A data processing system for allocating GPU

Abstract

The present invention relates to the field of GPU scheduling technology, and in particular to a data processing system for allocating a GPU, comprising: a data acquisition unit for performing video monitoring and acquisition on a monitoring area; a data analysis unit for periodically detecting the node usage status and determining a data analysis method according to the node usage status; a real-time processing unit for obtaining the operation routes of each device in the monitoring area corresponding to a type of node to determine an associated route and adjust the priority coefficient of the monitoring area corresponding to the associated route, and setting a secondary compensation condition for the priority coefficient of the monitoring area corresponding to the associated route and a secondary compensation method corresponding to the secondary compensation condition; a task scheduling unit for executing a task scheduling method in response to the task scheduling condition, and determining the scheduling task amount according to the estimated allowable coefficient of each schedulable node; the present invention improves the management efficiency of a GPU.

Description

Technical Field

[0001] The present invention relates to the technical field of GPU scheduling, and in particular to a data processing system for allocating GPUs. Background Art

[0002] GPU is a hardware specially designed for processing graphics calculations. With the development of technology, GPU is now also widely used in various computing tasks that require large-scale parallel processing capabilities, such as graphics rendering, image recognition, and visual monitoring. For smart parks, the control and scheduling of devices are often carried out by means of intelligent video recognition and monitoring. Among them, for the image processing part, the GPU allocation method has a great influence on the speed of data processing. However, the current data processing system for GPU allocation cannot be flexibly allocated according to the changes in the actual scene, resulting in low data processing efficiency. Therefore, how to allocate GPUs according to actual application scenarios to improve data processing efficiency is an urgent problem to be solved by technical personnel in this field.

[0003] Chinese patent publication number CN115145734A discloses a data processing system for allocating GPUs, which obtains a microservice information set, including the storage space occupied by the microservice and the target GPU list, and thereby obtains a GPU scheduling priority index list, and allocates the GPU according to the priority index. In the process of obtaining the target GPU scheduling priority index list, the invention takes into account the average utilization value of the GPU, thereby ensuring the computing speed of the microservice and improving the operating efficiency of the microservice. However, the above technical solution has the following problems: in view of the real-time changes in the data source in the actual application scenario, it is impossible to select different data analysis methods according to the actual situation, resulting in the data analysis method not being able to meet the actual use scenario, thereby making the GPU allocation management efficiency low. Summary of the invention

[0004] To this end, the present invention provides a data processing system for allocating GPUs to overcome the problem in the prior art that, in response to real-time changes in data sources in actual application scenarios, different data analysis methods cannot be selected according to actual conditions, resulting in data analysis methods not being able to conform to actual usage scenarios, thereby causing the GPU allocation management efficiency to be low.

[0005] To achieve the above object, the present invention provides a data processing system for allocating a GPU, comprising:

[0006] A data acquisition unit, used for video surveillance and acquisition of the monitoring area;

[0007] A data analysis unit connected to the data acquisition unit, for periodically detecting the node usage status and determining a data analysis method according to the node usage status, determining a real-time processing method according to the device status in the monitoring area corresponding to a type of node, and adjusting the monitoring period according to the quantity change status;

[0008] A real-time processing unit connected to the data analysis unit is used to obtain the operation routes of each device in the monitoring area corresponding to a type of node to determine the associated route and adjust the priority coefficient of the monitoring area corresponding to the associated route, and set a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition for the priority coefficient of the monitoring area corresponding to the associated route;

[0009] The task scheduling unit is connected to the data analysis unit and is used to execute the task scheduling method in response to the task scheduling conditions and determine the scheduling task amount according to the estimated allowable coefficient of each schedulable node.

[0010] Furthermore, the data analysis unit periodically detects the node usage status and determines the data analysis method according to the node usage status. The data analysis method includes performing task processing on a type of node, or determining whether to adjust the monitoring period according to the change in the number of a type of node.

[0011] Furthermore, the data analysis unit determines the node usage status according to the proportion of a type of nodes, and different node usage statuses correspond to different proportions of a type of nodes.

[0012] Furthermore, the data analysis unit detects the device status in the monitoring area corresponding to a type of node, and the device status is determined according to the device blocking time and the number of devices.

[0013] Furthermore, the data analysis unit determines a real-time processing method according to the device status in the monitoring area corresponding to a type of node. The real-time processing method includes adjusting the priority coefficient of the monitoring area corresponding to the associated route, or scheduling and allocating the request task corresponding to the monitoring area.

[0014] Furthermore, the real-time processing unit obtains the operation routes of each device in the monitoring area corresponding to a type of node to obtain a shared route, records the shared route with the most corresponding shared devices as an associated route, and increases and adjusts the priority coefficient of the monitoring area corresponding to the associated route;

[0015] The increase value of the priority coefficient of the corresponding monitoring area of ​​the associated route is positively correlated with the blocking time of the device.

[0016] Furthermore, the real-time processing unit is provided with a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition, wherein the secondary compensation method is to perform secondary increase adjustment on the priority coefficient of the monitoring area whose impact evaluation value in the associated route is greater than the preset impact evaluation value;

[0017] The secondary compensation condition is that the number of nodes of the first category corresponding to the associated route is greater than the preset number of nodes of the first category associated.

[0018] Furthermore, the task scheduling unit is provided with task scheduling conditions and task scheduling methods corresponding to the task scheduling conditions, and the task scheduling method is to detect the estimated allowable coefficient of the schedulable node;

[0019] The estimated allowable coefficient is related to the estimated number of personnel and the estimated number of devices in the monitoring area corresponding to the dispatchable node;

[0020] The task scheduling condition is that there are schedulable nodes.

[0021] Further, the task scheduling unit determines the scheduling task amount according to the estimated allowable coefficient of each schedulable node;

[0022] The scheduling task volume is positively correlated with the estimated allowable coefficient.

[0023] Further, the data analysis unit monitors the quantity change state of a type of nodes and reduces the monitoring period when the quantity change state is an increasing state.

[0024] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical solution of the present invention, the node usage status is determined according to the proportion of a type of nodes, the busyness of the current image processing task is reflected by the node usage status, and different data analysis methods are selected according to different node usage statuses, so that the determination of the data analysis method is more in line with the actual application scenario, thereby improving the GPU allocation management efficiency of the present invention.

[0025] Furthermore, the data analysis unit in the present invention determines the real-time processing mode according to the device status in the monitoring area corresponding to a type of node, wherein the device status is determined by the device congestion duration and the number of devices, accurately characterizing the operation busyness and operation congestion degree of the devices in the monitoring area, and correspondingly selecting the real-time processing mode, so that the priority coefficient or task allocation determined by the real-time processing mode can be more effective and accurate, thereby improving the GPU allocation management efficiency of the present invention.

[0026] Furthermore, when adjusting the priority coefficient of the monitoring area corresponding to the associated route in the present invention, the co-occupied route with the maximum probability of congestion is selected as the associated route, and the priority coefficient of the monitoring area corresponding to the associated route is adjusted, so that the priority of subsequent image processing is improved, thereby reducing the degree of congestion in the congested area, and further reducing the difficulty of subsequent GPU allocation management from a global perspective.

[0027] Furthermore, the real-time processing unit in the present invention is provided with a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition. In order to avoid the problem of poor efficiency in adjusting the priority coefficient due to too many nodes of a type in the associated route, the present invention selects part of the monitoring area of ​​the associated route in combination with a preset impact evaluation value to make a secondary adjustment of the priority coefficient, thereby improving the setting efficiency of the priority coefficient, further optimizing the priority coefficient of the present invention, and thereby improving the processing speed of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A unit connection diagram of a data processing system for which a GPU is allocated according to the present invention;

[0029] Figure 2 This is a flow chart of the present invention for determining a data analysis method according to a node usage status;

[0030] Figure 3 This is a flow chart of an embodiment of the present invention that determines a real-time processing method according to the device status in a monitoring area corresponding to a type of node. DETAILED DESCRIPTION

[0031] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0033] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0034] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0035] See also Figures 1 to 3 As shown, the present invention provides a data processing system for allocating a GPU, comprising:

[0036] A data collection unit, which includes several monitoring devices for performing video monitoring and collection on the monitoring area, and is used to perform video monitoring and collection on the monitoring area;

[0037] A data analysis unit connected to the data acquisition unit, for periodically detecting the node usage status and determining a data analysis method according to the node usage status, determining a real-time processing method according to the device status in the monitoring area corresponding to a type of node, and adjusting the monitoring period according to the quantity change status;

[0038] A real-time processing unit connected to the data analysis unit is used to obtain the operation routes of each device in the monitoring area corresponding to a type of node to determine the associated route and adjust the priority coefficient of the monitoring area corresponding to the associated route, and set a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition for the priority coefficient of the monitoring area corresponding to the associated route;

[0039] The task scheduling unit is connected to the data analysis unit and is used to execute the task scheduling method in response to the task scheduling conditions and determine the scheduling task amount according to the estimated allowable coefficient of each schedulable node.

[0040] The application scenario of the present invention is the analysis and processing of monitoring images of a smart park. The position distribution of the monitoring device of the video acquisition unit is set according to the actual use scenario. The monitoring device is not specifically limited, but it is required to have video acquisition and video communication transmission functions. In the present invention, a single node corresponds to the monitoring devices of at least two monitoring areas, which are used to process and analyze the monitoring videos collected by the monitoring devices of the monitoring areas. There are several unmanned working devices in the smart park. The working devices move along the working route according to the tasks sent by the user. The working devices include but are not limited to mobile robots, unmanned controlled vehicles and automatic inspection robots. After the monitoring device identifies the working device, it sends a request task to the node corresponding to the monitoring area. The request task is the video data collected by the monitoring device. The node is used to analyze the video data with a preset function. The preset function is set by the user. For example, the user needs to perform image analysis on the wear of the working device, the safety of the goods, and the failure of the working device in the video data. The setting positions of the monitoring devices are all movable roads in the smart park. The movable roads are roads that allow the working device to move, that is, roads that can be used as working routes. However, the specific setting number and the setting distance of adjacent monitoring devices are not specifically limited. Adaptive settings are made according to the actual application scenario. This is easy for technicians in this field to understand and will not be repeated here.

[0041] Specifically, the data analysis unit periodically detects the node usage status and determines the data analysis method according to the node usage status. The data analysis method includes performing task processing on a type of node, or determining whether to adjust the monitoring period according to the change in the number of a type of node.

[0042] Specifically, the data analysis unit determines the node usage status according to the proportion of a type of nodes, and different node usage statuses correspond to different proportions of a type of nodes;

[0043] Among them, one type of node is a node whose current resource usage ratio is greater than the preset resource usage ratio. For a single node, its corresponding resource usage ratio = the amount of data to be processed by the node / the maximum amount of data that the node can process. For the numerical setting of the preset resource usage ratio, the user can set it according to the actual scenario. It can be understood that due to the different data processing capabilities of each node in the actual usage scenario, the setting standards for the preset amount of video data to be processed are different. A setting method for the preset resource usage ratio is provided, and the preset resource usage ratio is 50% of the maximum amount of data that the node can process;

[0044] The node usage status includes a first preset node usage status and a second preset node usage status, the first preset node usage status is that the proportion of a class of nodes is greater than the preset class proportion, and the second preset node usage status is that the proportion of a class of nodes is less than or equal to the preset class proportion;

[0045] When the first preset node is in use, task processing is performed on a type of node;

[0046] When the second preset node is in use, it is determined whether to adjust the monitoring period according to the quantity change state of a type of nodes.

[0047] The proportion of a class of nodes is the number of a class of nodes / the number of all nodes, where all nodes include a class of nodes and other nodes that are not a class of nodes. The user can preset the value of the class of nodes proportion based on the actual application scenario. The greater the user's demand for data processing stability, the smaller the value of the preset class of nodes proportion. A method for setting a preset class of nodes proportion is provided. The preset class of nodes proportion is the average value of the class of nodes proportion corresponding to the records in the historical records whose data processing efficiency meets the user's needs. A value of the preset class of nodes proportion is provided, and the preset class of nodes proportion is 60%.

[0048] Specifically, the data analysis unit detects the device status in the monitoring area corresponding to a type of node, the device status is determined according to the device blocking time and the number of devices, and the device status includes a first device status, a second device status and a third device status;

[0049] The first device state is that the number of devices is greater than or equal to the preset number of devices and the device blocking time is greater than the preset device blocking time;

[0050] The second device state is that the number of devices is greater than or equal to the preset number of devices and the device blocking time is less than or equal to the preset device blocking time;

[0051] The third device state is that the number of devices is less than the preset number of devices. When in the third device state, the data analysis system sends a reminder message to the user to remind the user to perform manual and autonomous optimization on the node to improve the performance of the node itself;

[0052] Among them, for a single type-one node, the number of devices corresponding to the type-one node is the number of working devices that appear in a single traffic cycle in the monitoring area corresponding to the type-one node, and the device congestion duration is the appearance duration corresponding to the working device with the longest appearance duration in a single traffic cycle in the monitoring area. For the numerical settings of the preset number of devices and the preset device congestion duration, the user can set them according to the actual usage scenario. The smaller the number of devices that can be accommodated in the monitoring area corresponding to the type-one node, the smaller the numerical setting of the preset number of devices. A method for determining the preset number of devices is provided, and the preset number of devices is 50% of the number of devices that can be accommodated in the monitoring area; the device congestion duration reflects the difficulty of vehicle passage in the corresponding monitoring area. It can be understood that the slower the flow of the device, the more likely it is that accidents such as device collision will occur. The greater the monitoring demand of the monitoring area, the longer the continuous monitoring time of the monitoring device, which results in the monitoring device in the monitoring area needing to process more video information than the monitoring device corresponding to the monitoring area with strong device mobility and a small number. For the value setting of the preset device congestion duration, the user can record the average value of the congestion duration corresponding to the records that meet the user's transportation needs in the historical work records as the preset congestion duration.

[0053] Specifically, the data analysis unit determines a real-time processing method according to the device status in the monitoring area corresponding to a type of node, and the real-time processing method includes adjusting the priority coefficient of the monitoring area associated with the route, or scheduling and allocating the request task corresponding to the monitoring area;

[0054] Among them, when the device state is the first device state, the real-time processing mode selects to adjust the priority coefficient of the monitoring area corresponding to the associated route, and the node preferentially processes the video data of the request task of the monitoring area with a larger priority coefficient, and the initial value of the priority coefficient of each monitoring area is the same;

[0055] When the device state is in the second device state, the real-time processing mode selects scheduling and allocating the request task corresponding to the monitoring area.

[0056] Specifically, the real-time processing unit obtains the operation routes of each device in the monitoring area corresponding to a type of node to obtain a shared route, records the shared route with the most corresponding shared devices as an associated route, and increases and adjusts the priority coefficient of the monitoring area corresponding to the associated route;

[0057] The increase value of the priority coefficient of the monitoring area corresponding to the associated route is positively correlated with the blocking time of the device;

[0058] Among them, the shared route is the overlapping route section of the operation route of each device, and the shared device is the working device corresponding to the shared route, that is, there must be some sections of the working route of the shared device within the shared route. The acquisition of the position, working route and specifications of the working device in the present invention is not specifically limited. This is a conventional technical means mastered by technical personnel in this field. A method for obtaining information of a working device is provided, which can store the specifications of each working device and the current working route through a cloud platform, and obtain the position of the working device through a visual sensor.

[0059] Specifically, the real-time processing unit is provided with a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition, wherein the secondary compensation method is to perform a secondary increase adjustment on the priority coefficient of the monitoring area whose impact evaluation value in the associated route is greater than the preset impact evaluation value;

[0060] The secondary compensation condition is that the number of nodes of the first category corresponding to the associated route is greater than the preset number of nodes of the first category associated;

[0061] The quadratic increase value of the priority coefficient of the monitoring area is positively correlated with the impact assessment value;

[0062] A method for confirming the impact evaluation value of a single monitoring area is to record the monitoring area as a target area, obtain a monitoring area closest to the target area, and record the number of working devices in the monitoring area as the impact evaluation value.

[0063] The preset impact assessment value is the average of the impact assessment values ​​of each monitoring area in the associated route.

[0064] Specifically, the task scheduling unit is provided with task scheduling conditions and task scheduling methods corresponding to the task scheduling conditions, and the task scheduling method is to detect the estimated allowable coefficient of the schedulable node;

[0065] The estimated allowable coefficient is related to the estimated number of personnel and the estimated number of devices in the monitoring area corresponding to the dispatchable node;

[0066] The task scheduling condition is that there are schedulable nodes.

[0067] Specifically, the task scheduling unit determines the scheduling task amount according to the estimated allowable coefficient of each schedulable node;

[0068] The scheduling task volume is positively correlated with the estimated allowable coefficient, and the scheduling task volume is the data volume of the request task that can be scheduled to the schedulable node;

[0069] Wherein, the schedulable node is a non-class I node, and the estimated allowable coefficient is Y. , A1 is the number of personnel in the monitoring area of ​​the dispatchable node, A2 is the number of working devices in the monitoring area of ​​the dispatchable node, B iFor the numerical settings of the i-th weight coefficient, B1 and B2, users can determine the impact of A1 and A2 on the operation of devices in the smart park based on historical experience. The greater the impact, the larger the corresponding weight coefficient value.

[0070] Specifically, the data analysis unit monitors the quantity change state of a type of nodes and reduces the monitoring period when the quantity change state is an increasing state;

[0071] The data analysis unit of the present invention is provided with a cyclically running monitoring cycle, and a determination of the data analysis mode is performed once at the end of each monitoring cycle.

[0072] Among them, the data analysis unit has a monitoring cycle. At the end of a cycle, the number of a class of nodes in the most recent monitoring cycle minus the number of a class of nodes in the current monitoring cycle is calculated, and the value is recorded as the number change value of a class of nodes. When the numerical change value is greater than the preset number change value, the number change state is determined to be an increasing state. The reduction value of the monitoring cycle is positively correlated with the number change value. The value of the preset number change value can be set by the user according to actual needs. The greater the user's demand for the efficiency of park video data processing, the smaller the value of the preset number change value. A preset number change value is provided, and the preset number change value is 5.

[0073] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data processing system for allocating a GPU, characterized in that: include: A data acquisition unit, used for video surveillance and acquisition of the monitoring area; A data analysis unit connected to the data acquisition unit, for periodically detecting the node usage status and determining a data analysis method according to the node usage status, determining a real-time processing method according to the device status in the monitoring area corresponding to a type of node, and adjusting the monitoring period according to the quantity change status; A real-time processing unit connected to the data analysis unit is used to obtain the operation routes of each device in the monitoring area corresponding to a type of node to determine the associated route and adjust the priority coefficient of the monitoring area corresponding to the associated route, and set a secondary compensation condition and a secondary compensation method corresponding to the secondary compensation condition for the priority coefficient of the monitoring area corresponding to the associated route; A task scheduling unit connected to the data analysis unit, for executing a task scheduling method in response to a task scheduling condition, and determining a scheduling task amount according to an estimated allowable coefficient of each schedulable node; When the first preset node is in use, task processing is performed on a type of node; In the second preset node use state, determining whether to adjust the monitoring cycle according to the quantity change state of a type of nodes; When the device state is the first device state, the real-time processing mode selects to adjust the priority coefficient of the monitoring area corresponding to the associated route, and the node preferentially processes the video data of the request task of the monitoring area with a larger priority coefficient, and the initial value of the priority coefficient of each monitoring area is the same; When the device state is in the second device state, the real-time processing mode selects scheduling and allocating the request task corresponding to the monitoring area; The secondary compensation method is to increase the priority coefficient of the monitoring area whose impact evaluation value in the associated route is greater than the preset impact evaluation value for a secondary increase; The secondary compensation condition is that the number of nodes of the first category corresponding to the associated route is greater than the preset number of nodes of the first category associated; The quadratic increase value of the priority coefficient of the monitoring area is positively correlated with the impact assessment value; The scheduling task volume is positively correlated with the estimated allowable coefficient, and the scheduling task volume is the data volume of the request task that can be scheduled to the schedulable node.

2. The data processing system for allocating GPUs according to claim 1, characterized in that: The data analysis unit determines the node usage status according to the proportion of a type of nodes, and different node usage statuses correspond to different proportions of a type of nodes.

3. The data processing system for allocating GPUs according to claim 2, characterized in that: The data analysis unit detects the device status in the monitoring area corresponding to a type of node, and the device status is determined according to the device blocking time and the number of devices.

4. The data processing system for allocating GPUs according to claim 1, characterized in that: The real-time processing unit obtains the running routes of each device in the monitoring area corresponding to a type of node to obtain a shared route, records the shared route with the most corresponding shared devices as an associated route, and increases and adjusts the priority coefficient of the monitoring area corresponding to the associated route; The increase value of the priority coefficient of the corresponding monitoring area of ​​the associated route is positively correlated with the blocking time of the device.

5. The data processing system for allocating GPUs according to claim 1, characterized in that: The task scheduling unit is provided with task scheduling conditions and task scheduling methods corresponding to the task scheduling conditions, and the task scheduling method is to detect the estimated allowable coefficient of the schedulable node; The estimated allowable coefficient is related to the estimated number of personnel and the estimated number of devices in the monitoring area corresponding to the dispatchable node; The task scheduling condition is that there are schedulable nodes.

6. The data processing system for allocating GPUs according to claim 1, characterized in that: The data analysis unit monitors the quantity change state of a type of nodes and reduces the monitoring period when the quantity change state is an increasing state.

Images