JP2011198063A - Method, program and device for controlling data input/output - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique enabling high speed execution of input/output corresponding to an asynchronous input/output text of an application.SOLUTION: When the execution of an application starts, a thread generation part 125 generates a prescribed number of threads. A thread management information storage part 131 stores management information of the generated threads. When an asynchronous data input/output of the application designated by an asynchronous input/output text starts, a use data input/output determination part 126 selects processing to be executed by a thread-using input/output processing part 127 for the presence of a thread in an empty state but selects processing to be executed by a system-using input/output processing part 128 for the absence of a thread in an empty state. The thread-using input/output processing part 127 executes data input/output processing utilizing already generated threads. The system-using input/output processing part 128 executes data input/output processing utilizing asynchronous I/O of a system call and a system library.

Description

  The present invention relates to a data input / output control method, a data input / output control program, and a data input / output control device for controlling data input / output processing corresponding to an asynchronous input / output statement in an application.

For example, Fortran, a programming language, has an asynchronous input / output statement that specifies asynchronous data input / output processing. As an execution model of data input / output processing specified by such an asynchronous input / output statement, for example, a data input / output model using a system call or system library provided by a system such as an OS (Operating System), a thread, or the like There is a data input / output model that uses. Asynchronous I / O statements are also available in programming languages such as C and C ⁺⁺ .

  The data input / output model using system calls and system libraries provided by the system uses asynchronous I / O (AIO: Asynchronous Input / Output) technology for example. Data input / output is realized. In the data input / output model using threads, for example, a sub thread is generated separately from the main thread of the application, and data input / output is executed in parallel using the generated sub thread.

  A technique for generating one subtask (process) by a program and executing an asynchronous input / output statement by the generated process is known. In addition, a technique for generating a new thread for each file and executing an asynchronous input / output statement is known. Asynchronous I / O technology is also known in which the frequency of key access is managed, data is buffered in order of frequency, and the number of system calls issued is reduced.

Japanese Patent No. 25859797 JP 2000-194574 A JP-A-4-342011

  In data input / output using the system call or system library provided by the above system, there is a problem of waiting due to blocking in multiplexed I / O. In addition, in asynchronous I / O, system resources are shared with other applications, and performance degradation may be forced depending on the operation of other applications. In data input / output using threads, there are problems of thread generation costs and resource depletion due to excessive thread establishment.

  In order to execute data input / output specified in an asynchronous input / output statement of an application at high speed, it is necessary to use system resources efficiently and to execute efficient data input / output processing according to the situation. Is done.

  The present invention solves the above problems, uses system resources efficiently, and executes efficient data input / output processing according to the situation, thereby enabling data input / output corresponding to an asynchronous input / output statement of an application. It is an object of the present invention to provide a technique capable of executing the above at high speed.

  A data input / output control method that implements processing corresponding to an asynchronous input / output statement of an application created by a source program including an asynchronous input / output statement of a programming language by a computer is performed when the computer starts execution of the application. A process that generates a predetermined number of threads used to execute and stores management information for managing the generated threads in a storage device accessible by the computer, and a data input / output process corresponding to an asynchronous input / output statement in the application When starting, if there is an unused thread in the created thread by referring to the management information, it is determined that the thread use data I / O processing is performed, and the unused thread is created in the created thread. If there is no data, perform asynchronous I / O usage data input / output processing And the thread usage data I / O process, the management thread sets the unused thread status used for data I / O processing to the usage status until the data I / O processing is completed. Asynchronous I / O provided by a system call or system library when it is determined to execute data input / output processing that uses an unused thread and asynchronous I / O usage data input / output processing And a process of executing a data input / output process using the.

  With the above technology, it is possible to execute data input / output at high speed when executing an asynchronous input / output statement of an application by determining an appropriate execution model when executing data input / output processing.

It is a figure which shows the classification example of the model of a data input / output. It is a figure explaining multiplexed I / O. It is a figure explaining asynchronous I / O. It is a figure explaining the buffering of the data at the time of asynchronous I / O. It is a figure which shows the structure of the computer system by this Embodiment. It is a figure which shows the example of CPU. It is a figure explaining the compilation by this Embodiment. It is a figure which shows the example of the source program containing an asynchronous input-output sentence. It is a figure which shows the function structural example explaining the data input / output control function by this Embodiment. It is a figure which shows the example of the information stored in the thread management information storage part of this Embodiment. It is a figure which shows the example of the information stored in the buffer management information storage part of this Embodiment. It is a use data input / output determination process flowchart by the use data input / output determination part of this Embodiment. It is a figure explaining the example of the data input / output process using the already produced | generated thread by this Embodiment. It is a thread utilization data input / output processing flowchart by the thread utilization input / output processing unit of the present embodiment. It is a figure explaining the example of the data input process using the asynchronous I / O of the system by this Embodiment. It is a figure explaining the example of the data output process using the asynchronous I / O of the system by this Embodiment. It is a system utilization data input / output process flowchart by the system utilization input / output processing part of this Embodiment.

  Hereinafter, the present embodiment will be described with reference to the drawings.

  First, before explaining the data input / output control technique according to the present embodiment, a general data input / output technique will be briefly described.

  FIG. 1 is a diagram showing an example of classification of data input / output models.

  For example, as shown in FIG. 1, a general data input / output model can be classified according to whether it is synchronous or asynchronous, or whether blocking is performed or not.

  The synchronous data input / output model includes, for example, a data input / output model using system calls such as read and write. In the synchronous data input / output model, an application thread waits for a response from the kernel after issuing an I / O request to the kernel. The synchronous data input / output model includes a blocking model in which the application thread is blocked until the input / output processing in the kernel is completed, and processing to the application thread even if the input / output in the kernel is not completed. There is a non-blocking model in which is returned.

  Asynchronous data input / output models include, for example, multiple I / O that performs blocking and asynchronous I / O that does not perform blocking.

  FIG. 2 is a diagram for explaining multiplexed I / O.

  In FIG. 2, an application 500 is a user program including data input / output processing. The kernel 600 serves as a core of the OS, and provides basic functions such as management and control of computer hardware resources, control of hardware-software exchange, and interprocess communication.

  Multiplex I / O is a technique for handling a plurality of I / Os simultaneously by a single thread without blocking by read / write. In the multiple I / O, the application 500 executes system calls such as select (2), poll (2), and epoll (2), and waits for notification of an I / O executable state from the kernel 600.

  For example, as shown in FIG. 2, the application 500 executes read or write (S900), and returns an error from the kernel 600 with EAGAIN (S901). Note that EAGAIN indicates that there is no data ready for read / write in read / write non-blocking data input / output. The application 500 issues a system call select () (S 902) and waits for an I / O executable state notification from the kernel 600. After receiving the notification of the I / O executable state from the kernel 600 (S903), the application 500 performs data transfer by read or write (S904).

  As described above, in the multiplexed I / O, the application 500 is blocked while waiting for notification of the I / O executable state, and cannot perform any other processing. As a result, the waiting time for notification is wasted and data input / output performance cannot be improved.

  FIG. 3 is a diagram for explaining asynchronous I / O.

  In FIG. 3, a system library 700 is a library of functions provided by the OS. As an asynchronous I / O function included in the system library 700, for example, there is a POSIX (Portable Operating System Interface) AIO function. POSIX is a registered trademark.

  In asynchronous I / O, AIO-related system calls and POSIX AIO functions are used, and data input / output is executed without blocking. Since it does not block, the application 500 is characterized in that it can perform other processes in parallel even after issuing read / write. Asynchronous I / O includes, for example, asynchronous POSIX I / O that is commonly supported in the system and asynchronous I / O supported by the Linux kernel. Note that Linux is a registered trademark.

  In FIG. 3, the application 500 calls the AIO function of the system library 700 (S910). The system library 700 issues a data input / output system call to the kernel 600 by the AIO function (S911). At this time, in the application 500, the processing is not blocked by the call of the AIO function, so that other processing can be executed in parallel with the data input / output processing in the kernel 600.

  When the data input / output processing in the kernel 600 is completed, an I / O completion notification is sent from the kernel 600 to the system library 700 (S912), and the processing is returned from the system library 700 to the application 500 (S913).

  As described above, the asynchronous I / O can perform data input / output without blocking, so that the performance of data input / output processing is improved.

  However, in asynchronous I / O, the number of asynchronous I / O operations simultaneously queued and the maximum memory capacity limit that can be locked during asynchronous I / O transfer depend on the value applied to the entire system. Asynchronous I / O shares system resources with other applications, so depending on the operation of other applications that run in parallel, the queue and the maximum memory capacity that can be locked may exceed the limit and be in a wait state. Input / output performance may be degraded.

  FIG. 4 is a diagram for explaining data buffering during asynchronous I / O.

  Data buffering is a method in which, when data is exchanged between the kernel 600 and the application 500, the data is temporarily copied to a buffer held by the application 500, and some data is collected. By performing this method, the number of system calls issued can be reduced.

  For example, in FIG. 4, the application 500 transfers data to be output to a file from the memory area to the buffer by buffering processing (S920).

  When the data to be output to the file is collected in the buffer, the application 500 calls the AIO function of the system library 700 (S921). In the system library 700, a data input / output system call to the kernel 600 is issued by the AIO function (S922). By the processing of the kernel 600, the buffer data is output to a file.

  When the data output process in the kernel 600 is completed, an I / O completion notification is sent from the kernel 600 to the system library 700 (S923), and the process is returned from the system library 700 to the application 500 (S924).

  Such a buffering process can reduce the number of system call issuances, thereby reducing the number of system call queues. However, when buffering processing is performed in asynchronous I / O, there is a possibility that time from issuing the I / O to waiting for completion may be wasted.

  In the following, the data input / output control technique according to the present embodiment for solving the problems of the data input / output technique as described above will be described.

  FIG. 5 is a diagram showing the structure of a computer system according to this embodiment.

  The computer 1 includes hardware 30 such as a CPU (Central Processing Unit) 31, a memory 32 such as a RAM (Random Access Memory) and a ROM (Read Only Memory), and a disk 33 such as an HDD (Hard Disk Drive).

  In FIG. 5, a system area 20 is an area where a system such as the OS 21 operates. In the system area 20 shown in FIG. 5, a system library 22 that is a library of functions provided by the OS 21 is prepared. The program in the system area 20 is executed using the hardware 30.

  In FIG. 5, a user area 10 is an area that the user can use for running the application 11. In the user area 10 shown in FIG. 5, a compiler / library 12 that is a software package for creating an executable program from a source program of an application 11 written in a programming language is operated. In this embodiment, it is assumed that the compiler / library 12 is a compiler package that creates an executable program from a source program programmed in Fortran. The program in the user area 10 is executed using the hardware 30 through the OS 21 in the system area 20 or the like.

  FIG. 6 is a diagram illustrating an example of a CPU.

  The CPU 31 of the present embodiment is provided by a CPU chip 310 as shown in FIG. 6, for example. The CPU chip 310 is a physical package of the CPU 31. The CPU chip 310 is a multi-core processor having a plurality of CPU cores 311. In addition, a single CPU core 311 can be seen by a plurality of logical CPUs 312 from the software program by a multi-thread mechanism provided by the system.

  FIG. 7 is a diagram for explaining compilation according to the present embodiment.

  Here, an example will be described in which the source program 111 of the application 11 shown in FIG. 5 is compiled using the compiler / library 12 shown in FIG. 5 to generate the executable program 114 of the application 11. In FIG. 7, the compiler 121, the assembler 122, the linker 123, and a part of the runtime library 124 are included in the compiler / library 12 shown in FIG.

  The runtime library 124 is a set of programs that are generally used by the application 11 executed in the user area 10. The runtime library 124 of the present embodiment shown in FIG. 7 is used for general purposes in the system library 22 provided by the OS 21, the language providing library packaged in the compiler / library 12, and programs created by the user in the past. For example, an application library held for the purpose is included.

  The program that provides the data input / output control function of this embodiment is included in the runtime library 124. For example, the program that provides the data input / output control function of the present embodiment is provided by a language providing library packaged in the compiler / library 12.

  The source program 111 is the source code of the application 11. In the present embodiment, it is assumed that the source program 111 is programmed with Fortran and includes a Fortran asynchronous input / output statement. Fortran is a procedural programming language developed for scientific and engineering calculations.

  In the present embodiment, when the source program 111 is compiled, the compiler 121 intermediately converts the source program 111 into an assembler source 112. The assembler 122 assembles the assembler source 112 and generates an object code 113. The linker 123 links an object code 113 of the application 11 with a program such as a necessary function included in the runtime library 124 to generate an execution format program 114 of the application 11.

  In this embodiment, it is assumed that a dynamic link is used as the link. In other words, in the present embodiment, the linker 123 performs call resolution to a necessary program included in the runtime library 124 with respect to the object code 113 to generate an executable program 114. The program of the runtime library 124 is called on the memory area of the user area 10 when the execution format program 114 of the application 11 is executed.

  FIG. 8 is a diagram illustrating an example of a source program including asynchronous input / output statements.

  The source program 111 shown in FIG. 8 is described in Fortran and includes an asynchronous input / output statement that specifies asynchronous input / output of data.

  In the source program 111 shown in FIG. 8, the next two lines after “PROGRAM MAIN” are portions where variable declarations are described. The next four lines describe the variable initialization. Furthermore, the next five lines from the OPEN statement to the CLOSE statement are portions where data input / output is described.

  In the data input / output portion of the source program 111 shown in FIG. 8, “ASYNCHRONOUS = 'YES'” described in the OPEN statement, the WRITE statement, and the READ statement is a description for designating asynchronous input / output of data. That is, in the source program 111 shown in FIG. 8, an input / output statement including “ASYNCHRONOUS = 'YES'” is an asynchronous input / output statement.

  FIG. 9 is a diagram showing a functional configuration example for explaining the data input / output control function according to the present embodiment.

  In the computer 1, the source program 111 of the application 11 is converted into an execution format program 114 of the application 11 by the compiler 121, the linker 123, and the like. At this time, an interface necessary for calling the runtime library 124 is generated from the execution format program 114 of the application 11. An execution command for starting the execution format program 114 is also generated.

  In the runtime library 124, a thread generation unit 125, a use data input / output determination unit 126, a thread use input / output processing unit 127, and a system use input / output processing unit 128, depending on programs such as functions called from the execution format program 114 of the application 11. Is realized. Each of these functional units that provide the data input / output control function according to the present embodiment is realized by the hardware 30, such as the CPU 31, the memory 32, and the disk 33 provided in the computer 1, and a software program.

  When the execution format program 114 of the application 11 is activated by the execution command, execution start processing is performed. In the execution start process of the executable program 114, the thread generation unit 125 of the runtime library 124 generates a predetermined number of threads that are used in the execution of the application 11.

  When compiling the source program 111, the user designates whether the application 11 is executed in a multi-thread or a single thread. The thread generation unit 125 generates a plurality of threads when multithreading is specified at the time of compilation. The number of threads generated by the thread generation unit 125 is set in advance as appropriate in view of the performance of the CPU 31 and the memory 32. In the present embodiment, it is assumed that the number of threads generated by the thread generation unit 125 in the execution start process of the execution format program 114 of the application 11 is set to 8 including the main thread of the application 11. For example, the number of threads generated by the user when the execution format program 114 is executed may be specified.

  The thread generation unit 125 generates thread management information used for execution of the application 11 and stores it in the thread management information storage unit 131. The thread management information storage unit 131 is a storage device that can be accessed by the computer 1 and stores thread management information used in the execution of the application 11. The thread management information storage unit 131 stores information such as the process ID of the executed application 11, the number of threads generated for use in executing the application 11, and the state of each thread.

  When parallel processing is required in the execution of the application 11, in principle, the processing is executed using the thread generated by the thread generation unit 125. At this time, a program that performs parallel processing refers to the thread management information stored in the thread management information storage unit 131, selects an unused thread, and performs parallel processing. The thread used at this time is set to the use state in the thread management information stored in the thread management information storage unit 131. For example, in the source program 111 shown in FIG. 8, parallel processing using already generated threads is performed in the processing corresponding to the DO loop statements in the fourth to sixth lines.

  In the asynchronous data input / output start processing by the execution format program 114 of the application 11, the used data input / output determining unit 126 of the runtime library 124 determines the data input / output logic to be used. In the present embodiment, as the data input / output logic, a logic using an already generated thread by the thread use input / output processing unit 127 and a system call by the system use input / output processing unit 128 or an asynchronous I / O of the system library are used. Logic is provided in the runtime library 124. The asynchronous data input / output start process is a process corresponding to the OPEN statement of the source program 111 shown in FIG. 8, for example.

  In asynchronous data input / output execution processing by the execution format program 114 of the application 11, the thread use input / output processing unit 127 or the system use input / output processing unit 128 of the runtime library 124 executes data input / output processing.

  When the thread already generated by the thread generation unit 125 is empty, the thread use input / output processing unit 127 executes data input / output processing using the empty thread.

  The system use input / output processing unit 128 executes data input / output processing using asynchronous I / O provided by a system call or a system library.

  The buffer management information storage unit 132 is a storage device accessible by the computer 1 that stores management information of buffers prepared in the user area 10. In the data input / output processing by the thread use input / output processing unit 127 or the system use input / output processing unit 128, a buffer prepared in the user area 10 may be used. At this time, the thread use input / output processing unit 127 and the system use input / output processing unit 128 use the buffer in the user area 10 while referring to the buffer management information stored in the buffer management information storage unit 132.

  FIG. 10 is a diagram illustrating an example of information stored in the thread management information storage unit according to the present embodiment.

  A thread management table 135 illustrated in FIG. 10 is an example of thread management information generated for execution of the application 11 and stored in the thread management information storage unit 131. The thread management table 135 shown in FIG. 10 manages thread information for the application 11 whose process ID is “01” and the number of generated threads is N including the main thread. The process ID is identification information for uniquely identifying a process.

  The thread management table 135 has information such as a thread number and a thread state. The thread number is identification information for uniquely identifying a thread. The thread status is information indicating whether the thread is in an execution state, a wait state, or an unused state. In the thread management table 135 shown in FIG. 10, a thread whose thread state is “unused” is a free thread.

  FIG. 11 is a diagram illustrating an example of information stored in the buffer management information storage unit according to the present embodiment.

  The buffer management table 136 shown in FIG. 11 is an example of information for managing the buffer of the user area 10 that can be used by the application 11 and stored in the buffer management information storage unit 132. In the buffer management table 136 shown in FIG. 11, information on M buffers is managed.

  The buffer management table 136 has information such as buffer number and buffer usage status. The buffer number is identification information that uniquely identifies the buffer. The buffer usage status is information indicating whether or not the buffer is in use. In the buffer usage status of the buffer management table 136 shown in FIG. 11, “ON” indicates that it is in use, and “OFF” indicates that it is not being used.

  FIG. 12 is a use data input / output determination processing flowchart by the use data input / output determination unit of the present embodiment.

  The use data input / output determination unit 126 determines the data input / output logic to be used when the execution of the data input / output process corresponding to the asynchronous input / output statement of the source program 111 is started. For example, in the source program 111 shown in FIG. 8, “ASYNCHRONUS = 'YES'” indicating asynchronous input / output is set in the OPEN statement. As a process corresponding to the OPEN statement in which “ASYNCHRONUS = 'YES'” is set, the use data input / output determination unit 126 determines the data input / output logic to be used.

  As described above, the use data input / output determination unit 126 uses the data input / output logic using the already generated thread and the data input / output logic using the system call or the asynchronous I / O of the system library 22. Determine the data input / output logic to be performed. Hereinafter, an example of use data input / output determination processing by the use data input / output determination unit 126 will be described with reference to the flowchart of FIG.

  The usage data input / output determination unit 126 refers to the thread management information such as the thread management table 135 stored in the thread management information storage unit 131 (step S10). The usage data input / output determination unit 126 determines whether the processes can be executed in parallel (step S11). Here, the use data input / output determination unit 126 determines, for example, whether a plurality of threads are generated. In addition, the use data input / output determination unit 126 determines whether there is an idle thread (step S12). Here, the use data input / output determination unit 126 determines, for example, whether there is a thread whose thread state is “unused” in the thread management table 135 shown in FIG.

  If parallel execution is possible (YES in step S11) and there is a free thread (YES in step S12), the use data input / output determination unit 126 uses the thread use data input / output as the data input / output logic to be used. Processing is determined (step S13). At this time, the use data input / output determination unit 126 performs address registration of a module that realizes thread use data input / output processing. The thread use data input / output process is a data input / output process using an already generated thread by the thread use input / output processing unit 127. The use data input / output determination unit 126 determines a thread to be used for data input / output from the idle threads, and sets the state of the corresponding thread to the use state in the thread management table 135 stored in the thread management information storage unit 131. (Step S14). Note that in the case of data input / output processing using threads, in the thread management table 135 stored in the thread management information storage unit 131 when the processing corresponding to the CLOSE statement of the source program 111 shown in FIG. The thread state is returned to the unused state.

  When parallel execution is impossible (NO in step S11) or there is no idle thread (NO in step S12), the used data input / output determining unit 126 uses the system as the data input / output processing logic to be used. Usage data input / output processing is determined (step S15). At this time, the use data input / output determination unit 126 registers the address of the module that realizes the system use data input / output processing. The system use data input / output process is a data input / output process using a system call by the system use input / output processing unit 128 or an asynchronous I / O of the system library 22.

  In the present embodiment, if there is a free thread in the generated thread, the thread use data input / output processing by the thread use input / output processing unit 127 is preferentially selected. The already generated thread used for the data input / output process is a thread generated for executing the process of its own application 11.

  As described above, in the data input / output processing using the system library 22 and the asynchronous I / O of the system call, the maximum number of asynchronous I / O operations simultaneously queued and the maximum memory that can be locked during the asynchronous I / O transfer The capacity limit depends on the value applied throughout the system. Therefore, depending on the operation of other applications, the queue and the maximum memory capacity that can be locked exceed the limit, and there is a possibility that the performance of data input / output may be degraded.

  On the other hand, in the data input / output processing using the thread generated for executing the processing of the application 11, the maximum memory capacity that can be locked can be managed on the application 11 side. Therefore, it can be avoided that the maximum memory capacity that can be locked exceeds the limit, and the processing performance is hardly affected by the operation of other applications. As described above, when there is a free thread in the generated threads, the data input / output performance can be improved by giving priority to the data input / output using the thread.

  13 and 14, an example of data input / output processing using an already generated thread by the thread use input / output processing unit 127 when the thread use data input / output processing is determined by the use data input / output determination unit 126. Will be explained.

  FIG. 13 is a diagram for explaining an example of data input / output processing using an already generated thread according to the present embodiment.

  In FIG. 13, the main thread 115 indicates the main thread of the application 11. The sub thread A 116, the sub thread B 117, and the sub thread C 118 are threads that are generated at the start of execution of the application 11 and are in an idle state at the start of data input / output processing. The kernel 211 provides basic functions such as management and control of the hardware 30 of the computer 1, control of exchange between the hardware 30 and the user area 10, and interprocess communication as the core of the OS 21. The unit indicates a file.

  When the asynchronous input / output processing for the unit A is executed, the thread use input / output processing unit 127 performs data input / output using the sub thread A116 which is an empty thread instead of the main thread 115.

  The sub thread A116 of the application 11 calls a function for performing data input / output processing of the system library 22 (S100). Here, a function of synchronous input / output processing included in the system library 22 is called. A data input / output system call for unit A is issued by the called function of synchronous input / output processing (S101).

  Since the data input / output process for the unit A is executed by the sub thread A 116, the main thread 115 of the application 11 can execute other processes in parallel with the data input / output process by the sub thread A 116. In the example shown in FIG. 13, data input / output processing is performed by the sub thread A 116, but data input / output processing may be performed by the main thread 115, and other processing may be executed by the sub thread A 116.

  Similarly, when asynchronous I / O processing for unit B and asynchronous I / O processing for unit C are executed, if there is a free sub thread B 117 and sub thread C 118, the thread use I / O processing unit 127 Data input / output is performed using B117, sub thread C118. That is, the sub thread B 117 of the application 11 calls a function of synchronous input / output processing included in the system library 22 (S102). A data input / output system call for the unit B is issued by the function of the synchronous input / output processing included in the called system library 22 (S103). Similarly, the sub thread C118 of the application 11 calls a function of synchronous input / output processing included in the system library 22 (S104). A data input / output system call for unit C is issued by the function of the synchronous input / output process included in the called system library 22 (S105).

  While the data input / output processing is being performed by the sub thread A 116, the sub thread B 117, and the sub thread C 118, the main thread 115 of the application can execute other processing in parallel with the data input / output processing.

  A buffer prepared in the user area 10 may be used for data input / output processing. For example, the sub thread A 116 transfers the output data from the memory area of the application 11 to the buffer. The sub thread A 116 calls a data output processing function of the system library 22 and performs data output processing from the buffer to the file A. The sub thread A116 repeats these processes until all corresponding data on the memory area is output. Note that a plurality of buffers may be prepared, and data transfer to the buffer and function call of the system library 22 may be executed while alternately using the plurality of buffers.

  FIG. 14 is a thread use data input / output processing flowchart by the thread use input / output processing unit of this embodiment.

  The thread use input / output processing unit 127 causes the already-created thread in a free state assigned to the execution of the data input / output process to execute the process according to the flowchart shown in FIG. Here, a thread assigned to the execution of data input / output processing is referred to as a data input / output thread. The program that implements the thread use input / output processing unit 127 included in the runtime library 124 has, as arguments, the data on the memory area to be input / output, the address of the variable area, the length of the input / output data, The number of variables to be received.

  If the process to be executed is a data input process (input in step S20), the data input / output thread reads data from the disk 33 to the buffer in the user area 10 by the system call or the system library 22 (step S21). . The data input / output thread transfers the data on the buffer to the memory area of the application 11 (step S22). The data input / output thread repeats the processes of steps S21 and S22 until all input variables are completed (step S23).

  If the process to be executed is a data output process (output in step S20), the data input / output thread transfers the data to be output on the memory area of the application 11 to the buffer in the user area 10 (step S24). . The data input / output thread writes data from the buffer to the disk 33 by the system call or the system library 22 (step S25). The data input / output thread repeats the processing in steps S24 and S25 until the processing is completed for all data to be output on the memory area (step S26).

  In this embodiment, no thread is generated when data input / output processing using a thread is started. The thread used for the data input / output processing is a free thread among the threads generated by the thread generation unit 125 for general use in the execution of the application 11 when the execution of the application 11 is started. Since the cost of generating a thread for data input / output at the start of data input / output processing can be saved, the performance when executing data input / output processing is improved. Also, data input / output processing using threads is performed only when there is a free thread in the generated threads, and no new threads are added for data input / output processing. Therefore, the processing performance does not deteriorate due to an excessive number of threads for data input / output.

  A system call or asynchronous I / O of the system library 22 by the system usage input / output processing unit 128 when the system usage data input / output processing is determined by the usage data input / output determination unit 126 with reference to FIGS. An example of data input / output to be used will be described.

  In the system utilization data input / output processing by the system utilization input / output processing unit 128 according to the present embodiment, system calls provided from the system side such as the OS 21 and data input / output using asynchronous I / O supported by the system library 22 are performed. Done.

  However, the system utilization input / output processing unit 128 according to the present embodiment uses the buffer of the user area 10 and the non-use depending on the data input and output. That is, a plurality of buffers are used in data input processing using asynchronous I / O provided by the system, and no buffer is used in data output processing using asynchronous I / O provided by the system.

  In data input / output processing using asynchronous I / O of system calls and system library 22, data input and data output can be switched between using and not using the buffer in the user area 10 for faster data input / output Is possible.

  FIG. 15 is a diagram for explaining an example of data input processing using asynchronous I / O of the system according to this embodiment.

  In the data input process using the asynchronous I / O shown in FIG. 15, the AIO function provided by the system library 22 and M buffers are used to input data to the buffer, and from the buffer to the memory of the application 11. Data transfer to the area is performed in parallel. The buffer is managed by a buffer management table 136 stored in the buffer management information storage unit 132.

  When asynchronous input specifying buffer # 0 is executed, the application 11 calls an AIO function for performing data input processing of the system library 22 (S110). A system call for data input is issued by the called AIO function for performing data input processing (S111). Since the application 11 is asynchronous I / O, the application 11 can execute another process without waiting for completion of data input to the buffer # 0.

  Similarly, when asynchronous input designating buffer # 1 to buffer # M-1 is executed in sequence, an AIO function call for performing data input processing of system library 22 from application 11 (S112,... S114), AIO A system call for data input from the function is issued (S113,..., S115).

  When the data input to the buffer # 0 is completed, the kernel 211 returns an I / O completion notification for the buffer # 0 to the AIO function of the system library 22 (S116). The AIO function of the system library 22 returns processing to the application 11 (S117).

  Upon receiving the completion of data input to the buffer # 0, the application 11 executes data transfer processing from the buffer # 0 to the memory area of the application 11 in parallel with data input processing to other buffers. If there is still input data that cannot be stored in other buffers, asynchronous input of the next data specifying buffer # 0 is executed after completion of data transfer from buffer # 0 to the memory area.

  Similarly, when data input to the buffer # 1 is completed, the kernel 211 returns an I / O completion notification for the buffer # 1 to the AIO function of the system library 22 (S118), and the AIO function returns processing to the application 11. (S119). Upon receiving the completion of data input to the buffer # 1, the application 11 executes data transfer processing from the buffer # 1 to the memory area of the application 11 in parallel with data input processing to other buffers. Thereafter, the same processing is performed for the buffer # 2 and thereafter.

  As described above, in the data input using the asynchronous I / O of the system library 22 according to the present embodiment, in order to process as much data as possible on the application 11 side at the same time, Make input. The application 11 requests the AIO function of the system library 22 as much as necessary, and the data can be processed on the application 11 side when the data is transferred to the buffer. Thereby, for example, data input from the disk 33 to the buffer and processing of the input data by the application 11 can be executed in parallel. Therefore, the performance at the time of data input using the asynchronous I / O of the system is improved.

  FIG. 16 is a diagram for explaining an example of data output processing using asynchronous I / O of the system according to the present embodiment.

  In the data output processing using the asynchronous I / O shown in FIG. 16, the output target data on the memory area of the application 11 is output using the AIO function provided by the system library 22 without using the buffer. To do. Here, it is assumed that data output using asynchronous I / O is performed for each of the input variables X, Y, and Z.

  When asynchronous output for the input variable X is executed, the application 11 calls an AIO function for performing data output processing of the system library 22 (S120). At this time, information such as the address of the memory area in which the data to be output is stored and the length of the data to be output is passed to the AIO function that performs the data output process. A data output system call is issued by the called AIO function for performing data output processing (S121). Since the application 11 is asynchronous I / O, the application 11 can execute another process without waiting for the completion of data output for the input variable X.

  Similarly, when asynchronous output for the input variable Y and the input variable Z is sequentially executed, the application 11 calls the AIO function for performing the data output processing of the system library 22 (S122,... S123), and the data from the AIO function. An output system call is issued (S124,..., S125).

  When the data output for the input variable X is completed, the kernel 211 returns an I / O completion notification for the input variable X to the AIO function of the system library 22 (S126). The AIO function of the system library 22 returns processing to the application 11 (S127).

  Similarly, when the data output for the input variable Y and the input variable Z is completed, the kernel 211 returns an I / O completion notification for the input variable Y and the input variable Z to the AIO function of the system library 22 (S128, S130). The AIO function returns processing to the application 11 (S129, S131).

  As described above, in the data output using the asynchronous I / O of the system library 22 of this embodiment, the data in the memory area of the application 11 is output as it is without using the buffer. In data output using asynchronous I / O, when data is copied to the buffer, a wait state occurs while the copied data is passed to the system call. By passing data directly from the memory area of the application 11 to the system library 22 without using a buffer, it is possible to reduce the processing wait state of the application 11.

  FIG. 17 is a system use data input / output processing flowchart by the system use input / output processing unit of the present embodiment.

  If the process to be executed is a data input process (input in step S30), the system use input / output processing unit 128 refers to the buffer management table 136 stored in the buffer management information storage unit 132, and the buffer usage status is OFF. The following steps S31 to S34 are repeated for each buffer.

  The system use input / output processing unit 128 reads data from the disk 33 to the corresponding buffer in the user area 10 by the system call or the system library 22 (step S31). The system utilization input / output processing unit 128 sets the buffer usage status of the corresponding buffer to ON in the buffer management table 136 stored in the buffer management information storage unit 132 (step S32). The system use input / output processing unit 128 transfers the data on the buffer to the memory area of the application 11 (step S33). After receiving the I / O completion notification, the system usage input / output processing unit 128 sets the buffer usage status of the corresponding buffer to OFF in the buffer management table 136 stored in the buffer management information storage unit 132 (step S34).

  The system use input / output processing unit 128 repeats the processing for each buffer in steps S31 to S34 until all input variables are completed (step S35).

  If the process to be executed is the data output process (the output of step S30), the system use input / output processing unit 128 writes the data in the memory area of the application 11 to the disk 33 by the system call or the system library 22 (step S36). ). The system use input / output processing unit 128 repeats the processing in step S36 until the processing is completed for all data to be output in the memory area (step S37).

  By the data input / output control technology corresponding to the asynchronous input / output statement of this embodiment as described above, it is possible to determine an appropriate execution model at the start of data input / output processing and execute data input / output at high speed. It becomes.

  The processing by the application 11 or the runtime library 124 of the present embodiment described above can be realized by the hardware 31 such as the CPU 31 and the memory 32 provided in the computer and the software program, and the program can be read by the computer. It can be recorded on a medium or provided through a network.

  Although the present embodiment has been described above, the present invention can naturally be modified in various ways within the scope of the gist thereof.

DESCRIPTION OF SYMBOLS 1 Computer 10 User area 11 Application 111 Source program 112 Assembler source 113 Object code 114 Executable program 12 Compiler / library 121 Compiler 122 Assembler 123 Linker 124 Runtime library 125 Thread generation part 126 Use data input / output determination part 127 Thread use input / output Processing unit 128 System utilization input / output processing unit 131 Thread management information storage unit 132 Buffer management information storage unit 20 System area 21 OS
211 Kernel 22 System Library 30 Hardware 31 CPU
32 memory 33 disks

Claims (4)

A data input / output control method for realizing, by a computer, processing corresponding to the asynchronous input / output statement of an application created by a source program including an asynchronous input / output statement of a programming language,
The computer is
A process of generating a predetermined number of threads used to execute the application at the start of execution of the application, and storing management information for managing the generated threads in a storage device accessible by the computer;
At the start of data input / output processing corresponding to the asynchronous input / output statement in the application, referring to the management information, if there is an unused thread in the generated thread, thread utilization data is input. If it is determined that output processing is to be performed and there is no unused thread among the generated threads, processing to determine to perform asynchronous I / O utilization data input / output processing,
If it is determined that thread input / output data input / output processing is to be performed, the state of the unused thread to be used for data input / output processing in the management information is set to the use state until the data input / output processing is completed. Processing to execute data input / output processing using the thread in use,
When it is decided to perform data input / output processing using asynchronous I / O, a process for executing data input / output processing using asynchronous I / O provided by a system call or system library is executed. Data input / output control method. In the process of executing the data input / output process using the asynchronous I / O provided by the system call or the system library, the data input process inputs data using a plurality of buffers prepared in the user area of the computer. A data input / output control method comprising: performing data output without outputting a buffer prepared in a user area of the computer. A program for causing a computer to execute processing corresponding to an asynchronous input / output statement of an application created by a source program including an asynchronous input / output statement of a programming language,
In the computer,
A procedure for generating a predetermined number of threads used to execute the application at the start of execution of the application, and storing management information for managing the generated threads in a storage device accessible by the computer;
At the start of data input / output processing corresponding to the asynchronous input / output statement in the application, referring to the management information, if there is an unused thread in the generated thread, thread utilization data is input. A procedure for deciding to perform output processing and, if there are no unused threads among the generated threads, determining to perform asynchronous I / O utilization data input / output processing;
If it is determined that thread input / output data input / output processing is to be performed, the state of the unused thread to be used for data input / output processing in the management information is set to the use state until the data input / output processing is completed. A procedure for executing data input / output processing using a thread in use;
Data input / output to execute data input / output processing using asynchronous I / O provided by a system call or system library when it is decided to perform data input / output processing using asynchronous I / O Control program. A data input / output control device for executing processing corresponding to an asynchronous input / output statement of an application created by a source program including an asynchronous input / output statement of a programming language,
A thread generation unit that generates a predetermined number of threads used to execute the application at the start of execution of the application;
A thread management information storage unit for storing management information for managing the generated thread;
At the start of data input / output processing corresponding to the asynchronous input / output statement in the application, referring to the management information, if there is an unused thread in the generated thread, thread utilization data is input. A use data input / output determination unit that determines to perform an asynchronous I / O use data input / output process when there is no unused thread among the generated threads;
If it is determined that thread input / output data input / output processing is to be performed, the state of the unused thread to be used for data input / output processing in the management information is set to the use state until the data input / output processing is completed. A threaded input / output processing unit that executes data input / output processing using threads in use;
A system usage I / O processing unit that executes data I / O processing using asynchronous I / O provided by a system call or system library when it is decided to perform asynchronous I / O usage data I / O processing A data input / output control device.

Images