic processing allocation unit 34, and the processing result information providing unit 36 will be described with reference to FIG.

  First, the characteristic information and target end times of a plurality of jobs (JA1 to JA3) scheduled to be submitted by the user 2 (user A in FIG. 10) are transmitted to the present invention system 30 (# 1). Subsequently, the arithmetic processing information input unit 33 accepts input of characteristic information and target end times of a plurality of jobs (JA1 to JA3) (# 2). In response to the input, the first arithmetic processing allocation unit 34 is activated, and the first arithmetic processing allocation unit 34 first calculates the calculation amount [TFLOPS of each job based on the input characteristic information of each job (JA1 to JA3). ] Is estimated by a predetermined estimation algorithm (# 3). Next, the registered computer database 31 is accessed to extract the latest registration data (power calculation unit price, calculation speed, usage status) (# 4).

  Next, the calculation nodes 3 are rearranged in ascending order of the power calculation amount unit price of the extracted registration data, and the same number as the job (“3” in the example of FIG. 5) is provisionally selected (# 5). Next, in order from the job with the estimated large calculation amount, the calculation node 3 with the small electric power calculation unit price is allocated (# 6). For each job, the calculation time is calculated by dividing the estimated calculation amount by the calculation speed of the calculation node 3 to which the job is assigned, and the end time is calculated by adding the waiting time of the usage status of the calculation node 3 (0 hours in the empty state). Is predicted (# 7). Then, for some or all of the plurality of jobs, it is determined whether there is a job whose predicted end time exceeds the target end time (# 8). Here, for some or all of a plurality of jobs, if there is a job whose predicted end time exceeds the target end time, a job with a large calculation amount is calculated as a calculation node 3 with a high calculation speed or a calculation node 3 with a short waiting time. The process of replacing with the calculation node 3 with the smallest unit price of power calculation is performed (# 9), and the above steps # 7 to # 9 are repeated until the predicted end time does not exceed the target end time. The calculation node 3 is allocated within a range where the predicted end time does not exceed the target end time for all jobs. As a result, the allocation that minimizes the total power charge of the selected computation node 3 is realized in a range where the predicted end time does not exceed the target end time for all jobs. Next, the power charge of each job is calculated from the estimated power amount of the calculation node 3 assigned to each job and the estimated calculation amount (# 10).

  Next, the processing result information providing unit 36 processes the processing result information for each job (JA1 to JA3) (calculation nodes 3 assigned by the first arithmetic processing allocation unit 34 (in the example of FIG. 10, the calculation nodes N2, N3, N5). ), The predicted end time and the power charge) are transmitted to user 2 (user A in FIG. 10) (# 11). The user 2 examines whether or not to submit the jobs (JA1 to JA3) based on the processing result information, and if there is no problem, submits each job to the calculation node 3 assigned by the first arithmetic processing assignment unit 34 (# 12 At the same time, the input information indicating to which computing node each job is input or the confirmation information of the processing result information is transmitted to the system 30 of the present invention (# 13). As a result, the usage status of each calculation node 3 changes from a vacant state to a usage status, or the waiting time is extended, so that the registration data update unit 22 causes the calculation computer nodes N2, N3, The usage status of N5 is changed and the registered computer database 11 is updated (# 14).

  Subsequently, each computation node 3 (N2, N3, N5) executes the job that has been submitted, and when each job is completed, the result is transmitted to the user A who is the source of the jobs JA1 to JA3 (#). 15). The registration computer database 21 is updated by the registration data update unit 22 in response to a change in the usage status of each computation node 3 associated with the execution and termination of each job. The processing is as described in the second embodiment. Because there is, the explanation which overlaps is omitted.

  Next, processing operations of the arithmetic processing information input unit 33, the second arithmetic processing allocation unit 35, and the processing result information providing unit 36 will be described with reference to FIG.

  First, the user 2 (user A in FIG. 11) transmits the characteristic information of the plurality of jobs (JA1 to JA3) to be submitted and the upper limit value of the power charge to the present system 30 (# 1). Subsequently, the arithmetic processing information input unit 33 accepts the input of the characteristic information of the plurality of jobs (JA1 to JA3) and the upper limit value of the power charge (# 2). By receiving the input, the second arithmetic processing allocation unit 35 is activated, and the second arithmetic processing allocation unit 35 first calculates the calculation amount [TFLOPS of each job based on the input characteristic information of each job (JA1 to JA3). ] Is estimated by a predetermined estimation algorithm (# 3). Next, the registered computer database 31 is accessed to extract the latest registration data (power calculation unit price, calculation speed, usage status) (# 4).

  Next, a calculation node list in which the calculation nodes 3 are rearranged in the descending order of calculation speed of the registered data extracted for each job is created (# 5). Here, the calculation node 3 whose usage status is in the usage state performs a correction to reduce the calculation speed based on the calculation amount of the corresponding job. Next, in order from the estimated job with the largest calculation amount, the calculation node 3 with the fast calculation speed is allocated, the calculation node 3 is deleted from the calculation node list of other jobs, and the allocation is performed sequentially (# 6). For each job, a calculation time is calculated by dividing the estimated calculation amount by the calculation speed of the calculation node 3 to which the job is assigned, and an end time is predicted (# 7). If the job with the latest end time is not the job with the largest amount of calculation, replace the job with a larger amount of calculation with a calculation node to check whether the latest end time is earlier, and end the latest If the time is earlier, the replacement is executed to correct the assignment (# 8). Next, for each job, a power charge is calculated from the estimated power calculation amount of the assigned calculation node 3 and the estimated calculation amount, and a total is calculated for all jobs (# 9). It is determined whether or not the upper limit value is below (# 10). Here, if the total exceeds the upper limit, power calculation is performed for jobs with a large amount of calculation within the range that does not exceed the latest end time for jobs other than the job with the latest end time. The replacement of the calculation node 3 is performed so that the calculation node 3 with a low unit price is allocated (# 11), the total of the power charges is calculated again for all jobs (# 12), and the power charge for which the total is input It is determined whether it is below the upper limit value (# 13). Here, if the total still exceeds the upper limit value, including the job with the latest end time, the target value of the end time is gradually delayed while the calculation amount is within the range not exceeding the target value. Replacement of the calculation node 3 is performed so that the calculation node 3 with a low power calculation unit price is allocated to a large job (# 14), and the total of the power charges is calculated again for all jobs (# 15). It is determined whether or not the sum is equal to or less than the upper limit value of the input power charge (# 16). Here, if the total still exceeds the upper limit, the processes of steps # 14 to # 16 are repeated until the total is still below the upper limit. As a result, the allocation of the calculation node 3 with the earliest end time is realized within a range where the total power charge of all jobs does not exceed the upper limit value.

  Next, the processing result information providing unit 36 processes the processing result information for each job (JA1 to JA3) (calculation nodes 3 (calculation nodes N2, N3, N5 in the example of FIG. 11) allocated by the second arithmetic processing allocation unit 35). ), The predicted end time and the power charge) are transmitted to the user 2 (user A in FIG. 11) (# 17). Subsequent processing is the same as that described with reference to FIG. 10, and therefore redundant description is omitted.

  Hereinafter, another embodiment will be described.

<1> In each of the above embodiments, the power charge per unit calculation amount (unit price of power calculation amount) is used as the registration data of the registered computer databases 11, 21, 31. It may be consumption [Wh / TFLOPS] or CO ₂ generation [t / TFLOPS]. In this case, in the first embodiment, calculation node allocation that minimizes the power consumption or CO ₂ generation amount is executed. In the third embodiment, the first arithmetic processing allocation unit 34 ends prediction for all jobs. As long as the time does not exceed the target end time, the allocation that minimizes the sum of the power consumption or CO ₂ generation amount of the selected computation node 3 is executed. in a range that does not exceed the upper limit value of power consumption or CO ₂ the power consumption total is input generation amount or the amount of produced CO _2, early allocation of computing nodes 3 are realized most end time. The power consumption Sadamari the power consumption of the computer constituting each computing node is constant regardless of the time zone, CO ₂ emissions, even if different time zone in power consumption is the same, the power company Since the power generation equipment to be used varies depending on the total amount of power supplied, it varies depending on the time zone. Therefore, for each calculation node, based on the correspondence relationship between the power consumption of the computer, the power consumption for each time zone of the power company to be used, and the CO ₂ generation amount, the same as the power rate database 15, 25, 37. Using the database, the CO ₂ generation amount per unit power consumption is registered, and the registered data update units 12, 22, and 32 access the database, and the unit power for each time zone of each calculation node 3 The change time of the CO ₂ generation amount per consumption is stored in advance, and the CO ₂ generation amount per unit calculation amount of the calculation node 3 of the registered computer database 11 is updated every time the change time comes.

Further, the system 2 of the present invention 10, 20, 30 is configured so that the user 2 can select the power charge, the power consumption, and the CO ₂ generation amount as the optimization target related to the power consumption used for the allocation process of the calculation nodes. It is also preferable.

  <2> In each of the above embodiments, the power charge per unit calculation amount (power calculation unit price) is calculated based on the power consumption of each calculation node, the calculation speed, and the power unit price for each time zone. Although it is a unique value depending on the attribute, it is also preferable that the unit price of power calculation amount can be changed according to the intention of the owner of each calculation node. Specifically, the present invention systems 10, 20, and 30 accept the input of the bid power rate bid by each owner of the computation node (registered computer) as the power calculation amount unit price for each computation node, and accept the input. A bidding power rate input unit for registering the bidding power rate in the registered computer database is provided for each time.

  <3> In the first embodiment, the registered computer database 11 includes the latest power charge (unit price of power calculation amount) [yen / TFLOPS] per unit calculation amount for each calculation node 3 and the latest usage status (empty state). Alternatively, the calculation speed for each calculation node 3 may be stored as in the second or third embodiment.

  <4> In the second embodiment, the timing at which the registration data providing unit 24 provides the registration data to the user 2 is not limited to the timing illustrated in the second embodiment. In addition, when the registration data providing unit 24 receives a data provision request from each user 2, the registration data providing unit 24 provides registration data only for one of the cases where the registration data updating unit 22 updates the registered computer database 21. It doesn't matter.

  <5> In the third embodiment, the case where both the first arithmetic processing allocation unit 34 and the second arithmetic processing allocation unit 35 are provided has been described, but a configuration including only one of them may be employed. Absent. Further, the registration data providing unit 24 of the second embodiment may be provided in the configuration of the third embodiment. In this case, both functions of the processing result information providing unit 36 and the registration data providing unit 24 are combined to exhibit an information providing function for the user.

  <6> In the third embodiment, calculation node assignment processing procedures of the first arithmetic processing assignment unit 34 and the second arithmetic processing assignment unit 35 are not limited to the processing steps of the third embodiment.

  <7> In each of the above embodiments, an input of a power unit price for each time zone of an electric power company that supplies power to the location of each calculation node 3 is received in advance, and the power unit price for each time zone is determined as the location of each calculation node 3. It is also preferable to store them in the power rate databases 15, 25, 37 for each electric power company so that they can be searched. In this way, if the power unit price for each power company is stored in a database, the power unit price can be updated for each power company, so it can be used for multiple computing nodes that receive power from the same power company. Therefore, it is not necessary to individually update the power unit price and the power hour unit time, which is convenient.

The figure which shows the interconnection relation of the computer utilized for the grid computing in each embodiment of the multiple computer operation system which concerns on this invention The block diagram which shows schematic structure in 1st Embodiment of the multiple computer operation system which concerns on this invention. The figure which shows an example of the registration data of the registration computer database in 1st Embodiment of the multiple computer operation system which concerns on this invention The figure which shows an example of the registration data of the electric power rate database in each embodiment of the multiple computer operation system which concerns on this invention The block diagram which shows the flow of the process in 1st Embodiment between the multiple computer operation system which concerns on this invention, the computer which performs computer arithmetic processing, and a user The figure which shows an example of the registration data of the registration computer database in 1st Embodiment of the multiple computer operation system which concerns on this invention The block diagram which shows schematic structure in 2nd Embodiment of the multiple computer operation system which concerns on this invention. The figure which shows an example of the registration data of the registration computer database in 2nd or 3rd embodiment of the multiple computer operation system which concerns on this invention The block diagram which shows schematic structure in 3rd Embodiment of the multiple computer operation system which concerns on this invention. The block diagram which shows the flow of the 1st process in 3rd Embodiment between the multiple computer operation system which concerns on this invention, the computer which performs computer arithmetic processing, and a user between each other The block diagram which shows the flow of the 2nd process in 3rd Embodiment between the multiple computer operation system which concerns on this invention, the computer which performs computer arithmetic processing, and a user

Explanation of symbols

10, 20, 30: Multiple computer operation system according to the present invention 2: User (user terminal)
3: Compute node (computer)
4: Computer network (Internet)
5: Computers constituting a multiple computer operation system 11, 21, 31: Registered computer database 12, 22, 32: Registered data update unit 13: Operation processing input unit 14: Operation processing allocation unit 15, 25, 37: Electricity rate database 23, 33: Arithmetic processing information input unit 24: Registered data providing unit 34: First arithmetic processing allocating unit 35: Second arithmetic processing allocating unit 36: Processing result information providing unit

Claims (11)

A multiple computer operation system for selecting and assigning a plurality of processing computers for executing a plurality of computer arithmetic processes in parallel from a plurality of registered computers,
An amount to be optimized per predetermined unit calculation amount related to power consumption for each registered computer, and a registered computer database storing current usage conditions;
An arithmetic processing input unit for receiving inputs of the plurality of computer arithmetic processes;
Select the same number of processing computers as the plurality of input computer operation processes in ascending order of the amount of optimization per unit calculation amount at the time of input, and the plurality of computer operations input to the selected processing computer An arithmetic processing assignment unit to be assigned to the process;
A multi-computer operating system characterized by comprising:   2. The plurality of computers according to claim 1, wherein the arithmetic processing allocating unit selects the processing computers whose usage status is empty or waiting time within a predetermined time at the time of input from among the registered computers. Operational system.   The calculation speed of the processing computer is calculated from the calculation time and calculation amount of the computer calculation process executed by the processing computer selected by the calculation processing allocation unit, and the unit calculation amount per unit calculation amount is calculated based on the calculated calculation speed. The multi-computer operation system according to claim 1, further comprising a calculation speed calculation unit that calculates an amount to be optimized and registers it in the registered computer database. Multiple computer operation for providing information for selecting and assigning a plurality of processing computers for executing a plurality of computer computing processes in parallel from among a plurality of registered computers to a user requesting the plurality of computer computing processes A system,
A registered computer database that stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer;
A registration data update unit that detects a change in at least one of the amount to be optimized per unit calculation amount, the calculation speed, and the usage state, and updates the registered computer database;
A registered data providing unit that provides data registered in the registered computer database to the user terminal of the user;
A multi-computer operating system characterized by comprising:   5. The multiple computer operation system according to claim 4, wherein the registration data providing unit transmits registration data corresponding to a request time point to the user terminal in response to a data provision request from the user terminal.   6. The multiple computer operation system according to claim 4, wherein the registration data providing unit transmits registration data to the user terminal every time the registration data update unit updates the registration computer database. Multiple computer operation for providing information for selecting and assigning a plurality of processing computers for executing a plurality of computer computing processes in parallel from among a plurality of registered computers to a user requesting the plurality of computer computing processes A system,
A registered computer database that stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer;
A registration data update unit that detects a change in at least one of the amount to be optimized per unit calculation amount, the calculation speed, and the usage state, and updates the registered computer database;
An arithmetic processing information input unit that receives input of characteristic information and a target end time from the user terminal of the user that can specify the amount of calculation of the plurality of computer arithmetic processing;
The calculation amount of each computer calculation process is predicted from the characteristic information, and the sum of the amounts to be optimized of each computer calculation process is the latest expected end time of each computer calculation process as the target end time. A first arithmetic processing assigning unit that selects the same number of processing computers as the plurality of computer arithmetic processes so as to be a minimum within a range not exceeding, and assigns the selected processing computers to the plurality of computer arithmetic processes;
A processing result information providing unit for providing a processing result of the first arithmetic processing assigning unit to the user terminal;
A multi-computer operating system characterized by comprising: Multiple computer operation for providing information for selecting and assigning a plurality of processing computers for executing a plurality of computer computing processes in parallel from among a plurality of registered computers to a user requesting the plurality of computer computing processes A system,
A registered computer database that stores an amount to be optimized per predetermined unit calculation amount, a calculation speed, and a current use state related to the power consumption of each registered computer;
A registration data update unit that detects a change in at least one of the amount to be optimized per unit calculation amount, the calculation speed, and the usage state, and updates the registered computer database;
An arithmetic processing information input unit that receives input of characteristic information capable of specifying the calculation amount of the plurality of computer arithmetic processing and the upper limit value of the optimization target amount from the user terminal of the user;
The calculation amount of each computer calculation process is predicted from the characteristic information, and the expected end time of the plurality of computer calculation processes is within a range in which the total amount of the optimization target of each computer calculation process does not exceed the upper limit value. A second arithmetic processing assigning unit that selects the same number of processing computers as the plurality of computer arithmetic processes, and assigns the selected processing computers to the plurality of computer arithmetic processes so as to be the earliest;
A processing result information providing unit for providing a processing result of the second arithmetic processing assigning unit to the user terminal;
A multi-computer operating system characterized by comprising:   9. The registration data update unit according to any one of claims 4 to 8, wherein the calculation speed is received by receiving an input of a calculation time and a calculation amount of a computer calculation process executed by the processing computer. Multiple computer operating system as described. The multi-computer operation system according to any one of claims 1 to 9, wherein the optimized amount is any one of a power charge, a power consumption amount, and a CO ₂ generation amount. The amount to be optimized is a power charge,
As a power charge per unit calculation amount for each registered computer, an input of a bid power charge bid by each owner of the registered computer is accepted, and the bid power charge is sequentially registered in the registered computer database for each input. The multi-computer operation system according to any one of claims 1 to 9, further comprising a bid power charge input unit.

Images