JP7478918B2 - Task intelligent processing method based on distributed heterogeneous system - Google Patents

Description

The present disclosure relates to the field of task intelligent processing technology, and in particular to a task intelligent processing method based on a distributed heterogeneous system.

The task processing problem in distributed heterogeneous systems is of great significance to the realization of the system's parallel performance and to maintaining load balance. Existing task processing methods usually assign tasks to threads directly or randomly, and the threads execute the assigned tasks.
However, when the above method is adopted, the following technical problems often arise.
First, it directly assigns tasks to specified threads without considering the load and running state of threads, which leads to unbalanced thread loads and low parallel efficiency of multiple threads.
Second, by assigning tasks to designated threads without dividing data processing tasks into ranks, tasks with lower ranks are executed earlier and tasks with higher ranks are executed later, thereby affecting the execution order of data processing tasks.
Third, if the number of data processing tasks matches the number of threads that can execute the tasks,
Data processing tasks are assigned directly or randomly to each thread, and it often happens that a thread with a high load executes a task that involves a large amount of data processing, which frequently causes thread clogging.
Fourth, when the number of data processing tasks is greater than the number of threads that can execute the tasks, the data processing tasks are assigned to each thread directly or randomly, without considering how to create new threads, which reduces the efficiency of task processing.
Fifth, when the number of data processing tasks is greater than the number of executable task threads, data processing tasks are assigned directly or randomly to each thread, which may result in long latency of the data processing tasks.

Some embodiments of the present disclosure address one or more of the technical problems mentioned in the Background section above.
In order to solve one or more of the above problems, a task processing method, device, apparatus, and medium based on a distributed isomerization system are proposed.

In a first aspect, some embodiments of the present disclosure provide a task intelligent processing method based on a distributed heterogeneous system including a task delivery end, a pre-trained priority determination model, and a thread pool, the thread pool including a predetermined number of threads, the method includes: controlling the task delivery end to perform data processing task delivery processing to obtain a set of data processing tasks; inputting the set of data processing tasks into the pre-trained priority determination model to obtain a priority set; obtaining initial load information and state information of each thread in the thread pool, obtaining an initial load information set and a state information set; assigning each data processing task in the set of data processing tasks to a thread in the thread pool based on the initial load information set, the state information set and the priority set, each thread in the thread pool is assigned one data processing task, controlling thread activation of the data processing tasks assigned to the thread pool, and executing the assigned data processing tasks.

In a second aspect, some embodiments of the present disclosure provide a task intelligent processing device based on a distributed heterogeneous system, including a task handover end, a pre-trained priority determination model, and a thread pool, the device including a first control unit arranged to control the task handover end to perform data processing task handover processing and obtain a set of data processing tasks, an input unit configured to input the set of data processing tasks to the pre-trained priority determination model to obtain a priority set, an acquisition unit configured to acquire initial load information and state information of each thread in the thread pool and obtain an initial load information set and a state information set, an allocation unit configured to allocate each data processing task in the set of data processing tasks to a thread in the thread pool based on the initial load information set, the state information set, and the priority set, and a second control unit configured to control thread activation of the data processing tasks allocated to the thread pool and execute the allocated data processing tasks.

In a third aspect, some embodiments of the present disclosure are a task handoff end, a pre-trained prioritization model, a thread pool, one or more processors, and a storage device having stored thereon one or more programs that are executed by the one or more processors and cause the one or more processors to implement a method according to any of the first aspects above.

In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having stored thereon a computer program which, when executed by a processor, implements a method according to any of the first aspects above.

The above-mentioned embodiments of the present disclosure have the beneficial effect of allocating data processing tasks by the task intelligent processing method based on the distributed heterogeneous system of some embodiments of the present disclosure, thereby realizing the effect of thread load balancing and improving the parallel efficiency of multiple threads. Specifically, the reason why the load of threads becomes unbalanced and the parallel efficiency of multiple threads decreases is that the existing data processing task allocation method does not consider the load or running state of the threads. This is because the task is directly allocated to the specified thread. Based on this, a task intelligent processing method based on the distributed heterogeneous system of some embodiments of the present disclosure is provided. First, the task passing end is controlled to perform data processing task passing processing, and a data processing task set can be obtained. Thus, the received data processing tasks can be divided into grades. Next, the data processing task set is input into the pre-trained priority determination model to obtain a priority set. Thus, the data processing tasks with higher grades can be executed first, so that the tasks with higher time requirements can be executed and completed first. Then, the initial load information and state information of each thread in the thread pool are obtained, and an initial load information set and a state information set can be obtained. Thus, the load situation of each thread can be determined, and thus,
A task with a high time requirement can be assigned to a thread with a low load, and a task with a low time requirement can be assigned to a thread with a high load, thus balancing the load among the threads. Then, based on the initial load information set, the state information set, and the priority set, each data processing task in the data processing task set is assigned to a thread in the thread pool, one data processing task per thread. The data processing tasks can be assigned to threads in the thread pool. This improves the execution efficiency of each task in each thread. Finally, the thread startup of the data processing tasks assigned to the thread pool can be controlled, and the assigned data processing tasks can be executed. This realizes thread load balancing and improves the parallel efficiency of multiple threads by dividing the data processing tasks into ranks and assigning each data processing task to each thread based on the priority of the data processing task and the load status and execution status of each thread.

1 is a schematic diagram of an application scenario of a task intelligent processing method based on a distributed heterogeneous system according to some embodiments of the present disclosure; FIG. 1 is a flowchart of some embodiments of a task intelligent processing method based on a distributed heterogeneous system according to the present disclosure. FIG. 1 is a block diagram of some embodiments of a task intelligent processing device based on a distributed isomerization system according to the present disclosure. FIG. 1 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure.

The present disclosure will now be described in detail with reference to the accompanying drawings and with reference to the embodiments.
FIG. 1 is a schematic diagram of an application scenario of a task intelligent processing method based on a distributed heterogeneous system according to some embodiments of the present disclosure.
In the application scenario of Fig. 1, first, the computing device 101 controls the task passing end 102 to perform data processing task passing processing, and obtains a data processing task set 103. Next, the computing device 101 inputs the data processing task set 103 into the pre-training priority determination model 104 to obtain a priority set 105, and obtains initial load information and state information of each thread in the thread pool to obtain an initial load information set 106 and a state information set 107. The initial load information set 106 and the state information set 107
and assigning each data processing task in the data processing task set 103 to a thread in the thread pool 108 based on the priority set 105;
Each thread in the thread pool 8 is assigned one data processing task.
8 to control thread initiation for the data processing task assigned thereto, and executes the assigned data processing task.
It should be noted that the computing device 101 may be hardware or software. If the computing device is hardware, it may be implemented as a distributed cluster of multiple servers or terminal devices, or as a single server or single terminal device. If the computing device is embodied as software, it may be installed on the hardware devices listed above. It may be implemented, for example, as multiple software or software modules for providing distributed services, or as a single software or software module. No specific limitations are provided here.
It should be understood that the number of computing devices in Figure 1 is merely approximate, and any number of computing devices may be included as required by the implementation.

2, a flow 200 of some embodiments of a task intelligent processing method based on a distributed heterogeneous system according to the present disclosure is shown. The task intelligent processing method based on a distributed heterogeneous system includes the following steps:
In step 201, a task passing end is controlled to perform data processing task passing processing, and a data processing task set is obtained.
In some embodiments, the execution subject of the task intelligent processing method based on a distributed heterogeneous system (the computing device 101 shown in FIG. 1 ) can control the task passing end to perform data processing task passing processing to obtain a data processing task set, where the task passing end receives a data processing task sent from a client, and the execution subject can obtain the data processing task received by the task passing end. The data processing tasks in the set of data processing tasks may be any tasks related to data processing.
As an example, the data processing task may include:
The above data processing task may be performed at several points in time depending on the train schedule, and tickets may be issued in installments.
In step 202, the data processing task set is input to a pre-trained prioritization model to obtain a priority set.
In some embodiments, the execution entity may input the set of data processing tasks into the pre-trained prioritization model to obtain a set of priorities.
Here, the pre-trained priority decision model is obtained by the following training.
A sample set including sample data processing tasks and corresponding priorities for the sample data processing tasks is obtained.
Based on the above sample set, the following training procedure is carried out.
Input the sample data processing tasks of at least one sample in the sample set into an initial neural network, respectively, and obtain a priority corresponding to each sample in the at least one sample. Determine a loss value of the priority corresponding to each sample in the at least one sample based on the priority corresponding to each sample in the at least one sample and the priority corresponding to each sample in the at least one sample. In response to determining that the loss value converges to a predetermined threshold, set the initial neural network as a training completion priority determination model. In response to determining that the loss value does not converge to a predetermined threshold, adjust network parameters of the initial neural network, construct a training sample set using unused samples, and perform the training step again with the adjusted initial neural network as the initial neural network. Here, the initial neural network may be various neural networks, such as convolutional neural networks, deep neural networks, etc., that process tasks based on data and obtain priorities for the tasks.
The above step 202 is one of the inventive features of the embodiment of the present disclosure, and addresses technical problem 2 mentioned in the Background Art, "by not dividing data processing tasks into ranks but directly assigning tasks to designated threads, tasks with lower ranks are executed first and tasks with higher ranks are executed later,
The present invention solves the problem of affecting the execution order of data processing tasks. The cause of affecting the execution order of data processing tasks is often that tasks with low ranks are executed first and tasks with high ranks are executed later by allocating tasks to a specified thread without dividing the data processing tasks into ranks. By solving such a cause, each data processing task can be executed in order of priority. To achieve this effect, a pre-trained priority decision model is introduced to determine the priority of each data processing task. A training-completed priority decision model is obtained by training an initial neural network. In this way, the ranks of data processing tasks can be divided by the training-completed priority decision model, and task execution can be performed according to the rank of the data processing task, and therefore each data processing task can complete task execution in order of priority.
In step 203, the initial load information and state information of each thread in the thread pool are obtained to obtain an initial load information set and a state information set.
In some embodiments, the execution entity may obtain initial load information and state information for each thread in the thread pool to obtain a set of initial load information and a set of state information.
Here, the thread pool may include a predetermined number of threads. The thread pool may be a thread pool having a thread creation function.
In step 204, each data processing task in the set of data processing tasks is assigned to a thread in the thread pool based on the set of initial load information, the set of status information, and the set of priorities.
In some embodiments, the execution subject is the initial load information set, the state information set,
and based on the priority set, each data processing task in the set of data processing tasks may be assigned to a thread in the thread pool, where one data processing task is assigned to each thread in the thread pool.
In some alternative embodiments of some embodiments, the execution entity assigns each data processing task in the set of data processing tasks to a thread in the thread pool based on the set of initial load information, the set of state information, and the set of priority, wherein state information in the set of state information includes a running state and a not-running state, and may include the following steps:
In a first step, a thread in the thread pool is identified as a first thread having state information of a running thread, and a first thread set is obtained, where the first thread in the first thread set does not have a data processing task assigned to it. The first thread in the first thread set may be a thread having thread state information of a running state.
In a second step, the other threads in the thread pool other than the first thread set are identified as second threads to obtain a second thread set. Here, a data processing task is assigned to the second thread in the second thread set. The second thread in the second thread set may be a thread whose thread state information is in an unexecuted state, and the data processing task can be assigned to the unexecuted thread.
Optionally, the execution entity assigns each data processing task in the set of data processing tasks to a thread in the thread pool based on the set of initial load information, the set of state information and the set of priority, wherein state information in the set of state information includes a running state and a not-running state, and may include the following steps:
In the first step, the number of second threads in the second thread set is determined to obtain the number of allocated threads. For example, the number of second threads in the second thread set may be 5, i.e., the number of allocated threads may be 5.
Secondly, determine the number of data processing tasks in the set of data processing tasks to obtain the number of data processing tasks. For example, the number of data processing tasks in the set of data processing tasks may be 4, i.e. the number of data processing tasks may be 4.
In step 3, it is determined whether the number of data processing tasks is equal to or greater than the number of threads to be assigned.
In step 4, in response to the number of said data processing tasks being equal to or less than the number of said assigned threads, the following task allocation steps are performed for said set of data processing tasks.
From the set of data processing tasks, a data processing task having the highest priority (e.g., the highest priority of grade 1) is selected as the data processing task to be assigned. Based on each initial load information corresponding to the second thread set, the execution time length on each second thread in the second thread set of the data processing task to be assigned is determined, and a set of execution time lengths is obtained. Here, the execution time is obtained by referring to the historical execution time in threads of various load situations of the data processing task to be assigned (e.g., for a certain data processing task to be assigned, if the load of the thread that acquires the data processing task to be assigned is case 1, the multitask execution time of the data processing task to be assigned is long, and the average execution time of the multiple tasks is obtained, and the execution time in the thread with the load of the data processing task to be assigned is case 1). The shortest execution time is narrowed down from the set of execution time lengths as the target execution time length. The second thread corresponding to the target execution time length is determined as the target thread. Here, the target thread may be a thread for executing the data processing task to be assigned. The data processing task awaiting assignment is deleted from the data processing task set, which assigns the data processing task to be assigned to the target thread, and it is determined whether the data processing task set after the task deletion is empty, and in response to the data processing task set after the task deletion being empty, execution of the task allocation step is terminated. In response to the data processing task set after the task deletion being not empty, execution of the task allocation step is continued, with the data processing task set after the task deletion being the data processing task set.
The above first to fourth steps solve technical problem 3 mentioned in the background art as one of the inventive features of the embodiment of the present disclosure: "When the number of data processing tasks matches the number of threads capable of executing tasks, data processing tasks are directly or randomly assigned to each thread, and threads with high loads frequently execute tasks with large amounts of data processing, which results in high frequency of thread clogging." The reason why thread clogging occurs frequently is often because, when the number of data processing tasks matches the number of threads capable of executing tasks, data processing tasks are directly or randomly assigned to each thread, and threads with high loads frequently execute tasks with large amounts of data processing. By solving the above cause, the effect of reducing the frequency of thread clogging can be obtained. To achieve this effect, first, the number of threads to be assigned is determined, then the number of data processing tasks is determined, and the magnitude relationship between the number of the data processing tasks and the number of the assigned threads is determined, and when the number of the data processing tasks is equal to or less than the number of the assigned threads, a task allocation step can be performed for the set of data processing tasks. As a result, when the number of data processing tasks matches the number of threads capable of executing the tasks, each data processing task in the set of data processing tasks can be assigned to each thread, allowing one data processing task to be executed by each thread, reducing the frequency with which a high-load thread executes a task with a large amount of data processing, and reducing the frequency with which thread clogging occurs.
Optionally, the execution entity assigns each data processing task in the set of data processing tasks to a thread in the thread pool based on the set of initial load information, the set of state information and the set of priority, wherein state information in the set of state information includes a running state and a not-running state, and may include the following steps:
First, in response to the number of data processing tasks being greater than the number of assigned threads, determining a difference between the number of data processing tasks and the number of assigned threads. For example, the difference may be two.
Second, the thread pool is controlled to create a number of threads equal to the delta value, for example, two new threads can be created.
Third, the data processing tasks in the set of data processing tasks are assigned to the second set of threads and the created threads in descending order of priority, where the number of data processing tasks in the set of data processing tasks is the same as the number of the second set of threads and the created threads, thereby allowing one data processing task to be assigned to one thread. Thus, implementation is made such that one data processing task is executed per thread.
The above first to third steps are one of the inventive features of the embodiment of the present disclosure, which addresses Technical Problem 4 mentioned in the Background Art: "When the number of data processing tasks is greater than the number of threads capable of executing the tasks, assign data processing tasks directly or randomly to each thread;
The problem of "Task processing efficiency is reduced without considering the method of creating new threads" has been solved. The reason for the decrease in task processing efficiency is often that when the number of data processing tasks is greater than the number of threads that can execute the tasks, data processing tasks are assigned directly or randomly to each thread without considering the method of creating new threads. By solving the above causes,
The effect of improving the efficiency of task processing can be realized. To achieve this effect, first, the number of data processing tasks is greater than the number of assigned threads, and a difference between the number of data processing tasks and the number of assigned threads can be determined. Next, the thread pool is controlled to create threads whose number is equal to the difference value. Finally, data processing tasks in the set of data processing tasks can be assigned to the second set of threads and the created threads in descending order of priority. In this way, by creating new threads,
Each data processing task can be executed on its own thread, improving the efficiency of task processing.
Optionally, the execution entity assigns each data processing task in the set of data processing tasks to a thread in the thread pool based on the set of initial load information, the set of state information and the set of priority, wherein state information in the set of state information includes a running state and a not-running state, and may include the following steps:
In a first step, in response to the number of data processing tasks being greater than the number of assigned threads, data processing tasks in the set of data processing tasks are assigned to the second set of threads in descending order of priority to obtain an assignment result, the assignment result including a set of unassigned data processing tasks. Because the number of data processing tasks is greater than the number of threads, data processing tasks with higher priority are executed first.
In step 2, the following task reassignment steps are performed on the set of unassigned data processing tasks:
a step of detecting whether execution of a second thread in the second thread set has been completed, and determining the second thread that has been completed as a thread to be reassigned in response to the existence of a second thread in the second thread set that has been completed; a step of removing an unassigned data processing task with the highest priority from the unassigned data processing task set, which is assigned to the thread waiting for reassignment, and determining whether the data processing task set after the unassigned task removal is empty, and in response to the data processing task set after the unassigned task removal being empty, execution of the task reassignment step is terminated; and in response to the data processing task set after the unassigned task removal being not empty, execution of the task reassignment step is continued with the data processing task set after the unassigned task removal as the unassigned data processing task set. As a result, after execution of a relatively high priority data processing task has been completed, this thread is in an unexecuted state, and the unassigned data processing task with the highest priority from the set of unassigned data processing tasks can be assigned to the thread, and therefore, rapid execution of data processing tasks can be realized without creating a new thread.
The above first and second steps solve technical problem 5 mentioned in the Background Art, which is one of the inventive features of the embodiment of the present disclosure: "When the number of data processing tasks is greater than the number of threads of executable tasks, the data processing tasks are directly or randomly assigned to each thread, and the waiting time of some of the data processing tasks is increased." Among the data processing tasks,
When the number of data processing tasks is greater than the number of executable task threads, the waiting time for directly or randomly allocating data processing tasks to each thread may become long. By solving such reasons, the effect of reducing the waiting time of data processing tasks can be obtained. To achieve this effect, first, when the number of data processing tasks is greater than the number of the allocated threads, the data processing tasks in the set of data processing tasks can be allocated to the set of second threads in descending order of priority. A task reallocation step is performed for the unallocated data processing tasks. As a result, after the execution of the relatively high priority data processing task is finished, this thread is in an unexecuted state,
The highest priority unassigned data processing task from the set of unassigned data processing tasks can be assigned to this thread, thus without creating a new thread.
It is possible to realize the effects of quickly executing data processing tasks and reducing the waiting time of the data processing tasks.
In step 205, thread activation of the data processing task assigned to the thread pool is controlled to execute the assigned data processing task.
In some embodiments, the execution entity is capable of controlling thread initiation for data processing tasks assigned to the thread pool, and executing the assigned data processing tasks.
Alternatively, the method may further include detecting an execution state of a thread in the thread pool when executing an assigned data processing task. The execution state includes normal execution and abnormal execution. Then, a task copy process is performed on the data processing task executed by the thread executing the abnormal execution state to obtain a replica data processing task. Then, it can be determined whether or not a thread in an unexecuted state exists in the thread pool. In response to the existence of an unexecuted thread in the thread pool, the replica data processing task is assigned to the unexecuted thread in the thread pool for execution. In response to the existence of no unexecuted thread in the thread pool, the thread pool creation thread is controlled to assign the replica data processing task to the created thread for execution.
Further referring to FIG. 3, as an implementation of the methods shown in the above-mentioned figures, the present disclosure provides several embodiments of a task intelligent processing device based on a distributed isomer system corresponding to the embodiment of the method shown in FIG. 2, which can be specifically applied to various electronic devices.

As shown in Fig. 3, a task intelligent processing device 300 based on a distributed heterogeneous system in some embodiments includes a first control unit 301, an input unit 302, an acquisition unit 303, an allocation unit 304, and a second control unit 305. Here, the first control unit 301 is configured to control the task passing end to perform data processing task passing processing and obtain a data processing task set, the input unit 302 is configured to input the data processing task set into the pre-trained priority determination model to obtain a priority set, the acquisition unit 303 is configured to acquire initial load information and state information of each thread in the thread pool and obtain an initial load information set and a state information set, the allocation unit 304 is configured to allocate each data processing task in the data processing task set to a thread in the thread pool based on the initial load information set, the state information set and the priority set, and each thread in the thread pool is assigned a data processing task, and the second control unit 305 is configured to control thread startup of the data processing tasks assigned to the thread pool and execute the assigned data processing tasks.
It will be understood that the units depicted in the apparatus 300 correspond to the steps of the method described with reference to Fig. 2. Thus, the operations, features and beneficial effects produced as described for the method described above apply equally to the apparatus 300 and the units contained therein and will not be described further herein.

Referring now to FIG. 4, an electronic device suitable for implementing some embodiments of the present disclosure (e.g.,
For example, a structural schematic diagram of the computing device 101) 400 of Figure 1 is shown. The electronic device shown in Figure 4 is merely an example and does not pose any limitation on the functionality and scope of use of the embodiments of the present disclosure.
As shown in FIG. 4, the electronic device 400 may include a processing device (e.g., a central processor, a graphics processor, etc.) 401 capable of performing various suitable operations and processes according to a program stored in a read-only memory (ROM) 402 or loaded from a storage device 408 into a random access memory (RAM) 403.
The RAM 403 also stores various programs and data necessary for the operation of the electronic device 400. The processing device 401, the ROM 402, and the RAM 403 are connected to one another via a bus 404. An input/output (I/O) interface 405 is also connected to the bus 404.
Generally, the I/O interface 405 includes input devices 406, which may include a touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.
The electronic device 400 includes an output device 407 including a liquid crystal display (LCD), a speaker, a vibrator, etc., a storage device 408 including a magnetic tape, a hard disk, etc., and a communication device 409. The communication device 409 can enable the electronic device 400 to communicate wirelessly or by wire to exchange data with other devices. Although FIG. 4 shows the electronic device 400 with various devices, it should be understood that it is not necessary to implement or include all of the devices shown. Alternatively, more or fewer devices can be implemented or included. Each block shown in FIG. 4 may represent one device or multiple devices as needed.
In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product including a computer program carried on a computer-readable medium including program code for performing the methods shown in the flowcharts. In some such embodiments, the computer program may be downloaded and installed from a network via the communication device 409, or may be installed from the storage device 408, or may be installed from the ROM 402. When the computer program is executed by the processing device 401, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

Claims (2)

A method for intelligent task processing based on a distributed heterogeneous system, comprising:
The distributed heterogeneous system includes a task handoff end, a pre-trained prioritization model, and a thread pool;
the thread pool includes a predetermined number of threads;
controlling the task transfer end to perform a data processing task transfer process and obtain a data processing task set;
inputting the set of data processing tasks into the pre-trained prioritization model to obtain a set of priorities;
obtaining initial load information and state information for each thread in the thread pool to obtain a set of initial load information and a set of state information;
assigning each data processing task in the set of data processing tasks to a thread in the thread pool based on the set of initial load information, the set of status information and the set of priorities, where each thread in the thread pool is assigned a data processing task;
controlling thread activation of data processing tasks assigned to the thread pool and executing the assigned data processing tasks;
A method for intelligently processing tasks based on a distributed heterogeneous system, comprising: A computer-readable medium having a computer program stored thereon,
A computer readable medium, the computer program product of which, when executed by a processor, implements the method of claim 1 .