KR101775029B1 - System and method of scheduling - Google Patents

Abstract

BACKGROUND ART [0002] Techniques for scheduling on a multi-core system shared by various applications are disclosed. According to an aspect of the present invention, it is possible to provide a technique capable of overlapping a search time of a dependency resolving operation and a work while a core is operating.

Description

[0001] The present invention relates to a scheduling system and a scheduling method for efficiently processing an application,

The present invention relates to a system having a storage unit (Queue or Dependency queue) in which a work having a dependency descriptor is placed in a multi-core system shared by various applications and a method of operating such a system More specifically, to provide a dependency queue for requesting a work to a system, and to allow an application to provide a work to be performed in a multicore device from a dependency queue to a core, And a scheduling system and method for shortening the time until a work is executed in a multicore device.

The basic operation of scheduling a system that is shared by various applications is to remove the dependency on all other work waiting for the completed work when the core is idle, Quot; Next, all the waiting work queues are searched for in each dependency queue while searching all the dependency queues. This ensures that all work dependencies are removed. If there is a work whose dependency has been removed, it requests execution of the work to the idle core.

During the process of doing dependency resolving for the completed work and searching for all the dependency queues and searching for the work whose dependency has been removed, the core becomes idle continuously and the efficiency of the whole system becomes low.

According to the scheduling operation of the system, it increases the operation efficiency with the external system and reduces the time during which the core stays idle.

When all the preceding work among the waiting work is completed, an efficient multi-core system is provided through the storage unit storing the work whose dependencies have been removed.

It is possible to operate the core without a delay in performing a task of searching for a dependency elimination task and a work having no dependency by providing a storage unit for storing a completed preceding work.

A scheduling system according to an aspect of the present invention is a scheduling system that receives a work from an application and allocates the work to be executed in a plurality of cores, the scheduling system comprising: An execution work storage unit for storing the completed work and a completed work storage unit for executing the completed work in the core.

In addition, a temporary storage unit for storing a plurality of works received from one or more applications before storing them in the execution work storage unit, and a work stored in the execution work storage unit are transmitted to the core or executed in the core, Lt; RTI ID = 0.0 > I / O < / RTI >

Also, it searches for a work having no dependency among at least one work stored in each temporary storage unit, stores the work in the execution work storage unit, and transmits the work stored in the execution work storage unit through the input / output port to be executed in the idle core And a scheduler that is executed in the core and receives the completed work through the input / output port and stores the completed work in the completed work storage unit.

A scheduling method according to an aspect of the present invention is a scheduling method for receiving a work from an application and allocating the work to be executed in a plurality of cores, the scheduling method comprising the steps of, in consideration of denpendency between works, And storing the completed work in the core.

According to another aspect of the present invention, there is provided a scheduling method including an execution work storage unit for storing at least one work received from an application in order to be executed in the core in consideration of the denpendency between the works, A scheduling method for assigning a plurality of applications through a scheduling system including a completed work storage unit to use and share multicore,

Determining whether or not a work stored in the execution work storage unit exists; if there is a work stored in the execution work storage unit, determining whether there is an idle core capable of executing the work; A step of transferring a work stored in the execution work storage unit to be executed in an idle core, a step of determining whether a work stored in the execution work storage unit does not exist or a core in an idle state does not exist, Determining whether there is a completed work in the finished work storage, and removing the dependency from the at least one work transferred from each application to the work having dependency in relation to the completed work.

According to the scheduling operation of the system, it is possible to improve the operation efficiency with the external system and reduce the time during which the core stays idle, thereby improving the efficiency of the system as a whole.

The performance of the multicore system can be improved by providing a storage unit for storing work in a state where all dependencies are removed.

The storage unit for storing the completed preceding work is provided so that the efficiency of the multi-core system can be improved by performing a task of searching for a dependency removing operation and a work having no dependency with no time delay even when the core is operating.

1 is a block diagram illustrating a scheduling system according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a multicore system to which a scheduling system according to an embodiment of the present invention is applied.
FIG. 3 is a diagram showing the dependency between work and work stored in the temporary storage unit dep-Q of the scheduling system according to an embodiment of the present invention.
4 is a diagram illustrating a process (I) of a work in a multicore system to which a scheduling system according to an embodiment of the present invention is applied.
FIG. 5 is a diagram illustrating a process (II) of a work in a multicore system to which a scheduling system according to another embodiment of the present invention is applied.
FIG. 6 is a diagram illustrating a work process (III) in a multi-core system to which a scheduling system according to another embodiment of the present invention is applied.
FIG. 7 is a flowchart illustrating an operation process (I) of a multicore system to which a scheduling system according to an embodiment of the present invention is applied.
8A and 8B are flowcharts illustrating an operation procedure (II) of a multicore system to which a scheduling system according to an embodiment of the present invention is applied.
9 is a flowchart detailing the step (C) shown in FIG. 8B according to an embodiment of the present invention.

Hereinafter, one aspect of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating a scheduling system according to an embodiment of the present invention. Referring to FIG. 1, a scheduling system that allocates a plurality of applications to use and share multicore includes a scheduler 100, an execution work storage unit (runnable-Q) 110, a finish work storage unit (finish- 120, a temporary storage unit (dep-Q) 130, and an input / output port (IO-port) 140.

The relationship between the scheduling system that receives a work from an application and allocates it to be executed in a plurality of cores is largely divided into two. First, an operation relationship between an application and a scheduling system. Second, an operation relationship between a scheduling system and a multicore device Can be divided.

The former operation enqueues an application to be executed in parallel in a multi-core device into a dependency queue of the scheduling system, and the latter operation relationship is a work relationship between the scheduling system and the work received from the applications, A work that has no dependency is requested to be executed by an idle core, and a core whose work has been executed can be represented by informing the scheduling system of the completion of the work.

The scheduling system operates in a device having a core separate from the multicore device. It also has an execution flow independent of the application. That is, both the operation of enqueuing a work in a dependency queue and the execution of a work requested in the multicore device can all operate in parallel with the scheduling system.

The execution work storage unit 110 stores one or more work received from an application in order to be executed in the core in consideration of dependency between the work.

The completed work storage unit 120 stores the completed work executed in the core.

The temporary storage unit 130 stores a plurality of works received from one or more applications for each application before storing them in the execution work storage unit 110. [

The input / output port 140 transfers the work stored in the execution work storage unit 110 to the core or executes the work in the core to receive the completed work from the core.

The scheduler 100 stores the completed work in the completed work storage unit 120 when the completed work is returned from the core, and if the work received from the application has a dependency in relation to the stored work, And stores it in the storage unit 110.

The scheduler 100 searches for works having no dependency among at least one work stored in each of the temporary storage units 130 and stores them in the execution work storage unit 110. When the work stored in the execution work storage unit 110 is idle Output port 140 so as to be executed in the core in the state where the completed work is stored in the completed work storage unit 120. The completion work is executed in the core and is received via the input / output port 140 and stored in the completed work storage unit 120.

In addition, the scheduler 100 removes the dependency on the work that is executed in the core among the works stored in the temporary storage unit 130 and has a dependency in relation to the completed work while the work is being executed in the core, 130) and stores the retrieved work in the execution work storage unit (110).

The scheduling system can operate the input work from various applications according to the described dependency order through such a device.

FIG. 2 is a diagram illustrating a multicore system to which a scheduling system according to an embodiment of the present invention is applied. 2, a scheduling system to which an aspect of the present invention is applied includes a temporary storage unit (dep-Q (1), dep-Q (N)) for storing a work delivered by an application, A scheduler, an execution work storage unit (runnable-Q), and a completion work storage unit (finish-Q) are disposed between the input / output ports IO ports output from the I /

The execution work storage unit (runnable-Q) collects the work in the state where the dependency is completely removed, and the completion work storage unit (finish-Q) collects the completed work.

When the work executed in the core of the multicore device (Device A or Device B) is terminated, the completed work is stored in the finished work storage unit (finish-Q) of the scheduling system (enqueue) (idle). Immediately thereafter, the scheduler of the scheduler system dequeues the waiting work in the execution work storage (runnable-Q) and immediately requests the work execution to the idle core of the multicore device.

The scheduler system executes the work in the core of the multicore device, performs a dependency resolving operation on all completed work in the finished work storage (finish-Q), and then executes the temporary storage (dep-Q) . All of the waiting work in all the temporary storages (dep-Q (1),..., Dep-Q (N)) are searched and all the work that has no dependency is moved to the run workable storage unit (runnable-Q).

By doing so, when a core of a multicore device becomes idle, a workpiece waiting in a runnable-Q can be processed without a delay time by a dependency resolving operation or a searching runnable works process It can immediately request the execution of the work in the idle core of the multicore device.

FIG. 3 is a diagram showing the dependency between work and work stored in the temporary storage unit dep-Q of the scheduling system according to an embodiment of the present invention. Referring to FIG. 3, each work stored in the temporary storage unit dep-Q includes eight workloads in two temporary storage units dep-Q (1) and dep-Q (2) .

W1, w2, w5, and w6 have no dependency on any work, so they can be executed immediately. W3 and w4 are related to w1 And w7 and w8 are executable states (states having dependencies in relation to w5) after w5 has been executed.

The process of the work in the multi-core system shown in FIG. 4 to FIG. 6 will be described based on the relationship between the works stored in the temporary storage units dep-Q1 and dep-Q2 shown in FIG.

4 is a diagram illustrating a process (I) of a work in a multicore system to which a scheduling system according to an embodiment of the present invention is applied.

4 illustrates a multicolor system having two cores, in which two applications each request a work using a dep-Q (dependency queue).

The application enqueues w1 and w2 to dep-Q1 and w5 and w6 to dep-Q2 and then enqueues w3 and w4 to dep-Q1 and w7 and w8 to dep-Q2. , and the initial state of the core (core1, core2) of the device is idle. And if you have a work that does not have both dep-Q dependencies, it is assumed that the chance of being dequeued from two dep-Q is the same.

In order to request the idle core of the multicore device to execute the work, there is a process of performing dependency resolving of the already completed work and a process of searching for the work that does not have dependency from all dep-Qs ) Should be preceded.

The scheduling system confirms that core1 of the multicore device is idle and performs searching runnable works on all dep-Qs. Through these operations, w1, w2, w5 and w6 are dequeued from each dep-Q and enqueued to runnable-Q. The scheduling system dequeues w1 from runnable-Q and requests core1 of multi-core device to execute w1.

At the end of the request, the scheduling system confirms that core2 of the multicore device is idle, dequeues w5 from runnable-Q, and requests core2 to execute w5. When two cores of the multicore device are in operation, the scheduling system confirms that the cores of the multicore devices are not all idle, and continues the dependency resolving operation and the searching runnable works operation sequentially in the order of finish-Q and dep- Try it.

When core1 of multicore device completes execution of w1, w1 completed in finish-Q is transmitted (enqueue) and is changed into idle state. Immediately, the scheduling system dequeues w2 from the runnable-Q and requests core1 of multicore device to execute w2. At the same time, when the core 2 of the multicore device completes the execution of the w 5, it transmits the completed w 5 to the finish-Q and switches to the idle state. The scheduling system confirms that core2 of the multicore device is idle, dequeues w6 from runnable-Q, and requests w2 to core2 of multicore device.

When both cores of multicore devices are in operation, the scheduling system continuously performs dependency resolving and searching runnable works on w1 and w5. In this process, w3, w4, w7, w8 are all transferred from each dep-Q to runnable-Q.

By repeating this process, it is possible to overlap the execution time of the work in the multicore device with the execution time of the dependency resolving operation and the search runnable works. In addition, it is possible to maintain the parallelism between the scheduling system and the external system by first requesting the work execution in the multicore device rather than the dependency resolving work related to the completed work with finish-Q.

FIG. 5 is a diagram illustrating a process (II) of a work in a multicore system to which a scheduling system according to another embodiment of the present invention is applied.

5 illustrates a multicolor system having two cores, in which two applications each request a work using a dep-Q (dependency queue).

The application enqueues w1 and w2 to dep-Q1 and w5 and w6 to dep-Q2 and then enqueues w3 and w4 to dep-Q1 and w7 and w8 to dep-Q2. , and the initial state of the core (core1, core2) of the device is idle. And if you have a work that does not have both dep-Q dependencies, it is assumed that the chance of being dequeued from two dep-Q is the same.

5, the scheduling system of FIG. 5 confirms that the core 1 of the multicore device is idle and performs a searching runnable works operation for all the dep-Qs in the same manner as the process (I) of FIG. 4 . Through these operations, w1, w2, w5 and w6 are dequeued from each dep-Q and enqueued to runnable-Q. The scheduling system dequeues w1 from runnable-Q and requests core1 of multi-core device to execute w1.

At the end of the request, the scheduling system determines that core 2 of the multicore device is idle and dequeues w5 from runnable-Q. Request core2 to run w5. When two cores of the multicore device are in operation, the scheduling system confirms that the cores of the multicore devices are not all idle, and continues the dependency resolving operation and the searching runnable works operation sequentially in the order of finish-Q and dep- Try it.

When core1 of multicore device completes execution of w1, w1 completed in finish-Q is transmitted (enqueue) and is changed into idle state. Immediately, the scheduling system dequeues w2 from the runnable-Q and requests core1 of multicore device to execute w2.

Core 1 of multicore device is running w2, and core2 is running w5 continuously. When core2 completes the execution of w5, it transmits w5 completed in finish-Q (enqueue) and goes into idle state. The scheduling system confirms that core2 of the multicore device is idle, dequeues w6 from runnable-Q, and requests w2 to core2 of multicore device.

When both cores of the multicore device are in operation, the scheduling system first performs the dependency resolving operation with respect to the completed w1, and when the completed w5 is transferred and stored in the finish-Q, w6 is executed, and both of the cores are running again. Then, the execution of the dependency resolving operation and the searching runnable works are sequentially performed with respect to the completed w1 and w5. In this process, w3, w4, w7, w8 are all transferred from each dep-Q to runnable-Q.

By repeating this process, it is possible to overlap the execution time of the work in the multicore device with the execution time of the dependency resolving operation and the search runnable works.

FIG. 6 is a diagram illustrating a work process (III) in a multi-core system to which a scheduling system according to another embodiment of the present invention is applied.

6 illustrates a multicolor system having three cores, in which two applications request a work using dep-Q (dependency queues), respectively.

The application enqueues w1 and w2 to dep-Q1 and w5 and w6 to dep-Q2 and then enqueues w3 and w4 to dep-Q1 and w7 and w8 to dep-Q2. , and the initial state of the core (core1, core2) of the device is idle. And if you have a work that does not have both dep-Q dependencies, it is assumed that the chance of being dequeued from two dep-Q is the same.

The scheduling system in the processing step (III) of FIG. 6 confirms that the core 1 of the multicore device is idle and performs a searching runnable works work on all the dep-Qs in the same way as the processing (I) of FIG. . Through these operations, w1, w2, w5 and w6 are dequeued from each dep-Q and enqueued to runnable-Q. The scheduling system dequeues w1 from runnable-Q and requests core1 of multi-core device to execute w1.

At the end of the request, the scheduling system determines that core 2 of the multicore device is idle and dequeues w5 from runnable-Q. Request core2 to run w5. Since the core 3 of the multicore device is in a state of executing w0 from the past, the three cores of the multicore device are in operation, so that the scheduling system recognizes that the cores of the multicore device are not all idle, And dep-Q in order to continue the dependency resolving and searching runnable works.

When core1 of multicore device completes execution of w1, w1 completed in finish-Q is transmitted (enqueue) and is changed into idle state. Immediately, the scheduling system dequeues w2 from the runnable-Q and requests core1 of multicore device to execute w2.

At the same time, when the core 3 of the multicore device completes the execution of the w0, it completes the completion of the finish-Q and energizes the idle state. The scheduling system confirms that the core 3 of the multicore device is idle, dequeues w6 from the runnable-Q, and requests the core 3 of the multicore device to execute w6. Therefore, w2 is executed in the core 1 of the multicore device, w5 is continuously executed in the core 2, and w6 is executed in the core 3.

When all three cores of the multicore device are in operation, the scheduling system continuously performs dependency resolving and searching runnable works on w0 and w1. In this process, w3, w4, w7, w8 are all transferred from each dep-Q to runnable-Q.

Through this process, when w2, w5, w6 completed from the three cores of the multicore device are enqueued in finish-Q, the scheduling system dequeues w3, w4, w7 stored in runnable-Q, . Again all cores are running, and the scheduling system performs dependency resolving and searching runnable works in succession with respect to w2, w5 and w6.

By repeating this process, it is possible to overlap the execution time of the work in a plurality of core devices with the execution time of the dependency resolving operation and the search runnable works.

FIG. 7 is a flowchart illustrating an operation process (I) of a multicore system to which a scheduling system according to an embodiment of the present invention is applied. Referring to FIG. 7, the operation process of the multicore system is basically composed of six steps.

First, one or more work delivered from the application is stored in order to be executed in the core in consideration of dependency between the work (700). In the stored order, the work is transmitted to the idle core on the system for execution (710), and the delivered work is executed on the core.

The work which has been executed in the core is stored (720), and the next work (the work stored in step 700) stored in the order to be executed in the core is executed and transmitted to the idle core (730).

In operation 720, a dependency resolving operation of a work having a dependency among the works received from the application in relation to the stored work is performed (740), and a work having no dependency among the works received from the application is executed And stores them in the order to be executed in the core (750).

If the work having no dependency is stored in step 750, the process is restarted from step 710. If there is no stored work, the work is terminated because there is no work to be executed as long as there is no new work transferred from the application.

8A and 8B are flowcharts illustrating an operation procedure (II) of a multicore system to which a scheduling system according to an embodiment of the present invention is applied.

In the overall operation, the runnable-Q is first searched to assign the work to the idle core quickly, the dependency resolving work related to the completed work is performed, Move the removed work to runnable-Q. By repeating this process, the multicore system is scheduled.

In more detail, when the system is started, the runnable-Q is searched (801). If there is work stored in the runnable-Q, it is checked whether there is an idle core in the multicore device (802). If there is an idle core, the work is dequeued in the runnable-Q (803), the work is requested to the core (804), and the scheduling system checks again whether the runnable-Q is empty (801).

If there is no idle core in multicore device or if runnable-Q is empty, the scheduling system searches for finish-Q to check completed work (805). If there is finished work in finish-Q, this work is dequeued in finish-Q (806) and dependency resolving work is performed for that work (807).

When the dependency resolving operation is completed, it is checked whether there is work completed in finish-Q (805). If finish-Q is empty, the scheduling system looks for dependency Qs and finds work without dependency (808). If there is work that does not have a dependency on any dependency Q, it dequeues the work in dependency Q (810) and enqueues it to runnable-Q (811).

The scheduling system may repeat the process (812) to operate the jobs entered from the various applications according to the described order of dependency. In addition, unlike a system that operates only with dependency Q, it has runnable-Q and finish-Q to reduce the synchronization overhead between the scheduling system and external systems. In addition, dependency resolving and dependency Q Can be performed in parallel with the external system.

9 is a flowchart detailing the step (C) shown in FIG. 8B according to an embodiment of the present invention.

In step 808 of FIG. 8, if there is work that has no dependency on the dependency Q, one of the dependency Q (first dependency Q) in the scheduling system is acquired (900) and the work stored in the acquired dependency Q is obtained 901). (902). If all the work has been performed, the work having no dependency is dequeued (903) and the work is enqueued to the runnable-Q (904).

In order to perform an enqueue to the runnable-Q for all the work of the first dependency Q, it is checked whether the work enqueued in the runnable-Q is the last work existing in the first dependency Q (905) Otherwise, the next work stored in the first dependency Q is acquired (906), and the acquired work is performed again from the step 902. [

If it is the last work, it is determined whether the first dependency Q is the last dependency Q existing in the scheduling system (907). If it is not the last dependency Q, the next dependency Q existing in the system is acquired (908) If it is the last dependency Q, it is determined whether the entire execution process of the system is to be performed again (812), and the process is repeated from step 801 shown in FIG. 8A.

8A and 8B or 9, the steps 805 to 812 or the steps 900 to 907 are performed while the work is being executed in the core by the state on the multi-core system in step 804 or in the series of processes shown in FIGS. , It is possible to perform work of dependency resolving of the system and work without dependency at the same time that the previous work is executed in the core, and the delay time of the scheduling system can be minimized.

Meanwhile, the embodiments of the present invention can be embodied as computer readable codes on a computer readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored.

Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device and the like, and also a carrier wave (for example, transmission via the Internet) . In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily deduced by programmers skilled in the art to which the present invention belongs.

The present invention has been described in detail by way of examples. The foregoing embodiments are intended to illustrate the present invention and the scope of the present invention is not limited to the specific embodiments.

100 Scheduler
110 Execution work storage (runnable-Q)
120 Completion work storage (finish-Q)
130 temporary storage unit (dep-Q)
140 I / O port (IO-port)

Claims (19)

A scheduling system for receiving a work from an application and allocating the work to be executed in a plurality of cores,
An execution work storage unit for storing at least one work transferred from the application in an order to be executed in the core in consideration of dependencies between the works;
A completion work storage unit for executing the work in the core and storing the completed work; And
Removing the dependency from a work that has been delivered from the application and has not yet been executed among the works having a dependency in relation to the work stored in the finished work storage unit while the works are being executed in the core, A plurality of applications including a scheduler for searching for a work that does not have a dependency among unexecuted works and storing the same in the execution work storage unit.
The apparatus of claim 1, wherein the scheduler
The work is stored in the completed work storage unit when the completed work is returned from the core and the completed work is returned, and when the work received from the application has a dependency in relation to the work stored in the completed work storage unit, And stores it in the execution work storage unit.
The method according to claim 1,
A temporary storage unit for storing a plurality of works received from at least one application for each application before storing the programs in the execution work storage unit; And
And an input / output port for transferring the work stored in the execution work storage unit to the core or executing the work in the core to receive the completed work from the core. The apparatus of claim 3, wherein the scheduler comprises:
Searching for works having no dependency among at least one work stored in each of the temporary storages, storing them in the execution work storage,
The work stored in the execution work storage unit is transmitted through the input / output port to be executed in the idle core,
And a completed work executed in the core is received through an input / output port and stored in the completed work storage unit.
5. The apparatus of claim 4, wherein the scheduler
Wherein the dependency is removed from a work stored in the temporary storage, the work being executed in the core and having a dependency in relation to a completed work. 5. The apparatus of claim 4, wherein the scheduler
The work being executed in the core to receive a completed work, and removing a dependence on the work stored in the temporary storage unit in relation to the received work in real time, or in association with the work stored in the completion work storage unit And removes the dependency of the work stored in the temporary storage unit. 5. The apparatus of claim 4, wherein the scheduler
And delivers the work stored in the execution work storage unit to be executed in the core that has delivered the completed work when the completed work is executed in the core. 6. The apparatus of claim 5, wherein the scheduler
The work being executed in the core among the works stored in the temporary storage unit to remove the dependency on a work having a dependency in relation to the completed work while the works are being executed in the core, And stores the retrieved work in the execution work storage unit. A scheduling method for allocating a work to a scheduling system to be executed in a plurality of cores,
Storing at least one work transferred from the application in order to be executed in the core in consideration of dependency between the works;
Executing on the core to store a completed work; And
Removing the dependency from a work that has been delivered from the application and has not yet been executed among those that have dependencies in relation to the completed and stored work while the works are running on the core, A plurality of applications including a step of searching for and storing works having no dependency in the work, and allocating the plurality of applications to share and use the multi-core.
10. The method of claim 9,
And removing the dependency when the work received from the application after the step of storing the completed work has a dependency in relation to the completed and stored work.
10. The method of claim 9,
Wherein the storing in the order to be executed in the core includes searching and storing the works having no dependency among one or more work delivered from each application; And
And delivering the stored works to the core for execution in an idle core.

10. The method of claim 9,
And delivering the stored work in the order to be executed, to be executed in the core that delivered the completed work, when receiving the completed work executed in the core after the step of storing the completed work.
delete A scheduling system including an execution work storage unit for storing at least one work received from an application in order to be executed in the core in consideration of the dependency between the works, and a completion work storage unit for storing the completed work executed by the core A scheduling method for assigning a plurality of applications to use and share multicore,
Determining whether a work stored in the execution work storage unit exists;
If there is a work stored in the execution work storage, determining whether there is an idle core capable of executing the work;
Transferring the work stored in the execution work storage unit to be executed in the idle core when the idle core exists;
Determining whether a completed work exists in the completed work storage unit if there is no work stored in the execution work storage unit or the idle core is not present; And
Removing the dependency for a work having a dependency in relation to the completed work among at least one work delivered from each application while the works are being executed in the core, Searching for a work that has not yet been executed and has no dependency from the work delivered from the application and storing the work in the execution work storage unit.
15. The method of claim 14,
Determining whether there is a work that does not have a dependency among one or more work delivered from each application if the completed work does not exist in the finished work storage; And
And storing the work having no dependency in the execution work storage if the work does not have the dependency. 16. The method of claim 15,
To the step of determining whether there is a work stored in the execution work storage unit after performing the step of delivering to be executed in the idle core, or when there is no work that does not have the dependency. 16. The method of claim 15,
If there is a work having no dependency, acquiring all the work delivered from the application that transferred the work;
Removing dependencies for the obtained works in relation to the work stored in the finished work storage; And
And storing the work whose dependency has been removed in the execution work storage unit. 16. The method of claim 15,
Wherein the step of determining whether there is a work stored in the execution work storage is performed during the execution of the work in the core through the step of delivering the work to be executed in the idle core. 18. The method of claim 17,
If there is a work that does not have the dependency, if there is more than one application that delivered the work, removing the dependency for all the works for each application and storing the dependency in the execution work storage A scheduling method to perform.

Images