WO2012026582A1 - Simulation device, distributed computer system, simulation method and program - Google Patents

Simulation device, distributed computer system, simulation method and program Download PDF

Info

Publication number
WO2012026582A1
WO2012026582A1 PCT/JP2011/069294 JP2011069294W WO2012026582A1 WO 2012026582 A1 WO2012026582 A1 WO 2012026582A1 JP 2011069294 W JP2011069294 W JP 2011069294W WO 2012026582 A1 WO2012026582 A1 WO 2012026582A1
Authority
WO
WIPO (PCT)
Prior art keywords
processing
power consumption
time
processing nodes
power
Prior art date
Application number
PCT/JP2011/069294
Other languages
French (fr)
Japanese (ja)
Inventor
小林　大
真樹 菅
Original Assignee
日本電気株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電気株式会社 filed Critical 日本電気株式会社
Publication of WO2012026582A1 publication Critical patent/WO2012026582A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5094—Allocation of resources, e.g. of the central processing unit [CPU] where the allocation takes into account power or heat criteria
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present invention is based on the priority claim of Japanese Patent Application: Japanese Patent Application No. 2010-190641 (filed on Aug. 27, 2010), the entire contents of which are incorporated herein by reference. Shall.
  • the present invention relates to a simulation apparatus, a simulation method, and a program, and more particularly, to a simulation apparatus, a simulation method, and a program for simulating middleware that controls power consumption of processing nodes included in a distributed computer system.
  • Non-Patent Document 1 describes a technique for constructing, managing, and providing a data structure called a BigTable for a plurality of computers as a whole with a data unit composed of a plurality of column data called rows.
  • Non-Patent Document 1 In the technology described in Non-Patent Document 1, it is realized by software which data is allocated to which computer and which computer is used for processing, and the operation is dynamically changed with respect to the system state. The resource usage is adjusted to improve the performance of the distributed computer system for client computers (or users).
  • system control middleware Such software is referred to as “system control middleware”.
  • data migration moving data from one computer to another
  • data replication data replication is placed on multiple computers, increasing or decreasing the number of replicas
  • the resource usage adjustment function of the system control middleware focuses on improving the performance of the distributed computer system.
  • performance refers to the time required for the distributed computer system to process a request from a user, or the amount or number of requests that can be processed simultaneously in the distributed computer system. Therefore, the system control middleware observes the requested processing amount of each computer and the performance with respect to the request, and adjusts the resource usage to improve the system performance based on these.
  • feedback control dynamic resource usage adjustment from the current value in this way is referred to as “feedback control”.
  • Resource usage adjustment not only improves performance, but also turns on / off the entire distributed computer system, changes the operation mode of components such as CPUs and HDDs that make up the distributed computer system, and turns off power for each component It can be performed.
  • the system control middleware is required not only to improve the performance but also to reduce the power consumption of the distributed computer system while satisfying the given service requirements.
  • Patent Literature 1 measures the power consumption of computer elements constituting the system in advance, and based on the measured power consumption and the characteristics of the processing to be executed, Adjust power consumption.
  • the power consumption of the computer is non-linear with respect to the processing load, and the characteristics differ for each type of computer element. Therefore, instead of measuring the power consumption in advance, it is possible to adjust the resource usage amount by observing the power consumption during system operation and reducing the power consumption of the system based on the observed power consumption. According to the device selection system described in Japanese Patent Application No. 2009-239280, the power consumption of the computer is obtained based on the power consumption of the air conditioner that cools the computer, and the computer used for the next calculation is selected.
  • the simulation software uses a simplified model that simulates the operation of the computer, and estimates the future behavior and performance of the computer based on the behavior of the model with respect to a workload that simplifies the expected load.
  • a simulation technique for calculating the power consumption of the system is known.
  • the power consumption of software is obtained based on the operation of software and a hardware power consumption table acquired in advance.
  • the power consumption of the system is estimated from the operation status of the system components on the SoC (System On Chip).
  • the power consumption is estimated by counting the number of toggles of elements in the circuit.
  • the simulation techniques described in Patent Documents 3 to 5 are simulation techniques for small-scale hardware in a single chip.
  • Such a technique is used for the simulation of the power and performance of a distributed computer system composed of a large number of large and complex computers, there is also a problem that the simulation time increases.
  • the objective of this invention is providing the simulation apparatus, simulation method, and program which solve this subject.
  • the simulation apparatus is: A performance calculator that calculates the processing time required for each of the plurality of processing nodes included in the distributed computer system to process the assigned load; A power calculator that calculates power consumption required to process each of the loads assigned to each of the plurality of processing nodes; A system operation determination unit that updates allocation of loads to each of the plurality of processing nodes based on processing time and power consumption.
  • the simulation method is as follows.
  • the program according to the third aspect of the present invention is: A time calculation process for calculating a processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load; A power calculation process for calculating the power consumption required for each of the plurality of processing nodes to process the assigned load; Based on the processing time and power consumption, the computer is caused to execute an update process for updating the load assignment for each of the plurality of processing nodes.
  • the simulation apparatus According to the simulation apparatus, the simulation method, and the program according to the present invention, it is possible to simulate the operation of middleware that controls the assignment of loads to the processing nodes included in the distributed computer system.
  • a simulation apparatus According to the first development form, a simulation apparatus according to the first viewpoint is provided.
  • the power calculation unit changes each of the plurality of processing nodes to one of a plurality of operation modes defined by the processing speed and the power consumption per unit time.
  • a simulation apparatus is provided.
  • a simulation device in which the power calculation unit calculates power consumption according to power consumption per unit time in each current operation mode of a plurality of processing nodes.
  • a simulation device in which the performance calculation unit calculates the processing time according to the processing speed in each of the current operation modes of the plurality of processing nodes.
  • a simulation apparatus in which the power calculation unit calculates power consumption according to the processing time.
  • the performance calculation unit calculates the processing time based on a queuing network that models each of the plurality of processing nodes, and the power calculation unit is stored in the queue of the queuing network.
  • a simulation device is provided that calculates power consumption according to the number of loads.
  • a simulation apparatus in which the system operation determination unit updates the allocation of data and processing to each of a plurality of processing nodes based on processing time and power consumption.
  • a distributed computer system includes the simulation apparatus and a plurality of processing nodes whose operation mode and / or load allocation is updated by the simulation apparatus. .
  • the simulation method according to the second viewpoint is provided.
  • the simulation apparatus, simulation method, and program according to the present invention can simulate the operation of middleware that controls the load on processing nodes included in a distributed computer system. This is because by installing the middleware as a system operation determining unit of the simulation apparatus, the processing speed and power consumption in each processing node of the distributed computer system that is feedback-controlled by the middleware can be obtained.
  • FIG. 1 is a block diagram showing a configuration of a simulation apparatus according to the present embodiment.
  • the simulation apparatus includes a performance calculation unit 53, a power calculation unit 55, and a system operation determination unit 63.
  • the performance calculation unit 53 calculates a processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load.
  • the power calculator 55 calculates the power consumption required for processing the load assigned to each of the plurality of processing nodes.
  • the system operation determination unit 63 updates the load assignment to each of the plurality of processing nodes based on the above processing time and power consumption.
  • the power calculation unit 55 may change each of the plurality of processing nodes to any one of a plurality of operation modes defined by the processing speed and the power consumption per unit time. At this time, the power calculation unit 55 calculates the power consumption according to the power consumption per unit time in the current operation mode of each of the plurality of processing nodes. In addition, the performance calculation unit 53 calculates the processing time according to the processing speed in each current operation mode of the plurality of processing nodes.
  • the power calculator 55 may calculate power consumption according to the processing time calculated by the performance calculator 53.
  • the performance calculator 53 may calculate the processing time based on a queuing network that models each of a plurality of processing nodes. At this time, the power calculator 55 may calculate the power consumption according to the number of loads stored in the queue of the queue network.
  • the system operation determination unit 63 may update the allocation of data and processing to each of the plurality of processing nodes based on the above processing time and power consumption.
  • each of the plurality of processing nodes is preferably a computer included in the distributed computer system, or a CPU, storage, or memory included in the computer.
  • the simulation apparatus may further include an operation result log 120 that holds the processing time and power consumption.
  • FIG. 2 is a block diagram showing a configuration of a distributed computer system controlled by the simulation apparatus according to the present embodiment.
  • the distributed computer system includes a plurality of processing nodes 50A, 50B,. In these processing nodes 50A, 50B,..., 50N, the operation mode and / or load assignment is updated by the simulation apparatus of the present embodiment.
  • the distributed computer system may include a simulation device.
  • the simulation apparatus can simulate the operation of middleware that controls assignment of loads to processing nodes included in a distributed computer system. This is because by installing the middleware as the system operation determining unit 63 of the simulation apparatus of the present embodiment, it is possible to obtain the processing speed and power consumption in each processing node in the distributed computer system that is feedback controlled by the middleware.
  • the correctness of the operation of the system control middleware that monitors the current state of the overall system performance and power consumption and appropriately controls the value thereof, and An evaluation of the effect (ie, the correctness of the design of the system control middleware) can be simulated.
  • middleware that evaluates performance and power consumption in simulation system time, operates according to parameters that interact with each other, and sequentially operates the values in simulation system time. It is because it inputs with respect to.
  • the performance of the computer is represented by a queue
  • the power consumption of the computer is represented by the state transition of the operation mode
  • the trade-off between the power and performance of the computer can be represented easily.
  • FIG. 3 is a block diagram showing the configuration of the simulation apparatus according to this embodiment.
  • the simulation apparatus includes a simulator 100, a simulator control unit 110, and an operation result log 120. Each of these units may be a module in simulation software.
  • the simulator 100 includes a computer use load generation unit 30, a system configuration control unit 40, a node operation simulation unit 10, and a system efficiency improvement control unit 20.
  • the node operation simulation unit 10 includes a load reception unit 11, a control load generation unit 12, a performance calculation unit 13, and a power calculation unit 15.
  • the system efficiency improvement control unit 20 includes a performance information receiving unit 21, a power information receiving unit 22, a system operation determining unit 23, and a control command issuing unit 25.
  • the configuration and behavior (model) of a simulation target are mounted.
  • the simulator control unit 110 determines the behavior of the model in the simulator 100 at a certain point in time, calculates the mutual effect of the models due to the behavior, and when the calculation at a certain point in time for all the models is completed, Advance to the next time.
  • the simulator 100 models a distributed system composed of functions of a plurality of computers, communication for connecting computers, a client for applying a load on the computers, and system control middleware for operating the computers as a bundled system. .
  • the queuing network includes a first-in first-out (FIFO) queuing unit that queues processing loads, a service unit that holds the load for a certain period of time, and then transmits the load to the next connected element; And a line connecting them.
  • FIFO first-in first-out
  • FIG. 4 is a diagram showing a queuing network as an example.
  • the node operation simulation unit 10 may represent each computer node by one queue.
  • the node operation simulation unit 10 may include an element simulation unit that represents, in a queue, components that can process loads in parallel, such as CPUs and HDDs in each computer.
  • time consumption due to communication between computer nodes and resource utilization of communication paths can be expressed using a queue. In this way, the performance calculation unit 13 calculates the usage time and the usage resource amount for the load.
  • power consumption by each component in the computer is expressed in the simulation.
  • the power consumption of each computer can be classified into power consumption of electronic components such as CPU, HDD, memory, and other power consumption (power conversion loss, cooling, etc.).
  • the power consumption of an electronic component varies according to the load applied to the electronic component, and also varies according to the operation mode of the electronic component currently in operation. Such a phenomenon can be expressed by expressing an operation mode of each electronic component in a simulation program by a state transition diagram.
  • FIG. 5 is a diagram showing an example of the state transition of the operation mode.
  • the component has four types of modes mode0 to mode3.
  • power consumption coefficients (power) P0 to P3 corresponding to the load are defined.
  • the power consumption coefficient power may be either a constant or a function. It is also possible to define performance coefficients (speed) S0 to S3 in the currently operating mode. In addition, it is possible to define times (transition time) t01, t10, t02, t20, t03, and t30 required for transition between operation modes.
  • the power consumption coefficient power of each electronic component may be expressed as a function with a load as an input.
  • FIG. 6 is a diagram illustrating a case where power consumption is expressed as a function of load.
  • the power consumption per unit time with respect to the index w indicating the load level is shown.
  • the operation mode of the computer is defined as the power consumption of the entire computer instead of the power consumption of individual electronic components, and the power for the load is calculated. Also good.
  • the power calculator 15 calculates the power consumption of the node or electronic component.
  • the calculated performance information and power consumption are recorded as an operation result log 120 in the external storage device sequentially or at every real time timing or every simulation time timing.
  • the user of the simulation software can check the operation result log 120.
  • System control middleware generally operates as a distributed computer system by bundling multiple computers.
  • the distributed data store divides the data to be stored into a plurality of data fragments, and determines the number of replicas and the storage destination computer for each data fragment.
  • An access request such as READ or WRITE from the client to the stored data fragment reaches the system control middleware.
  • the system control middleware uses a distributed index structure, a consistent hash, and the like to solve the correspondence between the data fragment and the computer.
  • the system control middleware controls the access request so as to satisfy the required transactionality by using necessary lock acquisition / release and data version management.
  • the system control middleware issues instructions for changing the correspondence between data and computers by data transfer (data migration) and increasing / decreasing the number of data copies. To issue. As a result, it is possible to perform control such that system resources are efficiently used even after the access tendency changes.
  • System power consumption is reduced by changing the operation mode of each element such as CPU and HDD in the computer.
  • the operation mode of each element such as CPU and HDD in the computer.
  • the HDD in computer A is not accessed. Therefore, the system control middleware can change the operation mode of the HDD of the computer A to the low power mode.
  • system control middleware may be able to turn off power to computers other than computer A.
  • system control middleware allocates much of the frequently used data to the computer A and performs the resource usage control as described above, thereby reducing the power consumption of the entire system. Can do.
  • the system control middleware can perform power control such that data is arranged in the computer B, the load on the computer A is reduced, or the load is removed and the power is turned off.
  • the system control middleware can dynamically reduce the power consumption of the system by dynamically inputting the current power consumption information and feeding back the value to the resource usage control.
  • specific control contents and control algorithms are not particularly limited.
  • a large-scale calculation is divided in advance by a client into processing for a small single computer called a job.
  • the job reaches system control middleware (also called a scheduler).
  • the system control middleware selects one or a plurality of computers based on the scheduling algorithm in consideration of the already assigned job and the performance of the computer, and transfers the job to the selected computer.
  • the job is executed on the transfer destination computer.
  • the system control middleware determines which computer resource of the plurality of computers is used for processing in response to a processing request or resource consumption for the computer, and changes the state of the computer.
  • the system configuration control unit 40 determines which of the plurality of computers is to be used for processing.
  • the system efficiency control unit 20 changes the state of the computer.
  • the simulation apparatus 100 having the above configuration operates as follows upon receiving time update information from the simulator control unit 110.
  • the computer usage load calculation unit 30 that simulates a client to which a load is applied generates a load on the computer that generates the load according to a predetermined program.
  • the system configuration control unit 40 that determines the destination of the load receives the generated load.
  • the system configuration control unit 40 outputs a load to the node operation simulation unit 10 that simulates the operation of one or more computer nodes.
  • the load receiving unit 11 of the node operation simulation unit 10 receives a load.
  • the performance calculator 13 calculates the performance according to the load. Furthermore, the power calculator 15 calculates the power consumption during the simulation time regardless of the presence or absence of a load.
  • the system efficiency improvement control unit 20 of the system control middleware receives the performance information and the power consumption information sequentially or for each predetermined simulation time timing or any simulation time timing.
  • the system efficiency improvement control unit 20 determines whether or not to change the system configuration such as data arrangement based on the information.
  • the system efficiency improvement control unit 20 outputs a control command for changing the system configuration information to the node operation simulation unit 10 and the system configuration control unit 40.
  • the node operation simulation unit 10 and the system configuration control unit 40 consume the simulation time or a moment during the simulation time and change the system state.
  • the control load generator 12 may output the load to the load receiver 11.
  • an operation result log 120 in which the behavior of the system control middleware with respect to the system usage load is recorded is obtained.
  • the system control middleware developer or system user can evaluate the correctness and effect of the behavior of the system control middleware by analyzing the operation result log 120.
  • the performance and power consumption of each computer during the simulation operation change within the simulation operation by a system configuration change by the system control middleware that performs feedback control while using these values as inputs. Therefore, according to the simulation apparatus of the present embodiment, it is possible to evaluate the legitimacy and effect of the system control middleware that performs feedback control on the distributed computer system.
  • FIG. 7 is a block diagram showing the configuration of the simulation apparatus of the present embodiment.
  • the power coefficient and the performance coefficient are exchanged between the performance calculation unit 33 and the power calculation unit 35 of the node operation simulation unit 10.
  • the performance calculator 33 outputs a power coefficient that varies depending on the load to the power calculator 35.
  • the power calculation unit 35 outputs the performance coefficient to the performance calculation unit 33.
  • FIG. 8 is a block diagram showing the configuration of the simulation apparatus of this embodiment in more detail.
  • the performance calculation unit 33 is a queuing network including a queuing unit 41 and a service unit 42.
  • the power calculation unit 35 is represented by a state transition diagram of the operation mode as illustrated in FIG.
  • the power coefficient can be the number of processing waiting loads in the queue, and the performance coefficient can be a value defined in each operation mode.
  • the simulation apparatus 200 having the above configuration receives time update information from the simulator control unit 110 and operates as follows.
  • the operation until the node operation simulation unit 10 receives a load is the same as the operation of the simulation apparatus 100 according to the second embodiment, and thus the description thereof is omitted.
  • the load is input to the queue unit 41. If a processing load is already input in the queue, the load is input from the head of the queue to the service unit 42 every time the processing load of the service unit 42 is completed.
  • the load processing time is given by the time from when the load is registered in the queue until the service unit 42 is exited.
  • the performance calculation unit 33 calculates the time from when the load is input to the service unit 42 until it is output based on the performance coefficient output from the power calculation unit 35 during the simulation time.
  • the service time calculation method in the present embodiment is not limited to such a method.
  • the power calculator 35 calculates the power consumption at the simulation time based on the power coefficient output from the performance calculator 33.
  • the power calculator 35 can calculate the power f (T (n)). it can.
  • the function T (n) can be estimated by an experiment or the like in an actual machine. Note that the power consumption calculation method in the present embodiment is not limited to such a method.
  • the system control middleware developer or system user can evaluate the correctness and effect of the behavior of the system control middleware by analyzing the operation result log 120. Further, according to the simulation apparatus 200, the power consumption can be easily calculated based on the state parameters in the queue. Therefore, it is possible to evaluate the legitimacy and effect of the system control middleware with a small amount of calculation compared to the case of simulating the detailed behavior of each electronic component.
  • a distributed system according to the fourth embodiment will be described.
  • a distributed computer system including a plurality of computers is controlled by simulator built-in system control middleware.
  • the simulator built-in system control middleware is software that issues resource control instructions in the second and third embodiments. Further, when the system control middleware with built-in simulator determines the operation of the system using the input power coefficient and performance coefficient, the simulation software in the second or third embodiment built in the system operation determining unit includes these. Enter coefficients to predict future system conditions. Here, the simulation software may be executed a plurality of times by giving different random numbers. The system control middleware with built-in simulator determines a resource control instruction that will be optimal in the future based on the execution contents of the simulation software, and controls the system.
  • the simulation software, simulation method, and simulator built-in system control middleware according to the present invention are applied to computer system simulation software, simulators, evaluation devices, computer operation demonstrations, distributed system schedulers, database systems, storage systems, and distributed key-value stores. can do.
  • a performance calculation unit that calculates the processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load;
  • a power calculator that calculates power consumption required to process each of the loads assigned to each of the plurality of processing nodes;
  • a simulation apparatus comprising: a system operation determination unit that updates allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
  • the power calculation unit changes each of the plurality of processing nodes to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time.
  • Supplementary note 8 Any one of Supplementary notes 1 to 7, wherein the system operation determination unit updates allocation of data and processing to each of the plurality of processing nodes based on the processing time and the power consumption.
  • the simulation device according to any one of the above.
  • Supplementary note 9 Any one of Supplementary notes 1 to 8, wherein each of the plurality of processing nodes is a computer included in the distributed computer system, or a CPU, storage, or memory included in the computer.
  • each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time.
  • each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time.
  • the computer is further caused to execute a process of updating data and process assignments for each of the plurality of processing nodes based on the processing time and the power consumption.
  • the program as described in any one.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

A simulation device is provided with a performance calculation unit for calculating the processing time required for processing the load assigned to each of a plurality of processing nodes included in a distributed computer system, a power calculation unit for calculating the amount of power required to process the load assigned to each of the plurality of processing nodes, and a system operation determination unit for updating the assignment of the load with respect to each of the plurality of processing nodes on the basis of the processing time and the amount of power required. Hereby it is possible to simulate the operation of a middleware for controlling the assignment of a load with respect to the processing nodes included in a distributed computer system.

Description

シミュレーション装置、分散計算機システム、シミュレーション方法およびプログラムSimulation apparatus, distributed computer system, simulation method, and program
［関連出願についての記載］
　本発明は、日本国特許出願：特願２０１０－１９０６４１号（２０１０年８月２７日出願）の優先権主張に基づくものであり、同出願の全記載内容は引用をもって本書に組み込み記載されているものとする。
　本発明は、シミュレーション装置、シミュレーション方法およびプログラムに関し、特に、分散計算機システムに含まれる処理ノードの消費電力を制御するミドルウェアをシミュレートするシミュレーション装置、シミュレーション方法およびプログラムに関する。 [Description of related applications]
The present invention is based on the priority claim of Japanese Patent Application: Japanese Patent Application No. 2010-190641 (filed on Aug. 27, 2010), the entire contents of which are incorporated herein by reference. Shall.
The present invention relates to a simulation apparatus, a simulation method, and a program, and more particularly, to a simulation apparatus, a simulation method, and a program for simulating middleware that controls power consumption of processing nodes included in a distributed computer system.
　多数の計算機をネットワーク結合し、そのＨＤＤ（Ｈａｒｄ　Ｄｉｓｋ　Ｄｒｉｖｅ）やメモリを用いてデータ格納・利用する分散計算機システムが知られている。例えば、非特許文献１では、行と呼ばれる複数の列データから構成されるデータ単位を、複数の計算機全体に対してＢｉｇＴａｂｌｅと呼ぶデータ構造を構築し管理・提供する技術が記載されている。 A distributed computer system is known in which a large number of computers are connected to a network, and data is stored and used using an HDD (Hard Disk Drive) or a memory. For example, Non-Patent Document 1 describes a technique for constructing, managing, and providing a data structure called a BigTable for a plurality of computers as a whole with a data unit composed of a plurality of column data called rows.
　非特許文献１に記載された技術では、データをどの計算機に配置するか、処理をどの計算機で行うかをソフトウェアにより実現し、システムの状態に対して動作を動的に変更することで、システム内のリソース使用量を調整し、分散計算機システムのクライアント計算機（ないし利用者）に対する性能を向上している。以下、このようなソフトウェアを「システム制御ミドルウェア」という。 In the technology described in Non-Patent Document 1, it is realized by software which data is allocated to which computer and which computer is used for processing, and the operation is dynamically changed with respect to the system state. The resource usage is adjusted to improve the performance of the distributed computer system for client computers (or users). Hereinafter, such software is referred to as “system control middleware”.
　システム制御ミドルウェアによるリソース使用量の調整方法として、データマイグレーション（ある計算機から別の計算機へデータを移動する）、データレプリケーション（データの複製を複数の計算機に配置し、複製の個数を増減する）、仮想マシンやサーバ機能の割り当て／マイグレーション、データアクセスの複製への転送、アクセス順序の変更等がある。 As the method of adjusting the resource usage by the system control middleware, data migration (moving data from one computer to another), data replication (data replication is placed on multiple computers, increasing or decreasing the number of replicas), There are virtual machine and server function assignment / migration, data access transfer to replication, access order change, etc.
　システム制御ミドルウェアのリソース使用量調整機能は、分散計算機システムの性能を向上させることを主眼とする。ここで、「性能」とは、利用者からの要求を分散計算機システムが処理するのに要する時間、または、分散計算機システムで同時に処理可能な要求の量もしくは個数をいう。したがって、システム制御ミドルウェアは、各計算機の要求処理量や要求に対する性能を観測し、これらに基づいて、システムの性能を向上させるべくリソース使用量調整を行う。以下、このように現在の値から動的にリソース使用量調整を行うことを「フィードバック制御」という。 The resource usage adjustment function of the system control middleware focuses on improving the performance of the distributed computer system. Here, “performance” refers to the time required for the distributed computer system to process a request from a user, or the amount or number of requests that can be processed simultaneously in the distributed computer system. Therefore, the system control middleware observes the requested processing amount of each computer and the performance with respect to the request, and adjusts the resource usage to improve the system performance based on these. Hereinafter, dynamic resource usage adjustment from the current value in this way is referred to as “feedback control”.
　リソース使用量調整として、性能の向上のみならず、分散計算機システム全体の電源のオン／オフ、分散計算機システムを構成するＣＰＵ、ＨＤＤなどの構成要素の動作モードの変更、構成要素ごとの電源オフなどを行うことができる。 Resource usage adjustment not only improves performance, but also turns on / off the entire distributed computer system, changes the operation mode of components such as CPUs and HDDs that make up the distributed computer system, and turns off power for each component It can be performed.
　近年では、特に、分散計算機システムの消費電力を削減することが重要となってきている。したがって、システム制御ミドルウェアは、性能の向上を図るのみならず、与えられたサービス要件を満たしつつ、分散計算機システムの消費電力を削減する必要がある。 In recent years, it has become particularly important to reduce the power consumption of distributed computer systems. Therefore, the system control middleware is required not only to improve the performance but also to reduce the power consumption of the distributed computer system while satisfying the given service requirements.
　特許文献１に記載された計算機システムは、システムを構成する計算機要素の消費電力を予め測定しておいて、測定された消費電力と実行すべき処理の特徴とに基づいて、計算機のストレージ装置の消費電力を調整する。 The computer system described in Patent Literature 1 measures the power consumption of computer elements constituting the system in advance, and based on the measured power consumption and the characteristics of the processing to be executed, Adjust power consumption.
　ところで、計算機の消費電力は処理負荷に対して非線形であり、計算機要素の種類ごとに特性が異なる。したがって、事前に消費電力を測定する代わりに、システム運用中の消費電力を観測し、観測した消費電力に基づいてシステムの消費電力を削減するというリソース使用量調整を行うこともできる。特願２００９－２３９２８０号に記載された機器選択システムによると、計算機を冷却する空調の消費電力に基づいて計算機の消費電力を求め、次の計算に利用する計算機が選択される。 By the way, the power consumption of the computer is non-linear with respect to the processing load, and the characteristics differ for each type of computer element. Therefore, instead of measuring the power consumption in advance, it is possible to adjust the resource usage amount by observing the power consumption during system operation and reducing the power consumption of the system based on the observed power consumption. According to the device selection system described in Japanese Patent Application No. 2009-239280, the power consumption of the computer is obtained based on the power consumption of the air conditioner that cools the computer, and the computer used for the next calculation is selected.
　システム制御ミドルウェアを開発・設計・実装する場合には、その効果と正当性を事前に評価することが好ましい。システム制御ミドルウェアの正当性を評価する方法として、実機においてシステム制御ミドルウェアを動作させて確認する方法以外に、シミュレーションソフトウェアまたはシミュレーション装置を用いる方法がある。特許文献２に記載されたシミュレータによると、複数の計算機から成る分散処理システムに対するシステム制御ミドルウェアの性能を確認することができる。 When developing, designing, and implementing system control middleware, it is preferable to evaluate its effectiveness and validity in advance. As a method for evaluating the legitimacy of the system control middleware, there is a method of using simulation software or a simulation apparatus in addition to a method of operating and checking the system control middleware in an actual machine. According to the simulator described in Patent Literature 2, it is possible to confirm the performance of the system control middleware for the distributed processing system including a plurality of computers.
　シミュレーションソフトウェアは、計算機の動作を模擬する簡便化したモデルを用い、将来予想される負荷を簡便化したワークロードに対する、かかるモデルの挙動に基づいて、将来の計算機の振る舞いおよび性能を推定する。 The simulation software uses a simplified model that simulates the operation of the computer, and estimates the future behavior and performance of the computer based on the behavior of the model with respect to a workload that simplifies the expected load.
　また、システムの消費電力を算出するシミュレーション技術も知られている。特許文献３に記載された技術によると、ソフトウェアの動作と予め取得されたハードウェア電力消費の表とに基づいて、ソフトウェアの消費電力を求める。また、特許文献４に記載された技術では、ＳｏＣ（Ｓｙｓｔｅｍ　Ｏｎ　Ｃｈｉｐ）上のシステムの構成要素の動作状況から、そのシステムの電力消費量を見積もる。さらに、特許文献５に記載された論理回路のシミュレーションにおいては、回路内の要素のトグル回数をカウントして消費電力を見積もる。 Also, a simulation technique for calculating the power consumption of the system is known. According to the technique described in Patent Document 3, the power consumption of software is obtained based on the operation of software and a hardware power consumption table acquired in advance. In the technique described in Patent Document 4, the power consumption of the system is estimated from the operation status of the system components on the SoC (System On Chip). Furthermore, in the simulation of the logic circuit described in Patent Document 5, the power consumption is estimated by counting the number of toggles of elements in the circuit.
特開２００８－２１０００３号公報JP 2008-210003 A 特開２００６－３５０６５７号公報JP 2006-350657 A 特開２００９－０７５９６５号公報JP 2009-0795965 A 特開２００８－２０４３５０号公報JP 2008-204350 A 特開２００７－１０２３３７号公報JP 2007-102337 A
　上記の特許文献および非特許文献の各開示を、本書に引用をもって繰り込むものとする。以下の分析は、本発明者によってなされたものである。 The disclosures of the above patent documents and non-patent documents are incorporated herein by reference. The following analysis was made by the present inventors.
　特許文献２に記載されたシミュレータによると、電力情報を取得することができない。したがって、特許文献２に記載されたシミュレータによると、電力フィードバック制御を行うシステム制御ミドルウェアの正当性および効果を評価することができない。 According to the simulator described in Patent Document 2, power information cannot be acquired. Therefore, according to the simulator described in Patent Document 2, it is not possible to evaluate the validity and effect of system control middleware that performs power feedback control.
　また、特許文献３ないし５に記載されたシミュレーション技術によると、まず、構成要素の性能が評価され、シミュレーション終了時にシミュレーションの出力結果から電力が決定される。かかるシミュレーション方式によると、電力フィードバック制御を行うシステム制御ミドルウェアにシステム運用中の電力情報を入力することができない。すなわち、システム制御ミドルウェアを、設計者の意図したとおりに動作させることができない。したがって、特許文献３ないし５に記載されたシミュレーション技術によると、電力フィードバック制御を行うシステム制御ミドルウェアの正当性および効果を評価することができない。 In addition, according to the simulation techniques described in Patent Documents 3 to 5, first, the performance of the component is evaluated, and the power is determined from the simulation output result at the end of the simulation. According to such a simulation method, power information during system operation cannot be input to system control middleware that performs power feedback control. That is, the system control middleware cannot be operated as designed by the designer. Therefore, according to the simulation techniques described in Patent Documents 3 to 5, it is not possible to evaluate the validity and effect of the system control middleware that performs power feedback control.
　さらに、特許文献３ないし５に記載されたシミュレーション技術は、単一チップ内の小規模なハードウェアのシミュレーション技術である。かかる技術を大規模かつ複雑な多数の計算機によって構成される分散計算機システムの電力および性能のシミュレーションに用いた場合には、シミュレーション時間が増大するという問題もある。 Furthermore, the simulation techniques described in Patent Documents 3 to 5 are simulation techniques for small-scale hardware in a single chip. When such a technique is used for the simulation of the power and performance of a distributed computer system composed of a large number of large and complex computers, there is also a problem that the simulation time increases.
　そこで、分散計算機システムに含まれる処理ノードに対する負荷の割り当てを制御するミドルウェアの動作をシミュレートできるようにすることが課題となる。本発明の目的は、かかる課題を解決するシミュレーション装置、シミュレーション方法およびプログラムを提供することにある。 Therefore, it becomes an issue to be able to simulate the operation of middleware that controls the assignment of loads to the processing nodes included in the distributed computer system. The objective of this invention is providing the simulation apparatus, simulation method, and program which solve this subject.
　本発明の第１の視点に係るシミュレーション装置は、
　分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する性能計算部と、
　複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算部と、
　処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対する負荷の割り当てを更新するシステム動作決定部と、を備えている。 The simulation apparatus according to the first aspect of the present invention is:
A performance calculator that calculates the processing time required for each of the plurality of processing nodes included in the distributed computer system to process the assigned load;
A power calculator that calculates power consumption required to process each of the loads assigned to each of the plurality of processing nodes;
A system operation determination unit that updates allocation of loads to each of the plurality of processing nodes based on processing time and power consumption.
　本発明の第２の視点に係るシミュレーション方法は、
　コンピュータが、分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算工程と、
　複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算工程と、
　処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新工程と、を含む。 The simulation method according to the second aspect of the present invention is as follows.
A time calculation step in which the computer calculates a processing time required for each of the plurality of processing nodes included in the distributed computer system to process the assigned load;
A power calculation step of calculating power consumption required for processing each of the load assigned to each of the plurality of processing nodes;
And an updating step of updating the load assignment for each of the plurality of processing nodes based on the processing time and the power consumption.
　本発明の第３の視点に係るプログラムは、
　分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算処理と、
　複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算処理と、
　処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新処理と、をコンピュータに実行させる。 The program according to the third aspect of the present invention is:
A time calculation process for calculating a processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load;
A power calculation process for calculating the power consumption required for each of the plurality of processing nodes to process the assigned load;
Based on the processing time and power consumption, the computer is caused to execute an update process for updating the load assignment for each of the plurality of processing nodes.
　本発明に係るシミュレーション装置、シミュレーション方法およびプログラムによると、分散計算機システムに含まれる処理ノードに対する負荷の割り当てを制御するミドルウェアの動作をシミュレートすることができる。 According to the simulation apparatus, the simulation method, and the program according to the present invention, it is possible to simulate the operation of middleware that controls the assignment of loads to the processing nodes included in the distributed computer system.
第１の実施形態に係るシミュレーション装置の構成を示すブロック図である。It is a block diagram which shows the structure of the simulation apparatus which concerns on 1st Embodiment. 第１の実施形態に係るシミュレーション装置によって制御される分散計算機システムの構成を示すブロック図である。It is a block diagram which shows the structure of the distributed computer system controlled by the simulation apparatus which concerns on 1st Embodiment. 第２の実施形態に係るシミュレーション装置の構成を示すブロック図である。It is a block diagram which shows the structure of the simulation apparatus which concerns on 2nd Embodiment. 待ち行列ネットワークを一例として示す図である。It is a figure which shows a queue network as an example. 動作モードの状態遷移を一例として示す図である。It is a figure which shows the state transition of an operation mode as an example. 消費電力を負荷の関数として表す場合を一例として示す図である。It is a figure which shows the case where power consumption is represented as a function of load as an example. 第３の実施形態に係るシミュレーション装置の構成を示すブロック図である。It is a block diagram which shows the structure of the simulation apparatus which concerns on 3rd Embodiment. 第３の実施形態に係るシミュレーション装置の構成を詳細に示すブロック図である。It is a block diagram which shows in detail the structure of the simulation apparatus which concerns on 3rd Embodiment.
　第１の展開形態によると、上記第１の視点に係るシミュレーション装置が提供される。 According to the first development form, a simulation apparatus according to the first viewpoint is provided.
　第２の展開形態によると、電力計算部は、複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更する、シミュレーション装置が提供される。 According to the second development form, the power calculation unit changes each of the plurality of processing nodes to one of a plurality of operation modes defined by the processing speed and the power consumption per unit time. A simulation apparatus is provided.
　第３の展開形態によると、電力計算部は、複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて消費電力を計算する、シミュレーション装置が提供される。 According to the third development mode, a simulation device is provided in which the power calculation unit calculates power consumption according to power consumption per unit time in each current operation mode of a plurality of processing nodes.
　第４の展開形態によると、性能計算部は、複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて処理時間を計算する、シミュレーション装置が提供される。 According to the fourth development mode, a simulation device is provided in which the performance calculation unit calculates the processing time according to the processing speed in each of the current operation modes of the plurality of processing nodes.
　第５の展開形態によると、電力計算部は処理時間に応じて消費電力を計算する、シミュレーション装置が提供される。 According to the fifth development form, a simulation apparatus is provided in which the power calculation unit calculates power consumption according to the processing time.
　第６の展開形態によると、性能計算部は、複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて処理時間を計算し、電力計算部は、待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて消費電力を計算する、シミュレーション装置が提供される。 According to the sixth development mode, the performance calculation unit calculates the processing time based on a queuing network that models each of the plurality of processing nodes, and the power calculation unit is stored in the queue of the queuing network. A simulation device is provided that calculates power consumption according to the number of loads.
　第７の展開形態によると、システム動作決定部は、処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新する、シミュレーション装置が提供される。 According to the seventh development mode, a simulation apparatus is provided in which the system operation determination unit updates the allocation of data and processing to each of a plurality of processing nodes based on processing time and power consumption.
　第８の展開形態によると、上記シミュレーション装置と、上記シミュレーション装置によって、動作モード、および／または、負荷の割り当てが更新される複数の処理ノードと、を備えている、分散計算機システムが提供される。 According to an eighth development mode, a distributed computer system is provided that includes the simulation apparatus and a plurality of processing nodes whose operation mode and / or load allocation is updated by the simulation apparatus. .
　第９の展開形態によると、上記第２の視点に係るシミュレーション方法が提供される。 According to the ninth development form, the simulation method according to the second viewpoint is provided.
　第１０の展開形態によると、上記第３の視点に係るプログラムが提供される。 According to the tenth development form, a program according to the third viewpoint is provided.
　本発明に係るシミュレーション装置、シミュレーション方法およびプログラムによると、分散計算機システムに含まれる処理ノードに対する負荷を制御するミドルウェアの動作をシミュレートすることができる。ミドルウェアをシミュレーション装置のシステム動作決定部として実装することにより、当該ミドルウェアによってフィードバック制御された分散計算機システムの各処理ノードにおける処理速度および消費電力を求めることができるからである。 The simulation apparatus, simulation method, and program according to the present invention can simulate the operation of middleware that controls the load on processing nodes included in a distributed computer system. This is because by installing the middleware as a system operation determining unit of the simulation apparatus, the processing speed and power consumption in each processing node of the distributed computer system that is feedback-controlled by the middleware can be obtained.
　（第１の実施形態）
　第１の実施形態に係るシミュレーション装置について、図面を参照して説明する。 (First embodiment)
A simulation apparatus according to a first embodiment will be described with reference to the drawings.
　図１は、本実施形態に係るシミュレーション装置の構成を示すブロック図である。図１を参照すると、シミュレーション装置は、性能計算部５３、電力計算部５５およびシステム動作決定部６３を備えている。性能計算部５３は、分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する。電力計算部５５は、複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する。システム動作決定部６３は、上記の処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対する負荷の割り当てを更新する。 FIG. 1 is a block diagram showing a configuration of a simulation apparatus according to the present embodiment. Referring to FIG. 1, the simulation apparatus includes a performance calculation unit 53, a power calculation unit 55, and a system operation determination unit 63. The performance calculation unit 53 calculates a processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load. The power calculator 55 calculates the power consumption required for processing the load assigned to each of the plurality of processing nodes. The system operation determination unit 63 updates the load assignment to each of the plurality of processing nodes based on the above processing time and power consumption.
　電力計算部５５は、複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更するようにしてもよい。このとき、電力計算部５５は、複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて消費電力を計算する。また、性能計算部５３は、複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて処理時間を計算する。 The power calculation unit 55 may change each of the plurality of processing nodes to any one of a plurality of operation modes defined by the processing speed and the power consumption per unit time. At this time, the power calculation unit 55 calculates the power consumption according to the power consumption per unit time in the current operation mode of each of the plurality of processing nodes. In addition, the performance calculation unit 53 calculates the processing time according to the processing speed in each current operation mode of the plurality of processing nodes.
　電力計算部５５は、性能計算部５３によって計算された処理時間に応じて、消費電力を計算するようにしてもよい。 The power calculator 55 may calculate power consumption according to the processing time calculated by the performance calculator 53.
　性能計算部５３は、複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、処理時間を計算するようにしてもよい。このとき、電力計算部５５は、待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、消費電力を計算するようにしてもよい。 The performance calculator 53 may calculate the processing time based on a queuing network that models each of a plurality of processing nodes. At this time, the power calculator 55 may calculate the power consumption according to the number of loads stored in the queue of the queue network.
　システム動作決定部６３は、上記の処理時間および消費電力に基づいて、複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新するようにしてもよい。 The system operation determination unit 63 may update the allocation of data and processing to each of the plurality of processing nodes based on the above processing time and power consumption.
　ここで、複数の処理ノードは、それぞれ、分散計算機システムに含まれる計算機、または、当該計算機に含まれるＣＰＵ、ストレージもしくはメモリであることが好ましい。 Here, each of the plurality of processing nodes is preferably a computer included in the distributed computer system, or a CPU, storage, or memory included in the computer.
　また、シミュレーション装置は、処理時間および消費電力を保持する動作結果ログ１２０をさらに備えていてもよい。 In addition, the simulation apparatus may further include an operation result log 120 that holds the processing time and power consumption.
　図２は、本実施形態に係るシミュレーション装置によって制御される分散計算機システムの構成を示すブロック図である。図２を参照すると、分散計算機システムは、複数の処理ノード５０Ａ，５０Ｂ，…，５０Ｎを備えている。これらの処理ノード５０Ａ，５０Ｂ，…，５０Ｎは、本実施形態のシミュレーション装置によって、動作モード、および／または、負荷の割り当てが更新される。なお、分散計算機システムは、シミュレーション装置を含むようにしてもよい。 FIG. 2 is a block diagram showing a configuration of a distributed computer system controlled by the simulation apparatus according to the present embodiment. Referring to FIG. 2, the distributed computer system includes a plurality of processing nodes 50A, 50B,. In these processing nodes 50A, 50B,..., 50N, the operation mode and / or load assignment is updated by the simulation apparatus of the present embodiment. Note that the distributed computer system may include a simulation device.
　本実施形態に係るシミュレーション装置によると、分散計算機システムに含まれる処理ノードに対する負荷の割り当てを制御するミドルウェアの動作をシミュレートすることができる。ミドルウェアを本実施形態のシミュレーション装置のシステム動作決定部６３として実装することにより、当該ミドルウェアによってフィードバック制御された分散計算機システム内の各処理ノードにおける処理速度および消費電力を求めることができるからである。 The simulation apparatus according to the present embodiment can simulate the operation of middleware that controls assignment of loads to processing nodes included in a distributed computer system. This is because by installing the middleware as the system operation determining unit 63 of the simulation apparatus of the present embodiment, it is possible to obtain the processing speed and power consumption in each processing node in the distributed computer system that is feedback controlled by the middleware.
　本実施形態に係るシミュレーション装置によると、複数の計算機を備えた分散計算機システムにおいて、システム全体の性能と消費電力の現在状態を監視しその値を適正に制御するシステム制御ミドルウェアの動作の正当性および効果（すなわち、システム制御ミドルウェアの設計の正当性）の評価を、シミュレートすることができる。 According to the simulation apparatus according to the present embodiment, in a distributed computer system having a plurality of computers, the correctness of the operation of the system control middleware that monitors the current state of the overall system performance and power consumption and appropriately controls the value thereof, and An evaluation of the effect (ie, the correctness of the design of the system control middleware) can be simulated.
　本実施形態のシミュレーション装置によると、シミュレーション系内時間で性能および消費電力を評価し、相互に影響しあうパラメータによって動作し、その値をシミュレーション系内時間で逐次に、シミュレーション系内で動作するミドルウェアに対して入力するからである。 According to the simulation apparatus of the present embodiment, middleware that evaluates performance and power consumption in simulation system time, operates according to parameters that interact with each other, and sequentially operates the values in simulation system time. It is because it inputs with respect to.
　また、本実施形態のシミュレーション装置によると、計算機の性能を待ち行列で表現するとともに、計算機の消費電力を動作モードの状態遷移で表現し、簡便に計算機の電力と性能のトレードオフを表現することで、各機能要素を詳細にシミュレーションする従来の手法と比較して、少ない計算量でシステム制御ミドルウェアの動作の正当性および効果を検証することができる。 In addition, according to the simulation apparatus of the present embodiment, the performance of the computer is represented by a queue, and the power consumption of the computer is represented by the state transition of the operation mode, and the trade-off between the power and performance of the computer can be represented easily. Thus, the correctness and effect of the operation of the system control middleware can be verified with a small amount of calculation as compared with the conventional method of simulating each functional element in detail.
　（第２の実施形態）
　第２の実施形態に係るシミュレーション装置について、図面を参照して説明する。 (Second Embodiment)
A simulation apparatus according to the second embodiment will be described with reference to the drawings.
　図３は、本実施形態に係るシミュレーション装置の構成を示すブロック図である。図３を参照すると、シミュレーション装置は、シミュレータ１００、シミュレータ制御部１１０、および、動作結果ログ１２０を備えている。なお、これらの各部は、シミュレーションソフトウェア内のモジュールとしてもよい。 FIG. 3 is a block diagram showing the configuration of the simulation apparatus according to this embodiment. Referring to FIG. 3, the simulation apparatus includes a simulator 100, a simulator control unit 110, and an operation result log 120. Each of these units may be a module in simulation software.
　シミュレータ１００は、計算機利用負荷生成部３０、システム構成制御部４０、ノード動作模擬部１０、および、システム効率化制御部２０を備えている。ノード動作模擬部１０は、負荷受信部１１、制御負荷発生部１２、性能計算部１３、および、電力計算部１５を備えている。一方、システム効率化制御部２０は、性能情報受信部２１、電力情報受信部２２、システム動作決定部２３、および、制御命令発行部２５を備えている。 The simulator 100 includes a computer use load generation unit 30, a system configuration control unit 40, a node operation simulation unit 10, and a system efficiency improvement control unit 20. The node operation simulation unit 10 includes a load reception unit 11, a control load generation unit 12, a performance calculation unit 13, and a power calculation unit 15. On the other hand, the system efficiency improvement control unit 20 includes a performance information receiving unit 21, a power information receiving unit 22, a system operation determining unit 23, and a control command issuing unit 25.
　シミュレータ１００には、シミュレーションする対象となるものの構成および振る舞い（モデル）が実装される。 In the simulator 100, the configuration and behavior (model) of a simulation target are mounted.
　シミュレータ制御部１１０は、シミュレータ１００内のモデルのある時点の振る舞いを決定し、その振る舞いによるモデル相互の影響を計算し、すべてのモデルのある時点での計算が終了すると、シミュレータ１００内の時刻を次の時刻に進める。 The simulator control unit 110 determines the behavior of the model in the simulator 100 at a certain point in time, calculates the mutual effect of the models due to the behavior, and when the calculation at a certain point in time for all the models is completed, Advance to the next time.
　本実施形態では、複数の計算機を含む分散計算機システムのシミュレーションを行なうものとする。このとき、シミュレータ１００においては、複数の計算機、計算機間を接続する通信、計算機に負荷をかけるクライアント、複数の計算機を束ねシステムとして動作させるシステム制御ミドルウェアの各機能からなる分散システムがモデル化される。 In this embodiment, it is assumed that a distributed computer system including a plurality of computers is simulated. At this time, the simulator 100 models a distributed system composed of functions of a plurality of computers, communication for connecting computers, a client for applying a load on the computers, and system control middleware for operating the computers as a bundled system. .
　分散システムのモデルのうち、計算機に対する負荷（リソース利用要求）に対する性能は、例えば、待ち行列ネットワークにより簡易に表現し、計算することができる。待ち行列ネットワークは、処理待ち負荷を並べるファーストイン・ファーストアウト（ＦＩＦＯ）の待ち行列部と、負荷がある定められた時間保持され、その後、次につながれた要素に負荷を送信するサービス部と、これらをつなぐ線とを含む。 Among the distributed system models, the performance with respect to the load on the computer (resource usage request) can be expressed and calculated simply by a queuing network, for example. The queuing network includes a first-in first-out (FIFO) queuing unit that queues processing loads, a service unit that holds the load for a certain period of time, and then transmits the load to the next connected element; And a line connecting them.
　図４は、待ち行列ネットワークを一例として示す図である。例えば、ノード動作模擬部１０は、各計算機ノードをひとつの待ち行列により表現するようにしてもよい。また、ノード動作模擬部１０は、各計算機内のＣＰＵ、ＨＤＤのように負荷を並列に処理しうるコンポーネントをそれぞれ待ち行列で表現する要素模擬部を有していてもよい。さらに、計算機ノードの間の通信による時間消費や、通信路のリソース利用を待ち行列を利用して表現することもできる。性能計算部１３は、このようにして負荷に対する利用時間や利用リソース量を計算する。 FIG. 4 is a diagram showing a queuing network as an example. For example, the node operation simulation unit 10 may represent each computer node by one queue. In addition, the node operation simulation unit 10 may include an element simulation unit that represents, in a queue, components that can process loads in parallel, such as CPUs and HDDs in each computer. Furthermore, time consumption due to communication between computer nodes and resource utilization of communication paths can be expressed using a queue. In this way, the performance calculation unit 13 calculates the usage time and the usage resource amount for the load.
　本実施形態のシミュレーション装置においては、計算機中の各コンポーネントによる消費電力がシミュレーション中で表現される。各計算機の消費電力は、ＣＰＵ、ＨＤＤ、メモリなどの電子部品の消費電力と、それ以外の消費電力（電源変換ロス、冷却など）とに分類することができる。電子部品の消費電力は、その電子部品にかかる負荷に応じて変動するとともに、現在稼動している電子部品の動作モードに応じて変動する。このような現象は、シミュレーションプログラム中で各電子部品の動作モードを状態遷移図で表すことによって表現しうる。 In the simulation apparatus of the present embodiment, power consumption by each component in the computer is expressed in the simulation. The power consumption of each computer can be classified into power consumption of electronic components such as CPU, HDD, memory, and other power consumption (power conversion loss, cooling, etc.). The power consumption of an electronic component varies according to the load applied to the electronic component, and also varies according to the operation mode of the electronic component currently in operation. Such a phenomenon can be expressed by expressing an operation mode of each electronic component in a simulation program by a state transition diagram.
　図５は、動作モードの状態遷移を一例として示す図である。図５に示した例では、コンポーネントは、４種類のモードｍｏｄｅ０～ｍｏｄｅ３を有する。各モードにおいて、負荷に応じた消費電力係数（ｐｏｗｅｒ）Ｐ０～Ｐ３が定義される。消費電力係数ｐｏｗｅｒは、定数または関数のいずれであってもよい。また、現在動作中の動作モードにおける性能係数（ｓｐｅｅｄ）Ｓ０～Ｓ３を定義することもできる。また、動作モード間を遷移するために要する時間（遷移時間ｔｉｍｅ）ｔ０１，ｔ１０，ｔ０２，ｔ２０，ｔ０３，ｔ３０を定義することができる。また、各電子部品の消費電力係数ｐｏｗｅｒは、負荷を入力とする関数として表現してもよい。 FIG. 5 is a diagram showing an example of the state transition of the operation mode. In the example shown in FIG. 5, the component has four types of modes mode0 to mode3. In each mode, power consumption coefficients (power) P0 to P3 corresponding to the load are defined. The power consumption coefficient power may be either a constant or a function. It is also possible to define performance coefficients (speed) S0 to S3 in the currently operating mode. In addition, it is possible to define times (transition time) t01, t10, t02, t20, t03, and t30 required for transition between operation modes. The power consumption coefficient power of each electronic component may be expressed as a function with a load as an input.
　図６は、消費電力を負荷の関数として表す場合を一例として示す図である。図６を参照すると、負荷の大小を表す指標ｗに対する単位時間当りの消費電力が示されている。このとき、ある時点における負荷の値ｗを関数の引数に入力することで、その時点における単位時間当りの消費電力が得られる。なお、より簡便に計算機の消費電力を表現するために、個々の電子部品の消費電力の代わりに、計算機全体の消費電力として、計算機の動作モードを定義し、負荷に対する電力を算出するようにしてもよい。電力計算部１５は、このようにして、ノードまたは電子部品の消費電力を計算する。 FIG. 6 is a diagram illustrating a case where power consumption is expressed as a function of load. Referring to FIG. 6, the power consumption per unit time with respect to the index w indicating the load level is shown. At this time, by inputting the load value w at a certain time point as an argument of the function, the power consumption per unit time at that time point can be obtained. In order to more easily express the power consumption of a computer, the operation mode of the computer is defined as the power consumption of the entire computer instead of the power consumption of individual electronic components, and the power for the load is calculated. Also good. In this way, the power calculator 15 calculates the power consumption of the node or electronic component.
　計算された性能情報および消費電力量は、逐次、または、ある実時間タイミングもしくはあるシミュレーション時間タイミングごとに、外部記憶装置に動作結果ログ１２０として記録される。シミュレーションソフトウェアのユーザは、動作結果ログ１２０を確認することができる。 The calculated performance information and power consumption are recorded as an operation result log 120 in the external storage device sequentially or at every real time timing or every simulation time timing. The user of the simulation software can check the operation result log 120.
　システム制御ミドルウェアは、一般に、複数の計算機を束ねることで、分散計算機システムとして動作させる。 System control middleware generally operates as a distributed computer system by bundling multiple computers.
　第１の例として、複数の計算機をまとめて１つの分散データストアとするシステム制御ミドルウェアの場合について説明する。 As a first example, a case of system control middleware in which a plurality of computers are combined into one distributed data store will be described.
　分散データストアは、格納するデータを複数のデータ断片に分割し、各データ断片について、複製の個数および格納先の計算機を決定する。格納データ断片に対するクライアントからのＲＥＡＤ、ＷＲＩＴＥなどのアクセス要求は、システム制御ミドルウェアに到達する。システム制御ミドルウェアは、分散索引構造、コンシステントハッシュなどを利用し、データ断片と計算機との対応関係を解決する。 The distributed data store divides the data to be stored into a plurality of data fragments, and determines the number of replicas and the storage destination computer for each data fragment. An access request such as READ or WRITE from the client to the stored data fragment reaches the system control middleware. The system control middleware uses a distributed index structure, a consistent hash, and the like to solve the correspondence between the data fragment and the computer.
　ＲＥＡＤの場合には、複製のうちのいずれかを保持する計算機へのアクセスを行う。一方、ＷＲＩＴＥの場合には、データ間で予め定められた一貫性要件が保持されるように、複数の計算機へ同期的または非同期的にアクセスが行なわれる。 In the case of READ, access is made to the computer that holds one of the replicas. On the other hand, in the case of WRITE, a plurality of computers are accessed synchronously or asynchronously so as to maintain a predetermined consistency requirement between data.
　また、システムによっては、トランザクション性に関する要件が定められている場合がある。かかる場合には、システム制御ミドルウェアは、必要なロックの取得・解放、データの版管理などを利用し、求められるトランザクション性を満たすようにアクセス要求を制御する。 Also, depending on the system, there may be requirements for transactionality. In such a case, the system control middleware controls the access request so as to satisfy the required transactionality by using necessary lock acquisition / release and data version management.
　また、長期的スパンでデータに対するアクセスの傾向が変化する場合には、システム制御ミドルウェアは、データ転送によるデータと計算機の対応の変更（データマイグレーション）や、データ複製数の増減を行う命令を各計算機に発行する。これにより、アクセス傾向が変化した後においても、システムリソースが効率よく利用されるような制御が可能となる。 In addition, when the tendency of access to data changes over a long-term span, the system control middleware issues instructions for changing the correspondence between data and computers by data transfer (data migration) and increasing / decreasing the number of data copies. To issue. As a result, it is possible to perform control such that system resources are efficiently used even after the access tendency changes.
　システムリソースの効率性についての指標として、アクセスに対する処理時間を最小化（レイテンシを削減）し、システム同時処理アクセス数（スループット）を増加させる以外に、システム消費電力量を削減するためのリソース使用量制御を行う。 Resource usage to reduce system power consumption as an indicator of system resource efficiency, in addition to minimizing access processing time (reducing latency) and increasing the number of simultaneous system accesses (throughput) Take control.
　システム消費電力量削減は、計算機中のＣＰＵ、ＨＤＤなどの各要素の動作モードを変更することで行う。例えば、多くアクセスされるデータのすべてが特定の計算機（計算機Ａとする）のメモリ中に存在する場合には、計算機Ａ中のＨＤＤはアクセスされない。したがって、システム制御ミドルウェアは、計算機ＡのＨＤＤの動作モードを低電力モードに変更することができる。 System power consumption is reduced by changing the operation mode of each element such as CPU and HDD in the computer. For example, when all the frequently accessed data exists in the memory of a specific computer (referred to as computer A), the HDD in computer A is not accessed. Therefore, the system control middleware can change the operation mode of the HDD of the computer A to the low power mode.
　また、システム制御ミドルウェアは、計算機Ａ以外の計算機の電源をオフにすることができる場合もある。このような場合には、システム制御ミドルウェアは、高頻度に利用されるデータの多くを計算機Ａに配置し、上述のようなリソース使用量制御を行うことで、システム全体の消費電力を削減することができる。 In addition, the system control middleware may be able to turn off power to computers other than computer A. In such a case, the system control middleware allocates much of the frequently used data to the computer A and performs the resource usage control as described above, thereby reducing the power consumption of the entire system. Can do.
　さらに、実行中に計算機の消費電力を動的に取得し、計算機Ａと同一の性能を有する計算機（計算機Ｂとする）が計算機Ａよりも低消費電力で動作可能であることが判明した場合には、システム制御ミドルウェアは、データを計算機Ｂへ配置し、計算機Ａの負荷を低下させ、または、負荷を無くして電源をオフとするような電力制御を行うことができる。 Furthermore, when the power consumption of the computer is dynamically acquired during execution and it is found that a computer having the same performance as the computer A (referred to as computer B) can operate with lower power consumption than the computer A. The system control middleware can perform power control such that data is arranged in the computer B, the load on the computer A is reduced, or the load is removed and the power is turned off.
　このように、システム制御ミドルウェアは、動的に現在の消費電力情報を入力とし、その値をリソース使用量制御にフィードバックすることで、システムの消費電力を削減することができる。なお、本実施形態では、具体的な制御内容、制御アルゴリズムについては特に限定しない。 As described above, the system control middleware can dynamically reduce the power consumption of the system by dynamically inputting the current power consumption information and feeding back the value to the resource usage control. In the present embodiment, specific control contents and control algorithms are not particularly limited.
　次に、第２の例として、複数の計算機を用いて大規模な計算を並列に動作させるためのスケジューリングを行うシステム制御ミドルウェアの場合について説明する。 Next, as a second example, a case of system control middleware that performs scheduling for operating a large-scale calculation in parallel using a plurality of computers will be described.
　大規模な計算は、クライアントによって、ジョブと呼ばれる小規模な単一計算機に対する処理に予め分割されている。ジョブは、システム制御ミドルウェア（スケジューラともいう）に到達する。システム制御ミドルウェアは、すでに割り当てられたジョブおよび計算機の性能を考慮して、スケジューリングアルゴリズムに基づいて、１つ又は複数の計算機を選択し、選択した計算機にジョブを転送する。ジョブは、転送先の計算機で実行される。 A large-scale calculation is divided in advance by a client into processing for a small single computer called a job. The job reaches system control middleware (also called a scheduler). The system control middleware selects one or a plurality of computers based on the scheduling algorithm in consideration of the already assigned job and the performance of the computer, and transfers the job to the selected computer. The job is executed on the transfer destination computer.
　システム制御ミドルウェアは、計算機に対する処理要求またはリソース消費に対して、複数の計算機のうちのいずれの計算機のリソースを利用して処理を行うかを決定するとともに、計算機の状態を変更する。ここでは、システム構成制御部４０は、複数の計算機のうちのいずれの計算機のリソースを利用して処理を行うかを決定する。一方、システム効率化制御部２０は、計算機の状態を変更する。 The system control middleware determines which computer resource of the plurality of computers is used for processing in response to a processing request or resource consumption for the computer, and changes the state of the computer. Here, the system configuration control unit 40 determines which of the plurality of computers is to be used for processing. On the other hand, the system efficiency control unit 20 changes the state of the computer.
　シミュレーション上では、実機と同様の実装がされたソフトウェアが動作するようにしてもよい。一方、実機におけるデータ構造やアルゴリズムの一部が模擬的に実装された簡易なソフトウェアが動作するようにしてもよい。簡易なソフトウェアの場合には、実際のソフトウェアを評価する場合と比較して、実際とは異なる動作により、システム制御ミドルウェアの評価としての精度が低下するおそれがある。しかしながら、簡易なソフトウェアによると、シミュレーションの計算量が減少し、シミュレーション実装コストが減少するという効果がもたらされる。なお、システム制御ミドルウェアのシミュレーション上での表現方式は、特に限定しない。 In the simulation, software with the same implementation as the actual machine may be operated. On the other hand, simple software in which a part of a data structure or algorithm in an actual machine is implemented in a simulated manner may be operated. In the case of simple software, compared with the case of evaluating actual software, there is a possibility that the accuracy of evaluation of the system control middleware may be reduced by an operation different from the actual software. However, simple software has the effect of reducing the amount of simulation calculation and reducing the simulation implementation cost. In addition, the expression method on the simulation of the system control middleware is not particularly limited.
　以上の構成を備えたシミュレーション装置１００は、シミュレータ制御部１１０から時刻更新情報を受けて、次のように動作する。 The simulation apparatus 100 having the above configuration operates as follows upon receiving time update information from the simulator control unit 110.
　まず、負荷をかけるクライアントを模擬する計算機利用負荷計算部３０は、所定のプログラムに従って、負荷を発生させる計算機に対する負荷を生成する。 First, the computer usage load calculation unit 30 that simulates a client to which a load is applied generates a load on the computer that generates the load according to a predetermined program.
　システム制御ミドルウェアを模擬する部分のうちの、負荷の到達先を決定するシステム構成制御部４０は、生成された負荷を受ける。システム構成制御部４０は、１つまたは複数の計算機ノードの動作を模擬するノード動作模擬部１０へ負荷を出力する。ノード動作模擬部１０の負荷受信部１１は、負荷を受け取る。 Of the part that simulates the system control middleware, the system configuration control unit 40 that determines the destination of the load receives the generated load. The system configuration control unit 40 outputs a load to the node operation simulation unit 10 that simulates the operation of one or more computer nodes. The load receiving unit 11 of the node operation simulation unit 10 receives a load.
　性能計算部１３は、負荷に従って、性能を計算する。さらに、電力計算部１５は、負荷の有無に依らず、そのシミュレーション時間における消費電力を計算する。 The performance calculator 13 calculates the performance according to the load. Furthermore, the power calculator 15 calculates the power consumption during the simulation time regardless of the presence or absence of a load.
　システム制御ミドルウェアのシステム効率化制御部２０は、逐次、または、所定のシミュレーション時間タイミングもしくは任意のシミュレーション時間タイミングごと、性能情報および消費電力情報を受信する。システム効率化制御部２０は、これら情報に基づいて、データ配置などのシステム構成を変更するか否かを判定する。システム構成を変更する場合には、システム効率化制御部２０は、システム構成情報を変更する制御命令をノード動作模擬部１０およびシステム構成制御部４０に出力する。 The system efficiency improvement control unit 20 of the system control middleware receives the performance information and the power consumption information sequentially or for each predetermined simulation time timing or any simulation time timing. The system efficiency improvement control unit 20 determines whether or not to change the system configuration such as data arrangement based on the information. When changing the system configuration, the system efficiency improvement control unit 20 outputs a control command for changing the system configuration information to the node operation simulation unit 10 and the system configuration control unit 40.
　ノード動作模擬部１０およびシステム構成制御部４０では、シミュレーション時間またはシミュレーション時間中の一瞬を消費し、システムの状態を変更する。システムの状態を変更する際に各計算機のリソースを利用する場合には、制御負荷発生部１２は、負荷受信部１１に負荷を出力するようにしてもよい。 The node operation simulation unit 10 and the system configuration control unit 40 consume the simulation time or a moment during the simulation time and change the system state. When using the resources of each computer when changing the system state, the control load generator 12 may output the load to the load receiver 11.
　以上の処理を所定のシミュレーション実行時間内に繰り返すことで、システム利用負荷に対するシステム制御ミドルウェアの振る舞いを記録した動作結果ログ１２０が得られる。 By repeating the above processing within a predetermined simulation execution time, an operation result log 120 in which the behavior of the system control middleware with respect to the system usage load is recorded is obtained.
　システム制御ミドルウェア開発者やシステム利用者は、動作結果ログ１２０を解析することで、システム制御ミドルウェアの振る舞いの正当性および効果を評価することができる。ここで、シミュレーション動作中の各計算機の性能および消費電力は、これらの値を入力としつつフィードバック制御を行うシステム制御ミドルウェアによるシステム構成変更によって、シミュレーション動作内で変化する。したがって、本実施形態のシミュレーション装置によると、分散計算機システムに対するフィードバック制御を行うシステム制御ミドルウェアの正当性および効果を評価することができる。 The system control middleware developer or system user can evaluate the correctness and effect of the behavior of the system control middleware by analyzing the operation result log 120. Here, the performance and power consumption of each computer during the simulation operation change within the simulation operation by a system configuration change by the system control middleware that performs feedback control while using these values as inputs. Therefore, according to the simulation apparatus of the present embodiment, it is possible to evaluate the legitimacy and effect of the system control middleware that performs feedback control on the distributed computer system.
　（第３の実施形態）
　第３の実施形態に係るシミュレーション装置について、図面を参照して説明する。 (Third embodiment)
A simulation apparatus according to a third embodiment will be described with reference to the drawings.
　図７は、本実施形態のシミュレーション装置の構成を示すブロック図である。図７を参照すると、本実施形態のシミュレーション装置では、ノード動作模擬部１０の性能計算部３３と電力計算部３５との間で、電力係数および性能係数をやり取りする。性能計算部３３は、負荷によって変動する電力係数を電力計算部３５に出力する。一方、電力計算部３５は、性能係数を性能計算部３３に出力する。 FIG. 7 is a block diagram showing the configuration of the simulation apparatus of the present embodiment. Referring to FIG. 7, in the simulation apparatus according to the present embodiment, the power coefficient and the performance coefficient are exchanged between the performance calculation unit 33 and the power calculation unit 35 of the node operation simulation unit 10. The performance calculator 33 outputs a power coefficient that varies depending on the load to the power calculator 35. On the other hand, the power calculation unit 35 outputs the performance coefficient to the performance calculation unit 33.
　図８は、本実施形態のシミュレーション装置の構成をさらに詳細に示すブロック図である。図８を参照すると、性能計算部３３は、待ち行列部４１とサービス部４２を含む待ち行列ネットワークである。電力計算部３５は、一例として図５に示したよう動作モードの状態遷移図で表されているものとする。 FIG. 8 is a block diagram showing the configuration of the simulation apparatus of this embodiment in more detail. Referring to FIG. 8, the performance calculation unit 33 is a queuing network including a queuing unit 41 and a service unit 42. As an example, the power calculation unit 35 is represented by a state transition diagram of the operation mode as illustrated in FIG.
　このとき、電力係数を待ち行列内の処理待ち負荷数とし、性能係数を各動作モードで定義された値（ｓｐｅｅｄ）とすることができる。 At this time, the power coefficient can be the number of processing waiting loads in the queue, and the performance coefficient can be a value defined in each operation mode.
　以上の構成を備えたシミュレーション装置２００は、シミュレータ制御部１１０から時刻更新情報を受けて、次のように動作する。なお、ノード動作模擬部１０が、負荷を受信するまで動作は、第２の実施形態に係るシミュレーション装置１００の動作と同様であることから、説明を省略する。 The simulation apparatus 200 having the above configuration receives time update information from the simulator control unit 110 and operates as follows. The operation until the node operation simulation unit 10 receives a load is the same as the operation of the simulation apparatus 100 according to the second embodiment, and thus the description thereof is omitted.
　性能計算部３３が負荷を受け取った場合に、すでに他の負荷によってサービス部４２が利用されていたときには、負荷は待ち行列部４１に入力される。待ち行列中にすでに処理待ち負荷が入力されていた場合には、サービス部４２の処理負荷が終了するごとに、待ち行列の先頭からサービス部４２へ負荷が入力される。 When the performance calculation unit 33 receives a load and the service unit 42 has already been used by another load, the load is input to the queue unit 41. If a processing load is already input in the queue, the load is input from the head of the queue to the service unit 42 every time the processing load of the service unit 42 is completed.
　負荷の処理時間は、負荷が待ち行列に登録されてからサービス部４２を出るまでの時間によって与えられる。性能計算部３３は、当該シミュレーション時間において、電力計算部３５から出力された性能係数に基づいて、負荷がサービス部４２に入力されてから出力されるまでの時間を計算する。 The load processing time is given by the time from when the load is registered in the queue until the service unit 42 is exited. The performance calculation unit 33 calculates the time from when the load is input to the service unit 42 until it is output based on the performance coefficient output from the power calculation unit 35 during the simulation time.
　一例として、図５において動作モードがモードｍｏｄｅ０の場合、性能係数ｓｐｅｅｄはＳ０である。そこで、性能係数Ｓ０に対して、サービス部４２における単位負荷あたりの性能Ｂ、および、負荷の大きさｗを乗じたＳ０×Ｂ×ｗ（ミリ秒）を、その負荷のサービス時間とすることができる。なお、本実施形態におけるサービス時間の計算方法は、かかる方法に限定されない。 As an example, when the operation mode is mode mode 0 in FIG. 5, the performance coefficient speed is S 0. Therefore, S0 × B × w (milliseconds) obtained by multiplying the performance coefficient S0 by the performance B per unit load in the service unit 42 and the load size w may be set as the service time of the load. it can. Note that the service time calculation method in the present embodiment is not limited to such a method.
　電力計算部３５は、性能計算部３３から出力された電力係数に基づいて、当該シミュレーション時刻における消費電力を算出する。一例として、待ち行列の系内負荷数をｎとし、動作モードｍｏｄｅ０における消費電力は、負荷ｗを入力とする関数ｆ（ｗ）で求まるものとする。さらに、系内負荷数ｎを負荷の量ｗに変換する関数ｗ＝Ｔ（ｎ）が定義されている場合には、電力計算部３５は、電力ｆ（Ｔ（ｎ））を算出することができる。なお、関数Ｔ（ｎ）は、実機における実験等によって推測することができる。なお、本実施形態における消費電力の計算方法は、かかる方法に限定されない。 The power calculator 35 calculates the power consumption at the simulation time based on the power coefficient output from the performance calculator 33. As an example, suppose that the number of in-system loads in the queue is n, and the power consumption in the operation mode mode 0 is obtained by a function f (w) having the load w as an input. Furthermore, when a function w = T (n) for converting the number n of loads in the system into the load amount w is defined, the power calculator 35 can calculate the power f (T (n)). it can. Note that the function T (n) can be estimated by an experiment or the like in an actual machine. Note that the power consumption calculation method in the present embodiment is not limited to such a method.
　以上の処理を所定のシミュレーション実行時間内に繰り返すことにより、計算機の性能と消費電力が相互に関係しつつ変化するような現実的なシステムにおいても、システム利用負荷に対するシステム制御ミドルウェアの振る舞いを記録した動作結果ログ１２０が得られる。 By repeating the above processing within a predetermined simulation execution time, the behavior of the system control middleware with respect to the system usage load was recorded even in a realistic system where the computer performance and power consumption change in relation to each other. An operation result log 120 is obtained.
　システム制御ミドルウェア開発者やシステム利用者は、動作結果ログ１２０を解析することで、システム制御ミドルウェアの振る舞いの正当性および効果を評価することができる。また、シミュレーション装置２００によると、待ち行列における状態のパラメータによって消費電力を簡単に算出することができる。したがって、各電子部品の詳細な振る舞いをシミュレートする場合と比較して、少ない計算量でシステム制御ミドルウェアの正当性および効果を評価することができる。 The system control middleware developer or system user can evaluate the correctness and effect of the behavior of the system control middleware by analyzing the operation result log 120. Further, according to the simulation apparatus 200, the power consumption can be easily calculated based on the state parameters in the queue. Therefore, it is possible to evaluate the legitimacy and effect of the system control middleware with a small amount of calculation compared to the case of simulating the detailed behavior of each electronic component.
　（第４の実施形態）
　第４の実施形態に係る分散システムについて説明する。本実施形態の分散システムでは、複数の計算機を含む分散計算機システムをシミュレータ内蔵システム制御ミドルウェアによって制御する。 (Fourth embodiment)
A distributed system according to the fourth embodiment will be described. In the distributed system of the present embodiment, a distributed computer system including a plurality of computers is controlled by simulator built-in system control middleware.
　シミュレータ内蔵システム制御ミドルウェアは、第２および第３の実施形態におけるリソース制御命令を発行するソフトウェアである。また、シミュレータ内蔵システム制御ミドルウェアは、入力された電力係数および性能係数を用いてシステムの動作を決定する際、システム動作決定部に内蔵された第２または第３の実施形態におけるシミュレーションソフトウェアにこれらの係数を入力し、将来のシステムの状態を予測する。ここで、シミュレーションソフトウェアを複数回、異なる乱数などを与えて実行してもよい。シミュレータ内蔵システム制御ミドルウェアは、シミュレーションソフトウェアの実行内容に基づいて、将来最適となるリソース制御命令を決定し、システムを制御する。 The simulator built-in system control middleware is software that issues resource control instructions in the second and third embodiments. Further, when the system control middleware with built-in simulator determines the operation of the system using the input power coefficient and performance coefficient, the simulation software in the second or third embodiment built in the system operation determining unit includes these. Enter coefficients to predict future system conditions. Here, the simulation software may be executed a plurality of times by giving different random numbers. The system control middleware with built-in simulator determines a resource control instruction that will be optimal in the future based on the execution contents of the simulation software, and controls the system.
　本発明の全開示（請求の範囲を含む）の枠内において、さらにその基本的技術思想に基づいて、実施形態の変更・調整が可能である。また、本発明の請求の範囲の枠内において種々の開示要素の多様な組み合わせないし選択が可能である。すなわち、本発明は、請求の範囲を含む全開示、技術的思想にしたがって当業者であればなし得るであろう各種変形、修正を含むことは勿論である。 In the frame of the entire disclosure (including claims) of the present invention, the embodiment can be changed and adjusted based on the basic technical concept. Various combinations and selections of various disclosed elements are possible within the scope of the claims of the present invention. That is, the present invention of course includes various variations and modifications that could be made by those skilled in the art according to the entire disclosure including the claims and the technical idea.
　本発明に係るシミュレーションソフトウェア、シミュレーション方式、シミュレータ内蔵システム制御ミドルウェアは、計算機システムのシミュレーションソフトウェア、シミュレータ、評価装置、計算機動作のデモンストレーション、分散システムのスケジューラ、データベースシステム、ストレージシステム、分散キーバリューストアに適用することができる。 The simulation software, simulation method, and simulator built-in system control middleware according to the present invention are applied to computer system simulation software, simulators, evaluation devices, computer operation demonstrations, distributed system schedulers, database systems, storage systems, and distributed key-value stores. can do.
　上記実施形態の一部または全部は、以下の付記として記載することができるものであるが、これらに限定されるものではない。 Some or all of the above embodiments can be described as the following supplementary notes, but are not limited thereto.
　（付記１）分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する性能計算部と、
　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算部と、
　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新するシステム動作決定部と、を備えていることを特徴とするシミュレーション装置。 (Supplementary Note 1) A performance calculation unit that calculates the processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load;
A power calculator that calculates power consumption required to process each of the loads assigned to each of the plurality of processing nodes;
A simulation apparatus comprising: a system operation determination unit that updates allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
　（付記２）前記電力計算部は、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、付記１に記載のシミュレーション装置。 (Supplementary Note 2) The power calculation unit changes each of the plurality of processing nodes to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time. The simulation apparatus according to attachment 1, wherein the simulation apparatus is characterized.
　（付記３）前記電力計算部は、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、付記２に記載のシミュレーション装置。 (Supplementary note 3) The simulation according to supplementary note 2, wherein the power calculation unit calculates the power consumption according to power consumption per unit time in each of the current operation modes of the plurality of processing nodes. apparatus.
　（付記４）前記性能計算部は、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、付記２または３に記載のシミュレーション装置。 (Supplementary Note 4) The simulation apparatus according to Supplementary Note 2 or 3, wherein the performance calculation unit calculates the processing time according to a processing speed in each of the current operation modes of the plurality of processing nodes.
　（付記５）前記電力計算部は、前記処理時間に応じて、前記消費電力を計算することを特徴とする、付記１ないし４のいずれか一に記載のシミュレーション装置。 (Supplementary Note 5) The simulation apparatus according to any one of Supplementary Notes 1 to 4, wherein the power calculation unit calculates the power consumption according to the processing time.
　（付記６）前記性能計算部は、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、付記１に記載のシミュレーション装置。 (Supplementary note 6) The simulation apparatus according to supplementary note 1, wherein the performance calculation unit calculates the processing time based on a queuing network in which each of the plurality of processing nodes is modeled.
　（付記７）前記電力計算部は、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、付記６に記載のシミュレーション装置。 (Supplementary note 7) The simulation apparatus according to supplementary note 6, wherein the power calculation unit calculates the power consumption according to the number of loads stored in a queue of the queue network.
　（付記８）前記システム動作決定部は、前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新することを特徴とする、付記１ないし７のいずれか一に記載のシミュレーション装置。 (Supplementary note 8) Any one of Supplementary notes 1 to 7, wherein the system operation determination unit updates allocation of data and processing to each of the plurality of processing nodes based on the processing time and the power consumption. The simulation device according to any one of the above.
　（付記９）前記複数の処理ノードは、それぞれ、前記分散計算機システムに含まれる計算機、または、該計算機に含まれるＣＰＵ、ストレージもしくはメモリであることを特徴とする、付記１ないし８のいずれか一に記載のシミュレーション装置。 (Supplementary note 9) Any one of Supplementary notes 1 to 8, wherein each of the plurality of processing nodes is a computer included in the distributed computer system, or a CPU, storage, or memory included in the computer. The simulation apparatus described in 1.
　（付記１０）前記処理時間および前記消費電力を保持する動作結果ログをさらに備えていることを特徴とする、付記１ないし９のいずれか一に記載のシミュレーション装置。 (Supplementary note 10) The simulation apparatus according to any one of Supplementary notes 1 to 9, further comprising an operation result log that holds the processing time and the power consumption.
　（付記１１）付記１ないし１０のいずれか一に記載のシミュレーション装置と、
　前記シミュレーション装置によって、動作モード、および／または、負荷の割り当てが更新される複数の処理ノードと、を備えていることを特徴とする分散計算機システム。 (Supplementary Note 11) The simulation apparatus according to any one of Supplementary Notes 1 to 10,
A distributed computer system comprising: a plurality of processing nodes whose operation modes and / or load assignments are updated by the simulation apparatus.
　（付記１２）コンピュータが、分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算工程と、
　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算工程と、
　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新工程と、を含むことを特徴とするシミュレーション方法。 (Supplementary note 12) A time calculation step in which a computer calculates a processing time required for each of a plurality of processing nodes included in a distributed computer system to process an assigned load;
A power calculation step of calculating power consumption required for processing each of the plurality of processing nodes assigned load;
An updating step of updating allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
　（付記１３）前記電力計算工程において、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、付記１２に記載のシミュレーション方法。 (Supplementary note 13) In the power calculation step, each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time. The simulation method according to attachment 12, wherein the simulation method is characterized.
　（付記１４）前記電力計算工程において、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、付記１３に記載のシミュレーション方法。 (Supplementary note 14) The simulation according to supplementary note 13, wherein, in the power calculation step, the power consumption is calculated according to power consumption per unit time in a current operation mode of each of the plurality of processing nodes. Method.
　（付記１５）前記時間計算工程において、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、付記１３または１４に記載のシミュレーション方法。 (Supplementary note 15) The simulation method according to Supplementary note 13 or 14, wherein, in the time calculation step, the processing time is calculated according to a processing speed in each of the current operation modes of the plurality of processing nodes.
　（付記１６）前記電力計算工程において、前記処理時間に応じて、前記消費電力を計算することを特徴とする、付記１２ないし１５のいずれか一に記載のシミュレーション方法。 (Supplementary note 16) The simulation method according to any one of Supplementary notes 12 to 15, wherein, in the power calculation step, the power consumption is calculated according to the processing time.
　（付記１７）前記時間計算工程において、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、付記１２に記載のシミュレーション方法。 (Supplementary note 17) The simulation method according to supplementary note 12, wherein, in the time calculation step, the processing time is calculated based on a queuing network in which each of the plurality of processing nodes is modeled.
　（付記１８）前記電力計算工程において、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、付記１７に記載のシミュレーション方法。 (Supplementary note 18) The simulation method according to supplementary note 17, wherein in the power calculation step, the power consumption is calculated according to the number of loads stored in the queue of the queue network.
　（付記１９）コンピュータが、前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新する工程をさらに含むことを特徴とする、付記１２ないし１８のいずれか一に記載のシミュレーション方法。 (Supplementary note 19) The computer according to any one of supplementary notes 12 to 18, further comprising a step of updating data and processing assignments for each of the plurality of processing nodes based on the processing time and the power consumption. The simulation method according to claim 1.
　（付記２０）分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算処理と、
　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算処理と、
　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新処理と、をコンピュータに実行させることを特徴とするプログラム。 (Supplementary note 20) A time calculation process for calculating a processing time required for each of a plurality of processing nodes included in the distributed computer system to process an assigned load;
A power calculation process for calculating power consumption required for processing each of the plurality of processing nodes assigned to the load;
A program that causes a computer to execute update processing for updating allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
　（付記２１）前記電力計算処理において、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、付記２０に記載のプログラム。 (Supplementary Note 21) In the power calculation process, each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time. The program according to appendix 20, which is characterized.
　（付記２２）前記電力計算処理において、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、付記２１に記載のプログラム。 (Supplementary note 22) The program according to supplementary note 21, wherein, in the power calculation process, the power consumption is calculated according to power consumption per unit time in each of the current operation modes of the plurality of processing nodes. .
　（付記２３）前記時間計算処理において、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、付記２１または２２に記載のプログラム。 (Supplementary note 23) The program according to supplementary note 21 or 22, wherein, in the time calculation process, the processing time is calculated according to a processing speed of each of the plurality of processing nodes in a current operation mode.
　（付記２４）前記電力計算処理において、前記処理時間に応じて、前記消費電力を計算することを特徴とする、付記２０ないし２３のいずれか一に記載のプログラム。 (Supplementary note 24) The program according to any one of supplementary notes 20 to 23, wherein in the power calculation process, the power consumption is calculated according to the processing time.
　（付記２５）前記時間計算処理において、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、付記２０に記載のプログラム。 (Supplementary note 25) The program according to supplementary note 20, wherein in the time calculation process, the processing time is calculated based on a queuing network in which each of the plurality of processing nodes is modeled.
　（付記２６）前記電力計算処理において、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、付記２５に記載のプログラム。 (Supplementary note 26) The program according to supplementary note 25, wherein, in the power calculation process, the power consumption is calculated according to the number of loads stored in the queue of the queue network.
　（付記２７）前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新する処理を、さらにコンピュータに実行させることを特徴とする、付記２０ないし２６のいずれか一に記載のプログラム。 (Supplementary note 27) According to the supplementary notes 20 to 26, the computer is further caused to execute a process of updating data and process assignments for each of the plurality of processing nodes based on the processing time and the power consumption. The program as described in any one.
１０　　ノード動作模擬部
１１　　負荷受信部
１２　　制御負荷発生部
１３、３３、５３　　性能計算部
１５、３５、５５　　電力計算部
２０　　システム効率化制御部
２１　　性能情報受信部
２２　　電力情報受信部
２３、６３　　システム動作決定部
２５　　制御命令発行部
３０　　計算機利用負荷生成部
４０　　システム構成制御部
４１　　待ち行列
４２　　サービス部
４６　　動作モード制御部
５０Ａ～５０Ｎ　　処理ノード
１００、２００　　シミュレータ
１１０　　シミュレータ制御部
１２０　　動作結果ログ DESCRIPTION OF SYMBOLS 10 Node operation | movement simulation part 11 Load receiving part 12 Control load generating part 13, 33, 53 Performance calculation part 15, 35, 55 Power calculation part 20 System efficiency control part 21 Performance information receiving part 22 Power information receiving part 23, 63 System Operation determining unit 25 Control command issuing unit 30 Computer utilization load generating unit 40 System configuration control unit 41 Queue 42 Service unit 46 Operation mode control unit 50A to 50N Processing node 100, 200 Simulator 110 Simulator control unit 120 Operation result log

Claims (25)

  1. 　分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する性能計算部と、
    　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算部と、
    　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新するシステム動作決定部と、を備えていることを特徴とするシミュレーション装置。     A performance calculator that calculates the processing time required for each of the plurality of processing nodes included in the distributed computer system to process the assigned load;
    A power calculator that calculates power consumption required to process each of the loads assigned to each of the plurality of processing nodes;
    A simulation apparatus comprising: a system operation determination unit that updates allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
  2. 　前記電力計算部は、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、請求項１に記載のシミュレーション装置。 The power calculation unit changes each of the plurality of processing nodes to one of a plurality of operation modes defined by a processing speed and power consumption per unit time, respectively. The simulation apparatus according to claim 1.
  3. 　前記電力計算部は、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、請求項２に記載のシミュレーション装置。 The simulation apparatus according to claim 2, wherein the power calculation unit calculates the power consumption according to the power consumption per unit time in the current operation mode of each of the plurality of processing nodes.
  4. 　前記性能計算部は、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、請求項２または３に記載のシミュレーション装置。 4. The simulation apparatus according to claim 2, wherein the performance calculation unit calculates the processing time according to a processing speed in a current operation mode of each of the plurality of processing nodes.
  5. 　前記電力計算部は、前記処理時間に応じて、前記消費電力を計算することを特徴とする、請求項１ないし４のいずれか１項に記載のシミュレーション装置。 The simulation apparatus according to any one of claims 1 to 4, wherein the power calculation unit calculates the power consumption according to the processing time.
  6. 　前記性能計算部は、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、請求項１に記載のシミュレーション装置。 2. The simulation apparatus according to claim 1, wherein the performance calculation unit calculates the processing time based on a queuing network that models each of the plurality of processing nodes.
  7. 　前記電力計算部は、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、請求項６に記載のシミュレーション装置。 The simulation apparatus according to claim 6, wherein the power calculation unit calculates the power consumption according to the number of loads stored in the queue of the queue network.
  8. 　前記システム動作決定部は、前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新することを特徴とする、請求項１ないし７のいずれか１項に記載のシミュレーション装置。 8. The system operation determination unit according to claim 1, wherein the system operation determination unit updates allocation of data and processing to each of the plurality of processing nodes based on the processing time and the power consumption. The simulation apparatus described in 1.
  9. 　請求項１ないし８のいずれか１項に記載のシミュレーション装置と、
    　前記シミュレーション装置によって、動作モード、および／または、負荷の割り当てが更新される複数の処理ノードと、を備えていることを特徴とする分散計算機システム。     A simulation apparatus according to any one of claims 1 to 8,
    A distributed computer system comprising: a plurality of processing nodes whose operation modes and / or load assignments are updated by the simulation apparatus.
  10. 　コンピュータが、分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算工程と、
    　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算工程と、
    　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新工程と、を含むことを特徴とするシミュレーション方法。     A time calculation step in which the computer calculates a processing time required for each of the plurality of processing nodes included in the distributed computer system to process the assigned load;
    A power calculation step of calculating power consumption required for processing each of the plurality of processing nodes assigned load;
    An updating step of updating allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
  11. 　前記電力計算工程において、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、請求項１０に記載のシミュレーション方法。 In the power calculation step, each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time. The simulation method according to claim 10.
  12. 　前記電力計算工程において、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、請求項１１に記載のシミュレーション方法。 12. The simulation method according to claim 11, wherein, in the power calculation step, the power consumption is calculated according to power consumption per unit time in each current operation mode of the plurality of processing nodes.
  13. 　前記時間計算工程において、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、請求項１１または１２に記載のシミュレーション方法。 The simulation method according to claim 11 or 12, wherein, in the time calculation step, the processing time is calculated according to a processing speed in each current operation mode of the plurality of processing nodes.
  14. 　前記電力計算工程において、前記処理時間に応じて、前記消費電力を計算することを特徴とする、請求項１０ないし１３のいずれか１項に記載のシミュレーション方法。 The simulation method according to any one of claims 10 to 13, wherein, in the power calculation step, the power consumption is calculated according to the processing time.
  15. 　前記時間計算工程において、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、請求項１０に記載のシミュレーション方法。 11. The simulation method according to claim 10, wherein in the time calculation step, the processing time is calculated based on a queuing network in which each of the plurality of processing nodes is modeled.
  16. 　前記電力計算工程において、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、請求項１５に記載のシミュレーション方法。 The simulation method according to claim 15, wherein, in the power calculation step, the power consumption is calculated according to the number of loads stored in the queue of the queue network.
  17. 　コンピュータが、前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新する工程をさらに含むことを特徴とする、請求項１０ないし１６のいずれか１項に記載のシミュレーション方法。 17. The computer according to claim 10, further comprising a step of updating data and processing assignments for each of the plurality of processing nodes based on the processing time and the power consumption. The simulation method described in 1.
  18. 　分散計算機システムに含まれる複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する処理時間を計算する時間計算処理と、
    　前記複数の処理ノードのそれぞれが割り当てられた負荷を処理するのに要する消費電力を計算する電力計算処理と、
    　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対する負荷の割り当てを更新する更新処理と、をコンピュータに実行させることを特徴とするプログラム。     A time calculation process for calculating a processing time required for each of a plurality of processing nodes included in the distributed computer system to process the assigned load;
    A power calculation process for calculating power consumption required for processing each of the plurality of processing nodes assigned to the load;
    A program that causes a computer to execute update processing for updating allocation of a load to each of the plurality of processing nodes based on the processing time and the power consumption.
  19. 　前記電力計算処理において、前記複数の処理ノードのそれぞれを、処理速度と単位時間当たりの消費電力でそれぞれ定義された複数の動作モードのうちのいずれかの動作モードに変更することを特徴とする、請求項１８に記載のプログラム。 In the power calculation process, each of the plurality of processing nodes is changed to any one of a plurality of operation modes defined by a processing speed and power consumption per unit time. The program according to claim 18.
  20. 　前記電力計算処理において、前記複数の処理ノードのそれぞれの現在の動作モードにおける単位時間当たりの消費電力に応じて前記消費電力を計算することを特徴とする、請求項１９に記載のプログラム。 20. The program according to claim 19, wherein, in the power calculation process, the power consumption is calculated in accordance with power consumption per unit time in each of the current operation modes of the plurality of processing nodes.
  21. 　前記時間計算処理において、前記複数の処理ノードのそれぞれの現在の動作モードにおける処理速度に応じて前記処理時間を計算することを特徴とする、請求項１９または２０に記載のプログラム。 21. The program according to claim 19 or 20, wherein, in the time calculation process, the processing time is calculated according to a processing speed in each current operation mode of the plurality of processing nodes.
  22. 　前記電力計算処理において、前記処理時間に応じて、前記消費電力を計算することを特徴とする、請求項１８ないし２１のいずれか１項に記載のプログラム。 The program according to any one of claims 18 to 21, wherein in the power calculation process, the power consumption is calculated according to the processing time.
  23. 　前記時間計算処理において、前記複数の処理ノードのそれぞれをモデル化した待ち行列ネットワークに基づいて、前記処理時間を計算することを特徴とする、請求項１８に記載のプログラム。 The program according to claim 18, wherein in the time calculation process, the processing time is calculated based on a queuing network in which each of the plurality of processing nodes is modeled.
  24. 　前記電力計算処理において、前記待ち行列ネットワークの待ち行列に格納された負荷の個数に応じて、前記消費電力を計算することを特徴とする、請求項２３に記載のプログラム。 The program according to claim 23, wherein in the power calculation process, the power consumption is calculated according to the number of loads stored in a queue of the queue network.
  25. 　前記処理時間および前記消費電力に基づいて、前記複数の処理ノードのそれぞれに対するデータおよび処理の割り当てを更新する処理を、さらにコンピュータに実行させることを特徴とする、請求項１８ないし２４のいずれか１項に記載のプログラム。 The computer according to any one of claims 18 to 24, further causing a computer to execute a process of updating data and process assignments to each of the plurality of processing nodes based on the processing time and the power consumption. The program described in the section.
PCT/JP2011/069294 2010-08-27 2011-08-26 Simulation device, distributed computer system, simulation method and program WO2012026582A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2010190641 2010-08-27
JP2010-190641 2010-08-27

Publications (1)

Publication Number Publication Date
WO2012026582A1 true WO2012026582A1 (en) 2012-03-01

Family

ID=45723571

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2011/069294 WO2012026582A1 (en) 2010-08-27 2011-08-26 Simulation device, distributed computer system, simulation method and program

Country Status (1)

Country Link
WO (1) WO2012026582A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016189105A (en) * 2015-03-30 2016-11-04 日本電気株式会社 Storage system
CN108153967A (en) * 2017-12-22 2018-06-12 深圳市康必达控制技术有限公司 The realization system of powernet emulation
CN109977541A (en) * 2019-03-26 2019-07-05 东南大学 A kind of power information physics system union simulation platform and its synchronous method
CN115344388A (en) * 2022-08-22 2022-11-15 南方电网科学研究院有限责任公司 Power real-time simulation calculation task allocation method and device considering communication and calculation power

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH11338744A (en) * 1998-05-21 1999-12-10 Toshiba Microelectronics Corp Software development supporting device
JP2006099579A (en) * 2004-09-30 2006-04-13 Toshiba Corp Information processor and information processing method
JP2006155525A (en) * 2004-12-01 2006-06-15 Internatl Business Mach Corp <Ibm> Control unit, information processing system, control method, and program therefor
WO2009131592A1 (en) * 2008-04-21 2009-10-29 Cluster Resources, Inc. System and method for managing energy consumption in a compute environment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH11338744A (en) * 1998-05-21 1999-12-10 Toshiba Microelectronics Corp Software development supporting device
JP2006099579A (en) * 2004-09-30 2006-04-13 Toshiba Corp Information processor and information processing method
JP2006155525A (en) * 2004-12-01 2006-06-15 Internatl Business Mach Corp <Ibm> Control unit, information processing system, control method, and program therefor
WO2009131592A1 (en) * 2008-04-21 2009-10-29 Cluster Resources, Inc. System and method for managing energy consumption in a compute environment

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MAKOTO KOBAYASHI: "Performance Evaluation of Computer Systems : Overview", THE JOURNAL OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS, vol. 80, no. 5, 25 May 1997 (1997-05-25), pages 518 *
YOSHIYUKI ONO ET AL.: "Utilizing GPU nodes for Low-power Data Intensive Computing", IEICE TECHNICAL REPORT, vol. 110, no. 167, 28 July 2010 (2010-07-28), pages 3 - 4 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016189105A (en) * 2015-03-30 2016-11-04 日本電気株式会社 Storage system
CN108153967A (en) * 2017-12-22 2018-06-12 深圳市康必达控制技术有限公司 The realization system of powernet emulation
CN109977541A (en) * 2019-03-26 2019-07-05 东南大学 A kind of power information physics system union simulation platform and its synchronous method
CN109977541B (en) * 2019-03-26 2023-03-31 东南大学 Power information physical system joint simulation platform and synchronization method thereof
CN115344388A (en) * 2022-08-22 2022-11-15 南方电网科学研究院有限责任公司 Power real-time simulation calculation task allocation method and device considering communication and calculation power
CN115344388B (en) * 2022-08-22 2023-12-01 南方电网科学研究院有限责任公司 Power real-time simulation calculation task allocation method and device considering communication and calculation force

Similar Documents

Publication Publication Date Title
Elmougy et al. A novel hybrid of Shortest job first and round Robin with dynamic variable quantum time task scheduling technique
US10175745B2 (en) Optimizing power consumption by dynamic workload adjustment
Guérout et al. Energy-aware simulation with DVFS
Xu et al. λdnn: Achieving predictable distributed DNN training with serverless architectures
US10168953B1 (en) Dynamic scheduling of distributed storage management tasks using predicted system characteristics
CN111344688B (en) Method and system for providing resources in cloud computing
Kurowski et al. DCworms–A tool for simulation of energy efficiency in distributed computing infrastructures
Qureshi Profile-based power-aware workflow scheduling framework for energy-efficient data centers
Castiglione et al. Modeling performances of concurrent big data applications
WO2022110446A1 (en) Simulation method and apparatus for heterogeneous cluster scheduling, computer device, and storage medium
US20140012561A1 (en) Modeling and evaluating application performance in a new environment
Zhu et al. FPGA resource pooling in cloud computing
WO2012026582A1 (en) Simulation device, distributed computer system, simulation method and program
Klusáček et al. Alea–complex job scheduling simulator
Bhimani et al. New performance modeling methods for parallel data processing applications
US8468006B2 (en) Method of combined simulation of the software and hardware parts of a computer system, and associated system
Ataie et al. Modeling and evaluation of dispatching policies in IaaS cloud data centers using SANs
Silva et al. An investigation of latency prediction for NoC-based communication architectures using machine learning techniques
WO2015177691A1 (en) Thread performance optimization
Sotiriadis et al. Towards inter-cloud simulation performance analysis: Exploring service-oriented benchmarks of clouds in SimIC
US10592473B2 (en) Method for improving energy efficiency of map-reduce system and apparatus thereof
Dietze et al. Water-level scheduling for parallel tasks in compute-intensive application components
Zohrati et al. Flexible approach to schedule tasks in cloud‐computing environments
CN114115887A (en) Adaptive scheduling with dynamic partition load balancing for fast partition compilation
CN108009074B (en) Multi-core system real-time evaluation method based on model and dynamic analysis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11820042

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 11820042

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP