An agent-based data processing computing power scheduling method

Abstract

The application relates to the technical field of computing power scheduling, and discloses a data processing computing power scheduling method based on an intelligent agent, which comprises the following steps: acquiring basic data and real-time running data of each computing power device in an operation cluster, and acquiring real-time characteristic elements of each computing power device according to the basic data and the real-time running data; identifying each task based on the intelligent agent, acquiring task characteristic elements and priorities, determining a processing sequence according to the queue time and the priorities of current tasks and assigned tasks; and performing coincident matching according to the task characteristic elements of each task and the real-time characteristic elements of each computing power device, and assigning corresponding computing power devices to each task according to the coincident matching result. The real-time characteristic elements of each computing power device are acquired according to the basic data and the real-time running data of each computing power device, so that the computing power tasks and the computing power devices are accurately matched, and the performance optimization and the cost optimization of the computing power scheduling are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computing power scheduling, in particular to a data processing computing power scheduling method based on an intelligent agent. BACKGROUND

[0002] With the wide use of large-scale computing platforms (such as cloud computing, big data, and AI training platforms), the demand for computing power is also increasing. In order to ensure stable supply of computing power, many task requirements generally use computing power clusters to achieve computing power support. In this process, balanced scheduling of computing power can intelligently allocate computing tasks to the most suitable nodes, thereby avoiding the "uneven busy and idle" state where some nodes are overloaded and other nodes are idle, and thus optimizing the utilization of entire cluster resources.

[0003] In the prior art, the computing power scheduling scheme is determined according to the load state of the computing power device. When the load of the device is low, the computing power task is preferentially allocated to the computing power device. When the load of the device is high, the allocation priority of the computing power device is low. In this way, the load conditions of all devices are relatively consistent, thereby achieving relative balance of all computing power devices.

[0004] Due to the differences between different task requirements and the performance gap between different computing power devices, the existing computing power scheduling method cannot optimize the performance of computing power scheduling. In addition, due to the use cost gap between different computing power devices, the economic optimization of computing power scheduling cannot be guaranteed. Therefore, how to further improve the performance optimization and cost optimization of computing power scheduling is the fundamental problem to be solved by the present application. SUMMARY

[0005] In order to further improve the performance optimization and cost optimization of computing power scheduling, the present application provides a data processing computing power scheduling method based on an intelligent agent.

[0006] A data processing computing power scheduling method based on an intelligent agent, the method comprising:

[0007] Obtaining basic data and real-time running data of each computing power device in the computing cluster, and obtaining real-time characteristic elements of each computing power device according to the basic data and real-time running data;

[0008] Identifying each task based on an intelligent agent, and obtaining task characteristic elements and priority;

[0009] Determining the processing order according to the queue time and priority of the current task and the allocated task, and performing coincidence matching according to the task characteristic elements of each task and the real-time characteristic elements of each computing power device, and allocating a corresponding computing power device to each task according to the coincidence matching result.

[0010] Furthermore, the real-time feature elements include a transmission efficiency coefficient, a computation cost coefficient, a computation processing coefficient, and an image processing coefficient;

[0011] The task feature elements include transmission requirement coefficient, prediction computation time, computation processing requirement coefficient, and image processing requirement coefficient.

[0012] The overlapping matching process includes:

[0013] Through formula Calculate the matching value between the task feature elements and the i-th computing device. ;

[0014] According to the matching value Select the top m computing power devices in descending order; use the formula... Calculate the matching balance value between the task feature elements and the i-th computing device. Assign each task to a matching balance value The largest computing power device;

[0015] Where j = 1, 2, 3; These are the transmission weight coefficient, the computational processing weight coefficient, and the graphics processing weight coefficient, respectively. These are the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient of the i-th computing device, respectively. These are the transmission demand coefficient, the computational processing demand coefficient, and the image processing demand coefficient, respectively. For the first defined function, when hour, ,otherwise, , , Three groups The mean, For correction factor, These are reference values ​​for the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient, respectively. Let be the unit time cost of the i-th computing device. The predicted computation time for the current task. Cost unit quantity This is the cost adjustment factor.

[0016] Furthermore, the calculation process for the transmission efficiency coefficient of each computing device includes:

[0017] Through the formula:

[0018]

[0019] Calculate the transmission efficiency coefficient of the i-th computing device. ;

[0020] in, Let be the average latency of the i-th computing device. As a function to compare the effects of delay, Let i be the allowed throughput of the i-th computing device. Let i be the workload of the i-th computing device. , Let be the throughput of the k-th task of the i-th computing device. for Contention for individual running tasks and synchronization waiting affect throughput. This is the throughput unit value. For weight adjustment coefficients, x = 1, 2, 3, 4; Let $i$ be the average storage write speed, average storage write speed, average memory write speed, and average memory write speed of the $i$-th computing device, respectively. These are the baseline values ​​for storage write speed, storage write speed, memory write speed, and memory write speed, respectively. For the second defined function, when hour, ,otherwise, , The weighting coefficients for storage write, storage write-out, memory write, and memory write-out are respectively.

[0021] Furthermore, the calculation process of the operational processing coefficients includes:

[0022] Through the formula:

[0023]

[0024] Calculate the computational processing coefficients of the i-th computing device. ;

[0025] in, Let i be the computing power value of the i-th computing device. This represents the proportion of computing power used by the i-th computing device. for The percentage of computing power used corresponds to the decrease in computing power.

[0026] Furthermore, the calculation process of the image processing coefficients includes:

[0027] Through the formula:

[0028]

[0029] Calculate the image processing coefficients of the i-th computing device. ;

[0030] in, Let be the graphics processing capability value of the i-th computing device. The percentage of graphics processing computing power used by the i-th computing device. for The percentage of graphics processing computing power used corresponds to the decrease in graphics processing capability.

[0031] Furthermore, the process of determining the processing order includes:

[0032] The current task is compared with the priority of the assigned tasks. If the priority of the current task is not higher than the priority of the last assigned task in the sorted list, the processing order of the current task is determined according to the time order. If the priority of the current task is higher than the priority of the last assigned task in the sorted list, the processing order of the current task is determined according to the priority difference and the queue time of the last assigned task in the sorted list.

[0033] Furthermore, the process of determining the processing order of the current task based on the priority difference and the queue time of the last task in the assigned tasks includes:

[0034] Through the formula:

[0035]

[0036] Calculate the adjustment coefficient E. When E≥0, determine the processing order of the current task according to the time sequence. When E<0, adjust the current task before the last task in the assigned tasks.

[0037] in, The queue time for sorting the last task in the assigned tasks. For correction factor, Sort the assigned tasks by priority, with O representing the current task priority. Priority The corresponding reference queue limit time.

[0038] Furthermore, the method also includes:

[0039] Each computing device is monitored for faults, and if the monitoring results are abnormal, no task scheduling is performed on that computing device.

[0040] In summary, this application includes at least one of the following beneficial technical effects:

[0041] This invention uses an intelligent agent to identify each task, judge the requirements of each task, obtain task feature elements and priorities, and adjust the task processing priorities reasonably based on the priority and the queue time of the current task and the assigned tasks. In addition, by obtaining the real-time feature elements of each computing power device through basic data and real-time operation data, it can achieve accurate matching of computing power tasks and computing power devices, thereby achieving performance and cost optimization of computing power scheduling. Attached Figure Description

[0042] Figure 1 This is a flowchart of the steps involved in a data processing computing power scheduling method based on intelligent agents. Detailed Implementation

[0043] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0044] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0045] This application discloses a data processing computing power scheduling method based on intelligent agents. Please refer to [link to relevant documentation]. Figure 1As shown, the method includes: acquiring basic data and real-time operating data of each computing device in the computing cluster; acquiring real-time feature elements of each computing device based on the basic data and real-time operating data; identifying each task based on an intelligent agent, acquiring task feature elements and priorities; determining the processing order based on the queue time and priorities of the current task and the assigned tasks; performing overlap matching between the task feature elements of each task and the real-time feature elements of each computing device; and assigning corresponding computing devices to each task based on the overlap matching results. In this process, by using an intelligent agent to identify each task, the required characteristics of each task can be judged, task feature elements and priorities can be acquired, and based on the priorities and the queue time of the current task and the assigned tasks, the processing order can be determined. This allows for a more reasonable adjustment of task processing priorities. Furthermore, this embodiment obtains real-time feature elements for each computing device based on its basic data and real-time operational data. This enables accurate matching of computing tasks with computing devices, improving overall task processing efficiency and effectiveness. The intelligent agent is trained based on task sample data from experience data. By labeling, it can determine the demand level of each task for each feature element. Quantification is performed based on the numerical range of each real-time feature element to obtain task feature elements. Through a process of matching task feature elements with the real-time feature elements of computing devices one by one, the accuracy of matching computing tasks with computing devices is improved, thereby achieving performance and cost optimization in computing power scheduling.

[0046] In one embodiment, the real-time feature elements include a transmission efficiency coefficient, a computation cost coefficient, a computation processing coefficient, and an image processing coefficient; the task feature elements include a transmission requirement coefficient, a prediction computation time, a computation processing requirement coefficient, and an image processing requirement coefficient; the overlap matching process includes:

[0047] Through formula Calculate the matching value between the task feature elements and the i-th computing device. Where j = 1, 2, 3; These are the transmission weight coefficient, the computational processing weight coefficient, and the graphics processing weight coefficient, respectively. The settings are all selected based on the occupancy levels of different needs in the empirical data; These are the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient of the i-th computing device, respectively. These are the transmission requirement coefficient, the computational processing requirement coefficient, and the image processing requirement coefficient, respectively, which are obtained based on the agent's recognition of the task. For the first defined function, when hour, ,otherwise, , , Three groups The mean, As a correction factor, it is based on the test data. The numerical range and degree of influence are set. These are reference values ​​for the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient, respectively. Used for Numerical normalization is performed, therefore The value is based on the test values. The critical value setting is achieved through matching values. The acquisition process, when the computing power device has a stronger capacity to accommodate tasks in various features (i.e. The higher the value, the more balanced the computing power usage of various features of the task becomes (i.e., The smaller the value, the higher the matching degree (i.e., the smaller the value). The higher the value, the better, therefore according to the matching value The top m computing devices are selected in descending order of power. The value of m is determined based on the scale of the computing devices in the cluster; the larger the scale, the larger the value of m can be. These top m devices are better positioned to process the task. Then, the process is analyzed using a formula. Calculate the matching balance value between the task feature elements and the i-th computing device. ,in, Let be the unit time cost of the i-th computing device. The predicted computation time for the current task is obtained based on the agent's prediction. The unit of cost is based on The unit is determined. This is a cost adjustment factor, which is set based on test data and the degree of cost control, thus matching the equilibrium value. The larger the value, the better the economy is when the task is assigned to the i-th computing device for processing, while ensuring reasonable scheduling of computing power. Therefore, each task is assigned to a matching balance value. The largest computing power equipment can thus improve the performance and cost optimization of computing power scheduling.

[0048] In one embodiment, a process for calculating the transmission efficiency coefficient of each computing device is provided, including: using the formula:

[0049] Calculate the transmission efficiency coefficient of the i-th computing device. ;in, Let be the average latency of the i-th computing device. This is a delay influence control function, which sets corresponding influence values ​​based on the range of different delays in the test data, fits the control relationship, and thus obtains the delay influence control function. It should be noted that the greater the delay, the better. The smaller the value, Let i be the allowed throughput of the i-th computing device. Let i be the workload of the i-th computing device. , Let be the throughput of the k-th task of the i-th computing device. for The impact of task contention and synchronization waiting on throughput is set based on the average throughput decrease of the computing power device under different task loads. This value can improve the accuracy of real-time throughput judgment for the computing power device. This is the throughput unit value, used to remove the dimension of throughput. The weighting adjustment coefficient is set based on the fit of the test data, therefore... It can evaluate the transmission efficiency of computing devices based on their latency and throughput. Additionally, x = 1, 2, 3, 4. Let $i$ be the average storage write speed, average storage write speed, average memory write speed, and average memory write speed of the $i$-th computing device, respectively. These are the baseline values ​​for storage write speed, storage write speed, memory write speed, and memory write speed, respectively. Obtained according to the setting standards of computing power equipment. For the second defined function, when hour, ,otherwise, , These are the weighting coefficients for storage write, storage write, memory write, and memory write, respectively. These coefficients are adaptively set based on the different storage and memory requirements in the test data, and therefore, are determined through the transmission efficiency coefficient. It can comprehensively assess the transmission efficiency of computing devices by considering key factors related to transmission.

[0050] In one embodiment, the calculation process of the operation processing coefficient includes: using the formula:

[0051]

[0052] Calculate the computational processing coefficients of the i-th computing device. ;in, Let be the computing power value of the i-th computing device, which is obtained based on the existing CPU performance scoring scheme. This represents the proportion of computing power used by the i-th computing device. for The processing power utilization ratio corresponds to the computing power attenuation value, which is obtained by comparing the average historical CPU operation data of the i-th computing device. Therefore, the processing coefficient is calculated. It can determine the real-time CPU performance status of the i-th computing device.

[0053] In one embodiment, the calculation process of image processing coefficients includes: using the formula:

[0054]

[0055] Calculate the image processing coefficients of the i-th computing device. ;in, Let be the graphics processing capability value of the i-th computing device, which is obtained according to the existing GPU performance scoring scheme. The percentage of graphics processing computing power used by the i-th computing device. for The percentage of graphics processing power utilization corresponds to a decrease in graphics processing capability, which is obtained by comparing the average historical operating data of the GPU of the i-th computing device. Therefore, it is determined through the image processing coefficient. It can determine the real-time GPU performance status of the i-th computing device.

[0056] In one embodiment, a process for determining the processing order is provided, including: comparing the priority of the current task with that of the assigned tasks; if the priority of the current task is not higher than the priority of the last assigned task in the sorted sequence, determining the processing order of the current task according to time order; if the priority of the current task is higher than the priority of the last assigned task in the sorted sequence, determining the processing order of the current task based on the priority difference and the queue time of the last assigned task in the sorted sequence, specifically including: using the formula:

[0057]

[0058] The adjustment factor E is calculated, where, The queue time for sorting the last task in the assigned tasks. The correction factor is selected and set based on empirical data. Sort the assigned tasks by priority, with O representing the current task priority. Priority Regarding the corresponding reference queue limit time, it should be noted that the number of priority levels set in this embodiment is greater than 3, and different priorities have corresponding reference queue limit times. The higher the priority, the shorter the corresponding reference queue limit time. Therefore, by adjusting the value of the coefficient E, the balance between queue time and priority can be determined. When E≥0, the processing order of the current task is determined according to the time order. When E<0, the current task is adjusted to be before the last task in the assigned tasks, thereby making the task processing order more reasonable.

[0059] In one embodiment, the data processing computing power scheduling method further includes: performing fault monitoring on each computing power device. Fault monitoring can be performed using a common solution in the prior art. When the monitoring results are abnormal, the computing power device is not scheduled for tasks. Through this process, the faulty computing power device can be prevented from affecting the computing power scheduling process.

[0060] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A data processing computing power scheduling method based on intelligent agents, characterized in that, The method includes: Obtain basic data and real-time operating data of each computing device in the computing cluster, and obtain real-time feature elements of each computing device based on the basic data and real-time operating data; The agent identifies each task and obtains its feature elements and priority. The processing order is determined based on the queue time and priority of the current task and the assigned tasks; the task feature elements of each task are matched with the real-time feature elements of each computing power device, and the corresponding computing power device is assigned to each task based on the matching result; The real-time feature elements include transmission efficiency coefficient, computation cost coefficient, computation processing coefficient, and image processing coefficient; The task feature elements include transmission requirement coefficient, prediction computation time, computation processing requirement coefficient, and image processing requirement coefficient. The overlapping matching process includes: Through formula Calculate the matching value between the task feature elements and the i-th computing device. ; According to the matching value Select the top m computing power devices in descending order; use the formula... Calculate the matching balance value between the task feature elements and the i-th computing device. Assign each task to a matching balance value The largest computing power device; Where j = 1, 2, 3; These are the transmission weight coefficient, the computational processing weight coefficient, and the graphics processing weight coefficient, respectively. These are the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient of the i-th computing device, respectively. These are the transmission demand coefficient, the computational processing demand coefficient, and the image processing demand coefficient, respectively. For the first defined function, when hour, ,otherwise, , , Three groups The mean, For correction factor, These are reference values ​​for the transmission efficiency coefficient, computational processing coefficient, and image processing coefficient, respectively. Let be the unit time cost of the i-th computing device. The predicted computation time for the current task. Cost unit quantity This is the cost adjustment factor.

2. The data processing computing power scheduling method based on intelligent agents according to claim 1, characterized in that, The calculation process for the transmission efficiency coefficient of each computing device includes: Through the formula: Calculate the transmission efficiency coefficient of the i-th computing device. ;in, Let be the average latency of the i-th computing device. As a function to compare the effects of delay, Let i be the allowed throughput of the i-th computing device. Let i be the workload of the i-th computing device. , Let be the throughput of the k-th task of the i-th computing device. for Contention for individual running tasks and synchronization waiting affect throughput. This is the throughput unit value. For weight adjustment coefficients, x = 1, 2, 3, 4; Let $i$ be the average storage write speed, average storage write speed, average memory write speed, and average memory write speed of the $i$-th computing device, respectively. These are the baseline values ​​for storage write speed, storage write speed, memory write speed, and memory write speed, respectively. For the second defined function, when hour, ,otherwise, , The weighting coefficients for storage write, storage write-out, memory write, and memory write-out are respectively.

3. The data processing computing power scheduling method based on intelligent agents according to claim 2, characterized in that, The calculation process of the operation processing coefficients includes: Through the formula: Calculate the computational processing coefficients of the i-th computing device. ; in, Let i be the computing power value of the i-th computing device. This represents the proportion of computing power used by the i-th computing device. for The percentage of computing power used corresponds to the decrease in computing power.

4. The data processing computing power scheduling method based on intelligent agents according to claim 3, characterized in that, The calculation process of the image processing coefficients includes: Through the formula: Calculate the image processing coefficients of the i-th computing device. ; in, Let be the graphics processing capability value of the i-th computing device. The percentage of graphics processing computing power used by the i-th computing device. for The percentage of graphics processing computing power used corresponds to the decrease in graphics processing capability.

5. The data processing computing power scheduling method based on intelligent agents according to claim 1, characterized in that, The process of determining the processing order includes: The current task is compared with the priority of the assigned tasks. If the priority of the current task is not higher than the priority of the last assigned task in the sorted list, the processing order of the current task is determined according to the time order. If the priority of the current task is higher than the priority of the last assigned task in the sorted list, the processing order of the current task is determined according to the priority difference and the queue time of the last assigned task in the sorted list.

6. The data processing computing power scheduling method based on intelligent agents according to claim 5, characterized in that, The process of determining the processing order of the current task based on the priority difference and the queue time of the last task in the assigned tasks includes: Through the formula: Calculate the adjustment coefficient E. When E≥0, determine the processing order of the current task according to the time sequence. When E<0, adjust the current task before the last task in the assigned tasks. in, The queue time for sorting the last task in the assigned tasks. For correction factor, Sort the assigned tasks by priority, with O representing the current task priority. Priority The corresponding reference queue limit time.

7. The data processing computing power scheduling method based on intelligent agents according to claim 1, characterized in that, The method further includes: Each computing device is monitored for faults, and if the monitoring results are abnormal, no task scheduling is performed on that computing device.

Images