JP2007018268A - Task scheduling method, task scheduling device, and task scheduling program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve real-time property in task execution. <P>SOLUTION: In a task scheduling method which performs assignment of a plurality of tasks by a multiprocessor, the problem is solved by performing assignment of the above multiprocessor to a plurality of tasks based on; a task execution priority set up beforehand for each of a plurality of tasks; performance of each multiprocessor; turnaround time ratio of task execution performed in predetermined time among the multiprocessors; and the worst turnaround time ratio obtained from a result of measurement of the turnaround time ratio for predetermined number of times. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a task scheduling method, a task scheduling apparatus, and a task scheduling program, and more particularly to a task scheduling method, a task scheduling apparatus, and a task scheduling program for improving real-time performance in task execution.

  Generally, when a plurality of tasks are executed using a multiprocessor, it is necessary to schedule to which processor each task is assigned. FIG. 1 is a diagram illustrating an example of a schematic configuration of a multiprocessor system. As shown in FIG. 1, the multiprocessor system 10 includes a plurality of processors 11-1 to 11 -N, and the processors 11 are coupled to each other by a network 12 such as a shared bus. Further, a task scheduling device 13 is connected to the network 12, and tasks are assigned to the processors 11 by the task scheduling device 13.

  FIG. 2 is a diagram showing an example of a scheduling method of a conventional multiprocessor system. For example, as shown in FIG. 2, the scheduling method in the case where there are a plurality of tasks (task 0 to task 5) is normally statically assigned in advance to each processor 11-1 to 11-N in charge of each task. Only allowed to run on each assigned processor.

  In the scheduling method as described above, there is an execution permission time for each task. FIG. 3 is an example of a task execution permission time. As shown in FIG. 3, the task execution permission time indicates not the task execution time but the time from the start request generation to the task execution deadline. The task response time indicates the time from when a task activation request is generated until the task execution is completed. In this way, tasks to be executed statically are allocated in accordance with the processing capacity of the processor.

As a technique related to a multiprocessor system using such a scheduling method, there is one that reassigns a task for each processor by using a processor load as an index (see, for example, Patent Document 1).
JP 2004-192400 A

  By the way, some tasks may require real-time performance. Here, the real-time property indicates a property that task execution is completed within an execution permission time. At this time, since a task that can be executed for each processor is determined, there is a problem in that real-time performance cannot be ensured when many tasks having a long processing time are assigned to a certain processor.

  In addition, the scheduling method of a multiprocessor system as shown in Patent Document 1 reassigns tasks for each processor using the processor load as an index, and task assignment that can be executed for each processor is performed. Can be changed dynamically. However, in the scheduling method as shown in Patent Literature 1, although the calculation speed and response time reduction are considered as load indexes, the real time property is not dealt with, and task execution is not performed. There is a problem that a situation where the execution of the task is not completed within the permitted time may occur.

  The present invention has been made in view of the above-described problems, and provides a task scheduling method, a task scheduling apparatus, and a task scheduling program for improving real-time performance in task execution.

  In order to achieve the above-described object, the present invention provides a task scheduling method for assigning a plurality of tasks by a multiprocessor, a task execution priority preset for each of the plurality of tasks, and each of the multiprocessors. The plurality of tasks based on processing power, a response time ratio for task execution performed within a predetermined time of the multiprocessor, and a worst response time ratio obtained from a result of measuring the response time ratio for a predetermined number of times The multiprocessor is assigned to the above.

  Further, the present invention provides a task scheduling method for allocating a plurality of tasks by a multiprocessor, a processing capacity measuring step for measuring individual processing capacities of a plurality of processors constituting the multiprocessor, and for each of the plurality of tasks. A task is executed based on a task priority set in advance, an allocation step for initial allocation to each task from the processing capability result obtained in the processing capability measurement step, and a task allocation result obtained in the allocation step. A response time ratio measuring step for measuring a response time ratio of a task in time; a worst response ratio selecting step for measuring the response time ratio for a predetermined number of times and selecting a worst response time ratio from the measured response time ratios; , The worst response time ratio and setting of each task Based on the rate, and having a re-allocation step of performing the re-allocation of the processor to the task.

  As a result, it is possible to cause the optimum processor to execute the task based on the task priority, the processing capability of each processor, the response time ratio, and the worst response time ratio. Therefore, the real-time property in task execution can be improved.

  Furthermore, it is preferable that the reassignment step estimates a worst response time ratio in each task when reassigning the task, and reassigns the task to a processor having the lowest worst response time ratio. Thereby, the task can be allocated to the optimum processor by estimating the worst response time ratio of each task in advance.

  Furthermore, it is preferable that the reassignment step sets permission or prohibition of task reassignment for each of the plurality of processors. Thereby, permission or prohibition of reassignment can be set for the processor before reassignment is performed. Therefore, reallocation can be performed quickly without waste.

  Further, the present invention provides a task scheduling apparatus for allocating a plurality of tasks by a multiprocessor, a processing capacity measuring means for measuring individual processing capacities of a plurality of processors constituting the multiprocessor, and for each of the plurality of tasks. Based on the task priority set in advance and the processing capability result obtained by the processing capability measuring means, task allocation means for initial allocation to each task, and the task is executed based on the task allocation result obtained by the task allocation means. Response time ratio measuring means for measuring a response time ratio of a task at a predetermined time and further selecting a worst response time ratio from a plurality of response time ratios obtained by measuring the measured response time ratio for a predetermined number of times. And the task allocation means includes the worst response time ratio and the set ratio of each task. Based on, and performs reallocation of the processor to the task. As a result, it is possible to cause the optimum processor to execute the task based on the task priority, the processing capability of each processor, the response time ratio, and the worst response time ratio. Therefore, the real-time property in task execution can be improved.

  Furthermore, it is preferable that the task allocating unit estimates a worst response time ratio in each task when reassigning the task, and reassigns the task to a processor having the lowest worst response time ratio. Thereby, the task can be allocated to the optimum processor by estimating the worst response time ratio of each task in advance.

  Furthermore, it is preferable that the task assignment unit sets permission or prohibition of task reassignment for each of the plurality of processors. Thereby, permission or prohibition of reassignment can be set for the processor before reassignment is performed. Therefore, reallocation can be performed quickly without waste.

  Further, the present invention provides a task scheduling program for causing a computer to execute task scheduling for allocating a plurality of tasks by a multiprocessor, and measuring the processing capacity of each of the plurality of processors constituting the multiprocessor. Processing, task priority preset for each of the plurality of tasks, allocation processing for initial allocation to each task from the processing capability result obtained by the processing capability measurement processing, and task allocation result obtained by the allocation processing Based on the response time ratio measurement process for measuring the response time ratio of the task at a predetermined time, and measuring the response time ratio a predetermined number of times, and calculating the worst response time ratio from the measured response time ratios. The worst response ratio selection process to select and each Based on the above worst response time ratio and set the ratio of the disk, to perform a re-assignment processing for reallocation of the processor to the task to the computer. As a result, it is possible to cause the optimum processor to execute the task based on the task priority, the processing capability of each processor, the response time ratio, and the worst response time ratio. Therefore, the real-time property in task execution can be improved. In addition, task scheduling processing can be easily realized by installing an execution program in a computer.

  Furthermore, it is preferable that the reassignment process estimates a worst response time ratio in each task when reassigning the task, and reassigns the task to a processor having the lowest worst response time ratio. Thereby, the task can be allocated to the optimum processor by estimating the worst response time ratio of each task in advance.

  Furthermore, the reassignment process sets permission or prohibition of task reassignment for each of the plurality of processors. Thereby, permission or prohibition of reassignment can be set for the processor before reassignment is performed. Therefore, reallocation can be performed quickly without waste.

  According to the present invention, real-time performance in task execution can be improved.

  Hereinafter, preferred embodiments of a task scheduling method, a task scheduling apparatus, and a task scheduling program according to the present invention will be described with reference to the drawings. In the embodiment of the present invention described later, the general configuration of the multiprocessor system uses the same configuration as that in FIG.

  FIG. 4 is a diagram illustrating an example of a functional configuration of the task scheduling apparatus. The task scheduling apparatus 20 shown in FIG. 4 is configured to include a transmission / reception unit 21, a processing capability measurement unit 22, a task allocation unit 23, a response time ratio measurement unit 24, a storage unit 25, and a control unit 26. ing. The task scheduling device 20 is installed at the same position as the task scheduling device 13 shown in FIG.

  The transmission / reception means 21 transmits / receives control signals, data, and the like between the task scheduling apparatus 20 and the processors 11-1 to 11 -N via the network 12. Further, the processing capacity measuring means 22 measures the processing capacity (performance) of each processor when the system is activated. When measuring the processing capacity, a measurement task is used so that the processing capacity of each of the processors 11-1 to 11-N can be grasped by performing a preset process. The measurement task is stored in the storage unit 25 in advance. The processing capacity measurement unit 22 transmits the measurement task from the transmission / reception unit 21 to the processors 11-1 to 11-N via the network 12. The execution result obtained by being executed by each processor 11 is received by the transmission / reception means 21 via the network 12 and stored in the storage means 25. The above-described control is performed by the control means 26.

  The task allocation unit 23 is based on the priorities (task priorities) set in advance for a plurality of tasks to be actually executed from the processing capability results of the processors 11-1 to 11-N measured by the processing capability measuring unit 22. To perform initial task assignment. This priority may be set arbitrarily, or may be set based on the order of a plurality of tasks, the complexity of processing, and the like. Further, the task allocation unit 23 causes the control unit 26 to shift the multiprocessor system to the execution state after the initial allocation is completed. A plurality of tasks that are actually executed are stored in the storage unit 25 in advance.

  The task allocation unit 23 calculates a ratio of execution time (execution time ratio) within a preset execution permission time of the task, and each processor measured by the calculation result and a response time ratio measurement unit 24 described later. Based on the measurement result of the response time ratio to the task, the task that has not been executed yet and is stored in the storage unit 25 is reassigned. Note that task reallocation is performed using task priority, processor processing capacity, response time ratio, worst response time ratio, etc., for example, task execution is completed within the execution permission time, and the worst response time ratio is Tasks are reassigned so as to satisfy a preset condition that does not exceed a preset setting ratio for each of the processors 11-1 to 11-N. That is, the task is reassigned for each processor based on the worst response time ratio of each task and each processor processing capacity. Further, the task assignment unit 23 can set restrictions on whether to permit or prohibit task reassignment for each of the processors 11-1 to 11-N. A specific task assignment method will be described later.

  Next, the response time ratio measuring unit 24 measures the ratio (response time ratio) of the response time (time from the task activation request generation to the completion of task execution) within the preset task execution permission time. The response time ratio measuring means 24 repeatedly measures the response time ratio for a predetermined number of times set in advance, and selects the maximum one (worst response time ratio) among the measured response time ratios. Further, the response time ratio measuring unit 24 determines whether or not to reassign the processor to the task based on the worst response time ratio and the set ratio, and reassigns to the task allocation unit 23 as necessary. Output instructions. Further, the response time ratio measuring unit 24 stores the measurement result and the like in the storage unit 25.

  The storage unit 25 stores the above-described measurement task, stores the contents allocated by the task allocation unit 23, and stores the measurement results and the like in the response time ratio measurement unit 24. Further, when the multiprocessor system is terminated, the storage means 25 stores information on the processor assignment destination for each task at the present time. As a result, the processor assignment destination for each task set by the execution of the current system can be continued after the next system restart. As a result, the task can be executed quickly without having to redo the assignment process after restarting.

  In addition, the control unit 26 performs various operations and processes such as data input / output with each component in the task scheduling apparatus 20 and execution instructions, and controls the entire task scheduling apparatus 20. Specifically, the control unit 26, for example, shifts the multiprocessor system to an execution state or issues a task reassignment instruction.

  By using the task scheduling device 20 described above, it is possible to realize a scheduling method that dynamically changes task assignment for each processor so that task execution is completed within an execution permission time. Thereby, real-time property can be improved in a multiprocessor system.

  Here, the task scheduling apparatus 20 according to the present invention can perform task scheduling processing in a multiprocessor system using the above-described dedicated apparatus configuration, etc., but can execute a process in each configuration on a computer. By generating a program and installing the program in, for example, a general-purpose personal computer or server, the task scheduling process in the present invention can be realized.

<Hardware configuration>
Here, a hardware configuration example of a computer capable of executing the task scheduling process according to the present invention will be described with reference to the drawings. FIG. 5 is a diagram showing an example of a hardware configuration capable of realizing the task scheduling process according to the present invention.

  5 includes an input device 31, an output device 32, a drive device 33, an auxiliary storage device 34, a memory device 35, a CPU (Central Processing Unit) 36 for performing various controls, and a network connection device. 37, and these are connected to each other by a system bus B.

  The input device 31 has a pointing device such as a keyboard and a mouse operated by a user who uses a computer, and inputs various operation signals such as execution of a program from the user. The output device 32 has a display (monitor) for displaying various windows and data necessary for operating the computer main body for performing processing in the present invention, and the execution schedule of the task scheduling program is controlled by the control program of the CPU 36. And results can be displayed.

  Here, in the present invention, the execution program installed in the computer main body can be provided by the recording medium 38 such as a CD-ROM, for example. The recording medium 38 on which the program is recorded can be set in the drive device 33, and the execution program included in the recording medium 38 is installed in the auxiliary storage device 34 from the recording medium 38 via the drive device 33.

  The auxiliary storage device 34 is a storage unit such as a hard disk, and includes an execution program according to the present invention, a control program provided in a computer, a plurality of tasks such as the above-described measurement task, a processing performance measurement result, and a response time ratio measurement result. Etc. and can be input / output as necessary.

  Based on a control program such as an OS (Operating System) and an execution program read and stored by the memory device 35, the CPU 36 performs various operations and inputs / outputs data to / from each hardware component. Each process in the task scheduling program can be realized by controlling the process. Various information necessary during the execution of the program can be acquired from the auxiliary storage device 34 and can be stored.

  The network connection device 37 acquires an execution program from another terminal connected to the communication network by connecting to a communication network or the like, or an execution result obtained by executing the program or an execution in the present invention The program can be provided to other terminals. Further, the network connection device 37 can transmit / receive control signals and data to / from each processor.

  With the hardware configuration as described above, the task scheduling process according to the present invention can be realized at a low cost without requiring a special device configuration. In addition, task scheduling processing can be easily realized by installing a program.

<Task scheduling procedure>
Next, a task scheduling process procedure in the execution program according to the present invention will be described with reference to a flowchart. FIG. 6 is a flowchart of an example showing a task scheduling processing procedure in the present invention.

  When the task scheduling process is executed at the time of starting the system or in response to a program execution start instruction, it is first determined whether or not a processor assignment destination is set (stored) for each task (S01). In S12, which will be described later, the processor assignment destination for each task set by the previous system execution is made effective after the current system restart, so that it is not necessary to perform the assignment process every time the system is started. This is a process for executing a task.

  Here, when the processor assignment destination for each task is not set (NO in S01), the processing capacity of the processor is measured (S02). At this time, as described above, a measurement task is prepared in advance, and the processing task (performance) for each processor is measured by using the measurement task for each processor. Next, the task is initially assigned based on the processor processing capacity and the priority (task priority) given to each task (S03).

  Next, after the processing of S03 is completed or when the processor allocation destination is set in advance in the processing of S01 (YES in S01), the multiprocessor system is shifted to the task execution state (S04).

  Next, it is determined whether or not a predetermined period has been reached after execution (S05). If the predetermined period has been reached (YES in S05), the ratio of the response time of each task to the execution permission time (response time ratio) is calculated. To measure (S06). In addition, the response time ratio of each task measured by the process of S06 is stored by a storage unit or the like. Further, after the process of S06 is completed, it is determined whether the measurement (calculation) of the response time ratio exceeds a predetermined number (S07).

  Here, when the predetermined number of times is exceeded (YES in S07), that is, after repeating the measurement of the response time ratio for the predetermined number of times, the largest response time ratio (worst response time ratio) is selected. (S08). Further, after the process of S08 is completed, it is determined whether the task needs to be reassigned (S09). If the task needs to be reassigned (YES in S09), the worst response time ratio of each task and each The task is reassigned for each processor based on the processor processing capacity (S10). In the process of S10, if there is a processor whose worst response time ratio exceeds a preset setting ratio, the task is reassigned to a processor that does not exceed the setting ratio. Further, when performing task reassignment, the worst response time ratio of each task is estimated, and assignment is performed so that the task does not exceed the set ratio on the reassigned processor. Moreover, after the process of S10, it returns to S05 and repeats the process mentioned above.

  If the predetermined period has not been reached in the process of S05 (NO in S05), or if the predetermined number of times has not been exceeded in the process of S07 (NO in S07), there is no need to reassign tasks in the process of S09. In the case (NO in S09), or after the process of S10 is finished, it is determined whether or not the task scheduling process is finished (S11). If the process is not finished (NO in S11), the process returns to S05 and described above. Repeat the process.

  Further, when the process is terminated in the process of S11 (YES in S11), the assignment destination of the processor for each task is stored at this time (S12), and the task scheduling process is terminated.

  By the task scheduling process described above, it is possible to cause the optimum processor to execute the task based on the task priority, the processing capability for each processor, the response time ratio, and the worst response time ratio. Therefore, the real-time property in task execution can be improved. Further, load distribution of the processor can be realized with high accuracy. In addition, task scheduling processing can be easily realized by installing a program.

<Specific example of task scheduling processing>
Next, a specific example of task scheduling processing will be described with reference to the drawings. First, the processing capacity (performance) of each processor is measured at the time of system startup and recorded in a storage means such as a storage medium. Next, tasks are assigned to each processor based on the priority of each task and the processing capability of the processor. Here, FIG. 7 is a diagram illustrating an example of a correspondence table of processors and processing capabilities and setting contents for tasks. First, the correspondence table shown in FIG. 7A shows the processing capability of each processor ID by a numerical value. The processor ID is information for identifying each processor, and the numerical value and the number of digits are not particularly limited, and characters or the like may be used or numerical values and characters may be combined. Further, the numerical value of the processing capacity indicates how many times the processing capacity is set with respect to the processing capacity set in advance as “1” corresponding to the measurement task. That is, the higher the processing capability value shown in FIG. 7, the higher the processing capability.

  In the setting contents for the task shown in FIG. 7B, the task priority and the activation cycle are set in advance for each task ID. Here, the task ID is information for identifying the task, and the numerical value and the number of digits are not particularly limited. For example, as described above, a character or the like may be used. In addition, the task priority in FIG. 7B is set such that the smaller the numerical value is, the higher the priority is. However, the present invention is not limited to this, and may be reversed, for example. The task priority is not limited to a numerical value, and for example, characters such as “high”, “medium”, and “low”, or combinations thereof may be used.

  Furthermore, time information such as 2 [ms] or 3 [ms] is accumulated in the activation cycle shown in FIG. In the case of a single task that does not execute periodically, for example, a numerical value such as “0” or “non”, a character, or the like is set as the aperiodic activation information.

  Based on the priority of each task set as shown in FIG. 7 and the processing capability of each processor described above, task assignment for each processor is performed. Note that tasks are assigned one by one in order from the processor with the highest processing capability. The order of allocation includes, for example, task priority order, but is not particularly limited in the present invention.

  Here, FIG. 8 is a diagram illustrating an example of task assignment for each processor. In the example shown in FIG. 8, in order to explain the state of assignment more clearly, tasks with a high priority and processors with a high priority are arranged from the top of FIG. Here, first, each task is assigned to a processor in order from a task having a high priority to a processor having a high processing capability. In addition, as described above, after assigning one task to all the processors, the subsequent tasks (priority: maximum -N-1) are assigned from the processor 11-1 having the highest processing capability, and the subsequent tasks are also assigned. They are assigned in order of the processors with the highest processing capabilities.

  Here, when there are a plurality of tasks having the same priority, the task with the short activation cycle is assigned with priority, and the task that is not cyclically activated is set to the lowest priority. Further, after all task assignments are completed, the task assignments are recorded in a storage means such as a storage medium. Here, FIG. 9 is a diagram illustrating an example of an assignment correspondence table between task IDs and processors. As shown in FIG. 9, after the assignment is completed, the assignment destination processor ID corresponding to each task ID is stored. The processor ID “0” corresponds to the processor ID of the processor 11-1, and the processor ID “1” and later also correspond to the processor 11-2 and later.

  Moreover, processor ID should just be information which can identify a processor, and is not limited to a numerical value, For example, a character etc. may be sufficient. Note that for the assignment of processor IDs as shown in FIG. 9, it is necessary to determine the upper limit of the number of tasks in advance according to the processor capacity and the task load so that the processor does not become overloaded.

  Next, after assigning the tasks set as described above, the multiprocessor system is put into an execution state to execute each task. Moreover, the response time ratio of each task in a predetermined time is measured. Further, the measurement process is repeated a predetermined number of times, and is stored in a storage means such as a storage medium each time. Here, FIG. 10 is a diagram illustrating an example of a response time ratio measurement result. As shown in FIG. 10, the contents recorded by measuring the response time ratio are the task ID and the response time ratio corresponding to the task ID. Next, the response time ratio described above is recorded for a plurality of times, the worst response time ratio is selected from the recorded response time ratios, and the preset priority of each task is stored in the storage medium. Accumulate.

  Here, FIG. 11 is a diagram illustrating an example of the result of the worst response time ratio of each task to be stored. As shown in FIG. 11, the contents to be stored are the task priority and worst response time ratio for each task ID. Next, the presence / absence of a task whose worst response time ratio exceeds the set ratio for all assigned tasks is confirmed. Here, only a processor to which a task (including overrun) exceeding the set ratio is assigned performs a reallocation process described later.

  However, if there is only one task assigned under this condition, the reassignment prohibition flag set in advance for each of the processors 11-1 to 11-N is turned ON. Thereby, in the task reassignment in the task assigning means, it is removed from the processor to be assigned. That is, it is possible to set permission or prohibition of reassignment to a processor before performing reassignment, and reassignment can be performed quickly without waste. Further, concentration of processor load can be prevented.

  Therefore, when the reassignment prohibition flag is ON and when the worst response time ratio does not exceed the set ratio, the process proceeds to the process of measuring the response time ratio of each task for a predetermined time without performing the reallocation process described later. Returning, the process described above is repeated. In the present invention, the present invention is not limited to the prohibition flag. For example, if a permission flag is provided, the countermeasure can be taken by turning off the permission flag.

  Note that after the end of the task or after the end of the processing, the assignment destination processor ID shown in FIG. 9 may be updated in order to validate the contents assigned after the system restart. Thereby, a task can be assigned to an optimal processor.

<Task reassignment process>
Next, the task reallocation process described above will be described. First, the execution time of the task with the lowest priority in the processor in which the worst response time ratio of the task selected in the above-described process exceeds the set ratio is calculated. As a calculation procedure, first, the number of activations within the execution permission time of the lowest priority task is calculated for all tasks other than the lowest priority task. The execution time of tasks other than the lowest priority task is calculated based on the number of activations. Also, the execution time of the lowest priority task is calculated using the calculated execution time, the execution permission time of the lowest priority task, and the worst response time ratio of the lowest priority task.

  Here, a method for calculating the execution time of the lowest priority task will be described with reference to the drawings. FIG. 12 is a diagram illustrating an example of task assignment when the worst response time ratio exceeds the set ratio. In the example shown in FIG. 12, three tasks having different priorities are assigned to a certain processor. The assigned tasks are task 0 (startup cycle = 1 ms, priority = high), task 1 (startup cycle = 2 ms, priority = medium), and task 2 (startup cycle = 3 ms, priority = low). It has become. Further, the worst response time ratio for each task is assumed to be the worst response time ratio of task 0 = 0.6, the worst response time ratio of task 1 = 0.45, and the worst response time ratio of task 2 = 0.6. .

From the worst response time ratio R _wr1 of task 1 and the start cycle C ₁ , the worst response time T _wr1 per cycle of task 1 is T _wr1 = R _wr1 × C ₁ = 0.45 × 2 = 0.9 ms. _Further , the relationship between the task 0 activation cycle C ₀ and T _wr1 is “C ₀ > T _wr1 ”. For this reason, the number of activations of task 0 per cycle of task 1 is one.

Next, since task 0 has the highest priority, it is always executed with the worst response time. That is, it can be considered that execution time = worst response time. Here, due to the worst response time ratio R _{wr0 of} task 0 and the activation period C ₀ , the execution time T _e0 of task 0 in the worst response time of task 1 is T _e0 = R _wr0 × (1 × C ₀ ) = 0. 6 × (1 × 1) = 0.6 ms. As described above, the execution time T _e1 per cycle of the task 1 is T _e1 = T _wr1 −T _e0 = 0.9−0.6 = 0.3 ms. Further, the response time ratio R _e1 per cycle of the task 1 is R _e1 = T _e1 ÷ C1 = 0.3 ÷ 2 = 0.15.

Next, the worst case response time ratio _{R wr2} the activation period _{C 2} in the task 2 in FIG. 12, the worst response time _{T wr2} per one period of the task _{2, T wr2 = R wr2 × C} 2 = 0.6 × 3 = 1.8 ms. _Further , since the relationship between the activation cycle C _{0 of} task 0 and T _wr2 is “2 × Co> T _wr2 ”, the number of activations of task 0 per cycle of task 2 is two. _Further , since the relationship between the activation cycle C _{1 of} task 1 and T _wr2 is “C1> T _wr2 ”, the number of activations of task 1 per cycle of task 2 is one.

Also, the execution time T _e0 ′ of task 0 in the worst response time of task 2 is T _e0 ′ = R _wr0 × (2 × C ₀ ) = 0.6 × (2 × 1) = 1.2 ms. In addition, the execution time T _e1 ′ of task 1 in the worst response time of task 2 is T _e1 ′ = R _e1 × (1 × C ₁ ) = 0.15 × (1 × 2) = 0.3 ms.

Further, the execution time T _e2 ′ per cycle of the task 2 is T _e2 ′ = T _wr2 −T _e0 ′ −T _e1 ′ = 1.8−1.2−0.3 = 0.3 ms. Further, the execution time ratio R _e2 ′ per cycle of the task 2 is R _e2 ′ = T _e2 ′ / C ₂ = 0.3 ÷ 3 = 0.1.

  Next, a processor having the highest processing capability (performance) is selected from among the processors whose worst response time ratios do not exceed the set ratio for all the tasks assigned above. In addition, the execution time of all tasks assigned by the selected processor is calculated. As a procedure, first, for all tasks other than the lowest priority task, the number of activations within the execution permission time of the lowest priority task is calculated. Also, the execution time of tasks other than the lowest priority task is calculated based on the calculated number of activations.

Furthermore, the execution time in the new assignment destination processor is calculated for the task for changing the assignment destination (the task with the lowest priority among the tasks assigned to the processors exceeding the set ratio). In addition, when the execution time T _eOLD per one cycle of the task for changing the assignment destination obtained as described above, the processing capacity A _{OLD of the} currently assigned processor, and the processing ability A _NEW of the new assignment destination processor, the new assignment is obtained. The execution time T _eNEW in the previous processor becomes T _eNEW = T _eOLD × (A _NEW ÷ A _OLD ), and the execution time ratio R _NEW becomes R _eNEW = T _eNEW ÷ C from the start cycle C.

As a result, a new task assignment can be made based on the execution time _TNEW and the start cycle of the task to change the calculated assignment destination in the new assign destination processor and the execution time ratio and start cycle of all the tasks currently assigned to the processor. Estimate the worst response time ratio of each task in the performed state. Details of a method for estimating the worst response time ratio of each task in a state where task assignment is newly performed will be described later.

  Here, FIG. 13 is a diagram illustrating an example of task assignment after the worst response time ratio is estimated. After performing the above estimation, if one or more tasks exceeding the set ratio exist among newly assigned processors, as shown in FIG. 13A, the current task assignment is stopped and the next task assignment is stopped. Allocate a processor with high processing capacity to That is, the task (priority: maximum -K (K> N)) shown in FIG. 13A is assigned to the processor 11-2, but the task (priority: maximum -K) is assigned to the processor 11-2. Therefore, the task (priority: maximum -K) is assigned to the processor having the next highest processing capability (the processor 11-3 in FIG. 13A).

  If the task (priority: maximum -K) cannot be assigned to the processor with the next highest processing capability, and the subsequent processor 11-N cannot be assigned, the processor as shown in FIG. 13B. Reassign to 11-1 and change the prohibition flag to ON, and conditions such as until the task (priority: maximum -K) is completed or the worst response time ratio (estimated value) is within the set ratio To prevent other tasks from being assigned to the processor 11-1. Further, as described above, the response time ratio of each task for a predetermined time is measured. Here, if the worst response time ratio (estimated value) of each task is within the set ratio, the processor allocation destination for each reassigned task is stored and the response time ratio of each task for a predetermined time is measured, etc. The subsequent processing is performed.

<Estimation of worst response time ratio>
Here, a method for estimating the worst response time ratio of each task in a state where task assignment is newly performed will be described with reference to the drawings. FIG. 14 is a diagram for explaining an example of estimating the worst response time ratio. In the example of FIG. 14, first, a state (FIG. 14 (b)) in which a task C is newly assigned to the processor to which two tasks (task A and task B) are assigned in advance as shown in FIG. 14 (a) is shown. ing. Note that the three tasks shown in FIG. 14 are task A (activation cycle C _A = 1 ms, priority = high), task B (activation cycle C _B = 4 ms, priority = low), task C (activation cycle Cc = 3 ms, priority = medium). The execution time ratio of each task, the execution time ratio _{R eA} of the task A is 0.6, the execution time ratio _{R eB} task B is assumed to be 0.1. Further, it is assumed that the execution time ratio _ReC of task C is 0.15.

Since task A has the highest priority, the worst response time ratio R _wrA of task A is R _wrA ′ = R _eA = 0.6. Next, the free time T _fA per cycle of task A when only task A is executed is calculated. T _fA is T _fA = (1−R _wrA ′) × C _A = (1−0.6) × 1 = 0.4 ms.

Next, it is assumed that only task A and task C having the next highest priority are executed. In this assumption, regarded as the task C is always executable in T _fA. The execution time T _eC per cycle of the task C is calculated as T _eC = R _eC × Cc = 0.15 × 3 = 0.45 ms. Therefore, the ratio R _fA of T _fA in T _eC is R _fA = T _fA ÷ T _eC = 0.4 ÷ 0.45 = 0.89. That is, since 0.5 <R _fA <1, it can be seen that it is sufficient to have two cycles of task A to complete task C.

From the above-described relationship between the priorities of the task A and the task C, the waiting time until the completion of the execution of the task C is the execution time of the task A × 2. As a result, the worst response time T _wrC of task C becomes T _wrC = {(R _eA × C _A ) × 2} + T _eC = {(0.6 × 1) × 2} + 0.45 = 1.65. . Therefore, the worst response time ratio R _wrC of task C is R _wrC = T _wrC ÷ Cc = 1.65 ÷ 3 = 0.55.

Finally, let us consider a situation in which the remaining lowest priority task B is executed in addition to task A and task C. As described above, task A (startup request count = 2) and task C (startup request count = 1) are all completed and free time is available in the time of task A for two cycles (C _A × 2). Exists. At this time, the free time T _fAC is T _fAC = 2 × C _A −T _wrC = 2 × 1-1.65 = 0.35 ms. In addition, execution of task C is completed at time (C _A × 2) in the second cycle of task A. Therefore, it can be considered that all the idle time (T _fA = 0.4 ms) in the third period (time after C _A × 2) of task A can be allocated to the execution of task B. Furthermore, the execution time T _eB of task B per cycle of task B is T _eB = C _B × R _eB = 4 × 0.1 = 0.4 ms.

From the above-described contents, the relationship T _fAC <T _eB <(T _fAC + T _fA ) is established. Therefore, the worst response time T _wrB ′ of task B is T _wrB ′ = {(R _eA × C _A ) × 3} + T _eC + T _eB = {(0.6 × 1) × 3} + 0.45 + 0.4 = 2.65. Therefore, the worst response time ratio R _wrB ′ of task B is R _wrB ′ = T _wrB ′ / C _B = 2.65 ÷ 4 = 0.66.

  Here, FIG. 15 is a diagram illustrating an example of task reallocation. As shown in FIG. 15, the task (priority: maximum -K) set in the processor 11-1 is assigned to the processor 11-2 by performing the above-described calculation and the like. As described above, real-time performance in task execution can be improved by reassigning tasks.

  Further, the above-described assignment result is accumulated in the storage means or the like in order to validate the assignment contents even after the system is restarted. This reassigns tasks.

  As described above, according to the present invention, real-time performance in task execution can be improved. Specifically, in task execution scheduling using multiprocessors, task execution is completed within the execution permission time by using task priority, processor capacity, response time ratio, and worst response time ratio. The task can be reassigned so that the worst response time ratio does not exceed the set ratio.

  Note that the above-described task scheduling apparatus according to the present invention is applicable to a general machine control system, and is a large-scale parallel computer represented by a server or supercomputer from a small-scale system incorporated in a home appliance. Can be widely applied. Furthermore, the present invention is also effective when operating on a real-time system on a chip multiprocessor in which a plurality of arithmetic cores are provided in one CPU.

  The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to such specific embodiments, and various modifications, within the scope of the gist of the present invention described in the claims, It can be changed.

It is a figure which shows an example of schematic structure of a multiprocessor system. It is a figure which shows an example of the scheduling system of the conventional multiprocessor system. It is a figure of an example for demonstrating the execution permission time of a task. It is a figure which shows an example of a function structure of a task scheduling apparatus. It is a figure which shows an example of the hardware constitutions which can implement | achieve the task scheduling process in this invention. It is a flowchart of an example which shows the task scheduling processing procedure in this invention. It is a figure which shows an example of the setting content with respect to a correspondence table | surface of a processor and processing capability, and a task. It is a figure which shows an example of assignment of the task for every processor. It is a figure which shows an example of the allocation correspondence table | surface of task ID and a processor. It is a figure which shows an example of the measurement result of a response time ratio. It is a figure which shows an example of the result of the worst response time ratio of each task memorize | stored. It is a figure which shows an example of assignment of the task when the worst response time ratio exceeds a setting ratio. It is an example for demonstrating the mode of the allocation after estimating the worst response time ratio. It is a figure for demonstrating the example of estimation of the worst response time ratio. It is a figure which shows an example of reassignment of a task.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Multiprocessor system 11 Processor 12 Network 13, 21 Task scheduling apparatus 22 Processing capacity measurement means 23 Task allocation means 24 Response time ratio measurement means 25 Storage means 26 Control means 31 Input device 32 Output device 33 Drive device 34 Auxiliary storage device 35 Memory Device 36 CPU
37 Network connection device 38 Recording medium

Claims (10)

In a task scheduling method for assigning a plurality of tasks by a multiprocessor,
The priority of task execution preset for each of the plurality of tasks, the processing capability of each of the multiprocessors, the response time ratio for task execution performed within a predetermined time among the multiprocessors, and the response time ratio A task scheduling method, comprising: allocating the multiprocessor to the plurality of tasks based on a worst response time ratio obtained from a result measured for a predetermined number of times. In a task scheduling method for assigning a plurality of tasks by a multiprocessor,
A processing capability measuring step for measuring individual processing capabilities of a plurality of processors constituting the multiprocessor;
An assignment step for initial assignment to each task from a task priority set in advance for each of the plurality of tasks and a processing capability result obtained by the processing capability measurement step;
A response time ratio measuring step for executing a task based on a task assignment result obtained by the assigning step and measuring a response time ratio of the task at a predetermined time;
Worst response ratio selection step of measuring the response time ratio a predetermined number of times and selecting the worst response time ratio from a plurality of measured response time ratios;
A task scheduling method comprising: a reassignment step of reassigning a processor to a task based on the worst response time ratio and the set ratio of each task. The reassigning step includes
3. The task scheduling according to claim 2, wherein when performing the task reassignment, an estimation of a worst response time ratio in each task is performed, and the task is reassigned to a processor having the lowest worst response time ratio. Method. The reassigning step includes
4. The task scheduling method according to claim 2, wherein permission or prohibition of task reassignment is set for each of the plurality of processors. In a task scheduling device that assigns a plurality of tasks by a multiprocessor,
Processing capability measuring means for measuring individual processing capabilities of a plurality of processors constituting the multiprocessor;
Task allocation means for performing initial allocation to each task from the task priority set in advance for each of the plurality of tasks and the processing capability result obtained by the processing capability measuring unit;
A plurality of responses obtained by executing a task based on a task allocation result obtained by the task allocation means, measuring a response time ratio of the task at a predetermined time, and further measuring the measured response time ratio for a predetermined number of times. Response time ratio measuring means for selecting the worst response time ratio from the time ratio,
The task allocating means reallocates processors to tasks based on the worst response time ratio and setting ratio of each task. The task assignment means includes:
6. The task scheduling according to claim 5, wherein when performing the task reassignment, an estimation of a worst response time ratio in each task is performed, and the task is reassigned to a processor having the lowest worst response time ratio. apparatus. The task assignment means includes:
7. The task scheduling apparatus according to claim 5, wherein permission or prohibition of task reassignment is set for each of the plurality of processors. In a task scheduling program for causing a computer to execute task scheduling for assigning a plurality of tasks by a multiprocessor,
A processing capability measurement process for measuring individual processing capabilities of a plurality of processors constituting the multiprocessor;
A task priority set in advance for each of the plurality of tasks, and an allocation process for performing initial allocation to each task from the processing capability result obtained by the processing capability measurement process;
A response time ratio measurement process that executes a task based on the task assignment result obtained by the assignment process and measures a response time ratio of the task at a predetermined time;
Measuring the response time ratio a predetermined number of times, and selecting a worst response time ratio from a plurality of measured response time ratios;
A task scheduling program for causing a computer to execute a reassignment process for reassigning a processor to a task based on the worst response time ratio and the set ratio of each task. The reallocation process includes:
9. The task scheduling according to claim 8, wherein when performing the task reassignment, the worst response time ratio in each task is estimated, and the task is reassigned to a processor having the lowest worst response time ratio. program. The reallocation process includes:
10. The task scheduling program according to claim 8, wherein permission or prohibition of task reassignment is set for each of the plurality of processors.

Images