Distributed fusion perception task management method, device and system

Abstract

The invention relates to the field of task processing, and particularly discloses a distributed fusion sensing task management method, device and system. The method comprises the following steps: acquiring node report data of a plurality of management nodes; wherein the node report data is used for representing computing power resource data and task state data of a plurality of computing power nodes managed by the management node; determining the node state of each management node according to the computing power resource data and the task state data of the plurality of computing power nodes managed by each management node; determining the management node of which the node state is in the first state as a first management node, and determining the management node of which the node state is in the second state as a second management node; and sending the first task management instruction to the first management node, and sending the second task management instruction to the second management node. According to the mode, the current node state of each management node can be dynamically determined, so that tasks are reasonably distributed, and effective management of a large number of computing power nodes is realized.

Description

Technical Field

[0001] The present disclosure relates to the field of task processing, and in particular to a method, device and system for managing distributed fusion perception tasks. Background Art

[0002] Distributed fusion perception tasks refer to the process of perceiving and monitoring the state of the system, environmental changes, or other specific tasks through the collaborative work of multiple nodes in a distributed computing system. Each node can independently collect data, process information, and share the results with other nodes to achieve global perception and decision-making. In a distributed system, nodes are usually distributed in different locations of the network. These nodes can be physical machines, virtual machines, or containers, etc.

[0003] In related technologies, since a distributed system involves a large number of nodes, how to effectively manage multiple nodes becomes a technical problem that needs to be solved urgently. Summary of the invention

[0004] To this end, the present disclosure provides a management method, device, system and electronic device for a distributed fusion perception task to solve the problem that multiple nodes in a distributed system are inconvenient to manage.

[0005] In order to achieve the above-mentioned object, the first aspect of the present disclosure provides a method for managing a distributed fusion perception task, the method for managing a distributed fusion perception task comprising:

[0006] Obtain node reporting data of multiple management nodes; wherein each management node is used to manage multiple computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the multiple computing nodes managed by the management node;

[0007] Determine the node status of each management node according to the computing resource data and task status data of multiple computing nodes managed by each management node;

[0008] Determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the management node in the first state has different remaining computing resources and / or the number of unprocessed tasks from the management node in the second state;

[0009] A first task management instruction is sent to the first management node, and a second task management instruction is sent to the second management node; wherein an instruction type of the first task management instruction is different from an instruction type of the second task management instruction.

[0010] In an optional implementation, the first state is used to characterize that the remaining computing power resources of the management node are less than a first preset value and / or the number of unprocessed tasks is greater than a second preset value; the second state is used to characterize that the remaining computing power resources of the management node are greater than a third preset value and / or the number of unprocessed tasks is less than a fourth preset value;

[0011] Furthermore, the instruction type of the first task management instruction includes: a removal type for removing the target task that has been allocated in the first management node; and the instruction type of the second task management instruction includes: a reallocation type for reallocating the target task.

[0012] In an optional implementation, the multiple management nodes include: a terminal management node for managing terminal computing nodes, and a cloud management node for managing cloud computing nodes;

[0013] Furthermore, in a case where the first management node includes the terminal management node, the second management node includes the cloud management node.

[0014] In an optional implementation, the second management node includes: a plurality of second management nodes respectively arranged at a plurality of different geographical locations;

[0015] Furthermore, when the distributed fusion perception task is used to perform calculations on a preset terminal computing node, after sending the second task management instruction to the second management node, the method further includes:

[0016] If it is detected that the displacement of the preset terminal computing power node is greater than the preset displacement threshold, the second management node is re-determined from the multiple second management nodes, and a second task management instruction is sent to the re-determined second management node to reallocate the target task to the re-determined second management node.

[0017] In an optional implementation, the distributed fusion perception task includes: a vehicle-road cooperative computing task in a vehicle-road cooperative scenario, and the computing power node includes: a computing power device in a vehicle-mounted terminal and / or a mobile terminal, and a roadside sensor device;

[0018] In addition, the computing power node is configured with a preset program interface, and the preset program interface is used to send node resource data to the management node and receive node task data from the management node.

[0019] A second aspect of the present disclosure provides a computing resource scheduling device based on distributed fusion perception tasks, the device comprising:

[0020] An acquisition module, adapted to acquire node reporting data of a plurality of management nodes; wherein each management node is used to manage a plurality of computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the plurality of computing nodes managed by the management node;

[0021] A status determination module, adapted to determine the node status of each management node according to the computing resource data and task status data of the multiple computing nodes managed by each management node;

[0022] a node determination module, adapted to determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the management node in the first state and the management node in the second state have different remaining computing resources and / or the number of unprocessed tasks;

[0023] The management module is adapted to send a first task management instruction to the first management node and send a second task management instruction to the second management node; wherein the instruction type of the first task management instruction is different from the instruction type of the second task management instruction.

[0024] A third aspect of the present disclosure provides a computing resource scheduling system based on distributed fusion perception tasks, the system comprising: the above-mentioned computing resource scheduling device, multiple management nodes, and multiple computing nodes.

[0025] A fourth aspect of the present disclosure provides an electronic device, including:

[0026] one or more processors;

[0027] A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned distributed fusion perception task management method;

[0028] One or more I / O interfaces are connected between the processor and the memory and are configured to implement information interaction between the processor and the memory.

[0029] A fifth aspect of the present disclosure provides a computer-readable medium having a computer program stored thereon, which implements the above-mentioned method for managing distributed fusion perception tasks when the program is executed by a processor.

[0030] The present disclosure also provides a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0031] The present disclosure has the following advantages:

[0032] In the management method of the distributed fusion perception task provided in the embodiment of the present disclosure, first, the node reporting data of multiple management nodes is obtained; wherein each node reporting data is used to characterize the computing power resource data and task status data of the multiple computing power nodes managed by the corresponding management node; then, according to the computing power resource data and task status data of the multiple computing power nodes managed by each management node, the node state of each management node is determined, and the management node whose node state is in the first state is determined as the first management node, and the management node whose node state is in the second state is determined as the second management node; finally, the first task management instruction is sent to the first management node, and the second task management instruction is sent to the second management node. It can be seen that, on the one hand, the present disclosure sets up multiple management nodes, each management node is used to manage multiple computing power nodes, and accordingly, through the parallel processing of multiple management nodes, a large number of computing power nodes can be conveniently managed in a coordinated manner. On the other hand, by dynamically collecting the node reporting data of multiple management nodes, the current node state of each management node can be dynamically determined, so as to reasonably allocate tasks and realize the effective management of a large number of computing power nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings are used to provide further understanding of the present disclosure and constitute a part of the specification. They are used to explain the present disclosure together with the following detailed description, but do not constitute a limitation of the present disclosure.

[0034] Figure 1 A flowchart of a method for managing a distributed fusion perception task provided by an embodiment of the present disclosure;

[0035] Figure 2 A schematic diagram of the structure of a distributed computing power scheduling system in an example of the present disclosure is shown;

[0036] Figure 3 The flowchart of the computing power node management function module is shown;

[0037] Figure 4 A schematic diagram of a virtual network module is shown;

[0038] Figure 5 A schematic diagram of the process of the server-client mode is shown;

[0039] Figure 6 A schematic diagram of a backend management system composed of a management node and a network node is shown;

[0040] Figure 7 shows a block diagram of a scheduling system;

[0041] Figure 8 A block diagram of a computing task running status display module is shown;

[0042] Fig. 9 A schematic diagram of a distributed fusion perception task management device provided by an embodiment of the present disclosure;

[0043] Fig.10 A schematic diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0044] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0045] As used in this disclosure, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0046] The terms used in the present disclosure are only used to describe specific embodiments and are not intended to limit the present disclosure.As used in the present disclosure, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0047] When the terms “comprising” and / or “made of…” are used in the present disclosure, it specifies the existence of the stated features, integers, steps, operations, elements and / or components, but does not exclude the existence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0048] The embodiments described in the present disclosure may be described with reference to plan views and / or cross-sectional views by way of ideal schematic diagrams of the present disclosure. Therefore, the exemplary illustrations may be modified according to manufacturing techniques and / or tolerances.

[0049] Unless otherwise defined, all terms (including technical and scientific terms) used in this disclosure have the same meaning as those commonly understood by those of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning unless explicitly defined in this disclosure.

[0050] In a first aspect, an embodiment of the present disclosure provides a method for managing a distributed fusion perception task. The management method can be executed by various task management devices, for example, the task management device can be a server in the cloud. Figure 1 A flowchart of a distributed fusion perception task management method provided by an embodiment of the present disclosure, referring to Figure 1 , the management method of the distributed fusion perception task includes:

[0051] Step S110: Obtain node reporting data of multiple management nodes; wherein each management node is used to manage multiple computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the multiple computing nodes managed by the management node.

[0052] Typically, a management node is used to coordinate and manage multiple computing nodes within a preset geographical area. For example, in the case where the distributed fusion perception task includes a vehicle-road collaborative computing task in a vehicle-road collaborative scenario, the computing nodes in the present disclosure may include: computing devices in vehicle-mounted terminals and / or mobile terminals (such as mobile phones, etc.), roadside sensing devices (such as infrared sensors, radar sensors, cameras, etc.). Among them, the distributed fusion perception task aims to assign the fusion task to multiple distributed nodes for processing to improve the accuracy, reliability and efficiency of the perception system.

[0053] In specific implementation, a preset program interface can be configured in the computing node, and the preset program interface is used to send node resource data to the management node and receive node task data from the management node. For example, a preset software program can be pre-installed in each computing node so that the node resource data of the computing node (such as the total computing resources, used computing resources, remaining computing resources, etc. of the computing node) can be dynamically reported to the management node through the software program. In addition, the software program can also receive the node task data sent by the management node, and the node task data is used to indicate the tasks to be calculated assigned to the computing node.

[0054] It can be seen that each management node can dynamically obtain the node resource data of the multiple computing nodes it manages, and each management node can also dynamically obtain the node task data of the multiple computing nodes it manages (such as the total number of tasks assigned to the computing node, the number of completed tasks, the number of unfinished tasks, etc.). Accordingly, each management node aggregates the node resource data of multiple computing nodes to obtain the computing resource data of the multiple computing nodes managed by the management node (usually the sum of the node resource data of multiple computing nodes). Similarly, each management node aggregates the node task data of multiple computing nodes to obtain the task status data of the multiple computing nodes managed by the management node (used to characterize the overall completion status of the node task data of multiple computing nodes).

[0055] Correspondingly, each management node can periodically send node reporting data to the task management device, and the node reporting data includes computing resource data and task status data of multiple computing nodes managed by the management node.

[0056] Step S120: Determine the node status of each management node based on the computing resource data and task status data of multiple computing nodes managed by each management node.

[0057] The node status of each management node is used to characterize the matching degree between the current computing resource data of the management node and the task status data. For example, the node status may include: the first status (also called resource shortage status), the second status (also called resource idle status), and the third status (also called resource balance status).

[0058] For example, when the current remaining computing power resource data of the management node is much greater than the resources required for the assigned task (such as the difference between the remaining computing power resources and the resources required for the assigned task is greater than a first preset difference threshold), the management node is determined to be in a resource-idle state; when the current remaining computing power resource data of the management node is equal to or less than the resources required for the assigned task, the management node is determined to be in a resource-strained state; when the current remaining computing power resource data of the management node is not much different from the resources required for the assigned task (such as the difference between the remaining computing power resources and the resources required for the assigned task is within a preset interval), the management node is determined to be in a resource-balanced state.

[0059] Of course, the node status of the management node can also be divided in many other ways, and the type and number of node status can also be flexibly set according to the actual scenario, which is not limited in the present disclosure.

[0060] Step S130: Determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the remaining computing resources and / or the number of unprocessed tasks of the management node in the first state are different from those of the management node in the second state.

[0061] For example, the remaining computing resources of the management node in the first state and the management node in the second state are different. Specifically, the management node with smaller remaining computing resources can be used as the management node in the first state, and the management node with larger remaining computing resources can be used as the management node in the second state.

[0062] For another example, the management node in the first state has a different number of unprocessed tasks from the management node in the second state. Specifically, the management node with a larger number of unprocessed tasks can be used as the management node in the first state, and the management node with a smaller number of unprocessed tasks can be used as the management node in the second state.

[0063] Of course, the status of the management node can also be comprehensively determined by combining the remaining computing resources and the number of unprocessed tasks.

[0064] For example, in an optional implementation, the first state is used to characterize that the remaining computing resources of the management node are less than a first preset value and / or the number of unprocessed tasks is greater than a second preset value. The second state is used to characterize that the remaining computing resources of the management node are greater than a third preset value and / or the number of unprocessed tasks is less than a fourth preset value. Among them, the first preset value may be equal to or less than the third preset value, and the second preset value may be equal to or greater than the fourth preset value. In addition, the first preset value, the second preset value, the third preset value, and the fourth preset value may be in numerical form or in ratio form (for example, represented by the ratio between the remaining resources and the total resources), and the present disclosure does not limit this.

[0065] In summary, the present disclosure does not limit the specific division method of the first state and the second state. The focus of the present disclosure is to dynamically determine the real-time state of each management node so as to perform targeted task management operations according to different real-time states.

[0066] Step S140: sending a first task management instruction to the first management node, and sending a second task management instruction to the second management node; wherein the instruction type of the first task management instruction is different from the instruction type of the second task management instruction.

[0067] The difference in instruction type between the first task management instruction and the second task management instruction can be reflected in multiple aspects, for example, the task operation mode can be different, such as multiple operation modes including task removal, task addition, task modification, etc. Another example is that the task parameters or task magnitudes contained in the task are different, such as tasks of different types or tasks of different data magnitudes.

[0068] It can be seen that in an optional implementation method, the first task management instruction and the second task management instruction are both used to assign tasks, the difference being that the types and data levels of the tasks assigned by the two are different, so that tasks that are easy to process or have a smaller data level can be assigned to management nodes with limited resources, and tasks that are time-consuming or have a larger data level can be assigned to management nodes with idle resources.

[0069] In another optional implementation, the first task management instruction is used to delete or remove tasks to reduce the load of management nodes with tight resources. The second task management instruction is used to add or reallocate tasks to increase the load of management nodes with idle resources. For example, the instruction type of the first task management instruction may include: a removal type, which is used to remove the target task that has been allocated in the first management node; the instruction type of the second task management instruction may include: a reallocation type, which is used to reallocate the target task, thereby transferring the target task in the first management node to the second management node, so as to achieve reasonable scheduling of computing resources between management nodes.

[0070] In addition, those skilled in the art may also make various modifications and variations to the above implementations:

[0071] In an optional implementation, the multiple management nodes may include: a terminal management node for managing terminal computing nodes, and a cloud management node for managing cloud computing nodes. Correspondingly, in the case where the first management node includes a terminal management node, the second management node includes a cloud management node. Among them, the terminal management node can be various management nodes located on the vehicle side or the road side, which are used to manage various terminal computing nodes on the vehicle side or the road side, such as vehicle-mounted terminals, mobile terminals, etc. The cloud management node can be various management nodes located on the cloud server side, which are used to manage various cloud computing nodes such as cloud servers. Considering that the computing power resources of cloud computing nodes are usually large, the tasks in the terminal computing nodes can be transferred to the cloud computing nodes to use the computing power resources of the cloud to realize complex computing processing, thereby improving computing efficiency.

[0072] In addition, it should be noted that whether the terminal management node fails or the computing resources of the terminal management node are insufficient, the assigned tasks in the terminal management node can be dynamically transferred to the cloud management node, thereby using the resources of the cloud to ensure the reliable execution of computing tasks and improve the robustness of the system.

[0073] In another optional implementation, the second management node may include: a plurality of second management nodes respectively arranged at a plurality of different geographical locations. For example, the second management node may include a plurality of cloud management nodes respectively located in different road sections to manage cloud computing nodes in different road sections. Accordingly, in the case where the distributed fusion perception task is used to perform calculations for a preset terminal computing node, after sending a second task management instruction to the second management node, the following operations may be further performed: if it is detected that the displacement of the preset terminal computing node is greater than a preset displacement threshold, the second management node is re-determined from the plurality of second management nodes, and a second task management instruction is sent to the re-determined second management node, so that the target task is reallocated to the re-determined second management node. Among them, the preset terminal computing power node can be an on-board computing power node in a moving vehicle. Since the vehicle moves at a fast speed and has a large displacement, if it is detected that the displacement of the preset terminal computing power node is greater than the preset displacement threshold (the displacement can specifically be the displacement generated relative to the position corresponding to the time point when the second task management instruction was first sent to the second management node), it means that the vehicle has moved a long distance, and the cloud computing power node corresponding to the second management node previously assigned to it is already far away from the vehicle. Therefore, a new second management node is re-assigned to it to schedule the cloud computing power node closer to the current position of the vehicle for calculation, thereby facilitating the improvement of computing efficiency and reducing transmission time.

[0074] It can be seen that the above method can dynamically switch multiple second management nodes in the scenario of vehicle movement to achieve flexible scheduling of management nodes.

[0075] To facilitate understanding, an example is used below to introduce the implementation of the above embodiment.

[0076] This example is implemented based on a distributed computing power scheduling system. Specifically, through cloud-edge collaboration and edge computing platforms, the computing power of various computing powers and resources is integrated, and various devices and smart terminals are included in the scheduling system for unified management. Among them, various devices and smart terminals may include: X86 servers, ARM devices, GPU devices, smart terminals, mobile handheld devices, storage devices, low-power devices, etc. Devices can be connected to the computing power scheduling system through the network or 5G, etc., to form a virtualized resource pool that is distributed in physical form and unified in logical form. According to the resource type, the resource pool can be divided into multiple atomic resources such as computing power, network, storage, etc., and spliced ​​according to actual needs to form a complete computing power unit. One or more computing power units can constitute a computing power node mentioned above. The resource pool has monitoring and self-healing capabilities. By matching the supply and demand of the demand side and the supply side, the resource pool is dynamically managed and expanded, which can not only meet the needs of different applications, but also improve operation and maintenance efficiency and save operating costs.

[0077] This platform can be divided into multiple parts in a modular design to implement different functional modules. Among them, due to the particularity of computing power nodes, such as dispersion and heterogeneity, computing power nodes (also called computing power access nodes, computing power units, or computing power devices) need to support devices of various specifications and have flexible network access capabilities. Considering the flash disconnection and frequent shutdown of computing power equipment, it is also necessary to register and stabilize the computing power nodes. To this end, this example integrates the two functional modules of computing power node access and management into a minimum deployment unit (the minimum deployment unit is a management node mentioned above). Through horizontal expansion, the number of management nodes can be expanded, thereby expanding the access capacity and number of computing power nodes. The minimum deployment unit can be further configured with a computing task distribution module, a computing power node resource management function module, a computing power node management function module, etc. Therefore, the minimum deployment unit is also called a management node.

[0078] In addition, this example implements centralized management, unified monitoring, centralized scheduling and other functions in terms of scheduling and monitoring management, and can monitor and globally schedule across management nodes. In terms of design, the first deployment is completed at the same time as the first management node. When new management nodes are added in the future, they can be connected to the monitoring system through the API. Therefore, the computing power scheduling module and the computing task running status display module can also be integrated together as a monitoring node for development and deployment.

[0079] In addition, there are many requirements between management nodes, such as task scheduling, data interaction, and system multi-activity. Therefore, the communication between management nodes needs to build a virtual network on the ordinary wide area network, encrypt and protect the network, and ensure the security of the pipeline. Between each management node, it is necessary to independently develop a virtual network module between computing nodes and build a network management module.

[0080] To sum up, the whole system consists of three parts, namely: management node, monitoring node and network module. Figure 2 A schematic diagram of the structure of the distributed computing power scheduling system in this example is shown. Figure 2 As shown, the distributed computing power scheduling system includes: a computing task running status display module, a computing power scheduling module (also called a task scheduling module or a task management module), a virtual network module between computing power nodes, and two management nodes, namely management node 1 and management node 2. Each management node further includes the computing task distribution module, computing power node resource management function module, and computing power node management function module mentioned above. It should be noted that Figure 2 In the example, two management nodes are used for illustration. In actual situations, the number of management nodes can be large and can be expanded horizontally as needed. In addition, the computing node resource management function module and the computing node management function module in the management node can also be merged into one management function module, which is not limited in this example. In addition, the computing task running status display module and the virtual network module between computing nodes are not necessary and can be omitted.

[0081] It can be seen from this that Figure 2 The management nodes in the system are mainly used to access and pool computing units, and to virtualize scattered and unreliable computing resources into a resource pool with SLA capabilities. The computing power scheduling module is mainly used to aggregate information from multiple management nodes, form a unified monitoring view upward, and provide managers with global information display through the computing task running status display module; resources and tasks can be scheduled between multiple management nodes downward. In addition, the management node is scalable and can be expanded according to actual conditions and performance requirements when in use.

[0082] The following are respectively Figure 2 The specific working process of each module or node in the following is introduced:

[0083] (1) Computing power node management function module:

[0084] The computing node management function module is the core component of the system. Specifically, it is a submodule in the management node, which is mainly used to manage the computing nodes (i.e. computing nodes) in the computing cluster. Through this module, the system has the ability to adjust computing power, which can improve computing efficiency and resource utilization. In addition, through the multi-node deployment method, the entire system has redundancy and a certain degree of fault self-healing ability, ensuring the stability and reliability of the computing cluster.

[0085] This module includes the following functions:

[0086] Node registration and deregistration: Manage nodes in the computing cluster through automatic discovery or manual operation, with the ability to add or delete nodes.

[0087] Task scheduling and allocation: Assign tasks to computing nodes and return results to the master node after the tasks are completed.

[0088] Resource management and optimization: When node resources are limited, resources are allocated and utilized reasonably to maximize computing efficiency.

[0089] Error and fault management: Monitor the operating status of computing nodes, detect and handle node failures and errors in a timely manner, and ensure the stability and reliability of the computing cluster.

[0090] Figure 3 The following figure shows the workflow diagram of the computing power node management function module. Figure 3 As shown in the figure, the entire management system adopts a device discovery mechanism, and completes the discovery and update of nodes on the device side through IP reporting and scheduled query mechanisms. When a newly added node is started, it will send a registration request to the cluster management software, which contains basic information about the computing node, such as IP address, host name, hardware configuration, etc. After receiving the request, the cluster management software will save the information of the node to its database for subsequent management. Once an IP is reported, the system will match it with the existing database. If the main fields of the node, such as IP address, MAC address, user information, etc., are not found, it is regarded as a new node and the node registration process begins. After the node is registered, the system will detect the online status of the system by sending heartbeat packets at regular intervals, and detect the network health, including latency, packet loss rate, routing information, etc., based on the data packets. If the node does not respond to the heartbeat packet for a long time, the cluster management software will consider the node to be offline and it needs to be marked as unavailable.

[0091] After confirming that the node is available, the cluster management software will intelligently allocate resources and load balance the nodes according to the hardware configuration, current load, and other conditions of each computing node. When scheduling tasks, the cluster management software will consider the status and available resources of each node and assign tasks to the most suitable node. When the task is completed, or when the computing node needs to exit from the entire system, the management node can be adjusted by logging out. If the computing power node actively applies for offline, it will send a logout request to the cluster management software. After receiving the request, the cluster management software will delete the node's information from its database and release its resources. If the computing power node does not meet the requirements, such as hardware matching, network quality, software version, etc., the cluster management software can also actively initiate an offline operation to remove the computing power node from the system until it completes the transformation and meets the requirements and goes back online.

[0092] (2) Computing node resource management function module:

[0093] The computing node resource management function module is a submodule within the management node, which is mainly used to access and manage computing nodes scattered in various places, including x86, ARM, MIPS, GPU and various mobile devices, portable terminals, etc. These computing resources are connected to the platform service layer through the network layer, and are uniformly managed by the platform service layer, providing computing power guarantee for various tasks through virtualization and other technical means.

[0094] It can be seen that through the computing power node resource management function module, computing power resources are connected to the platform server through the network layer, and through virtualization, computing power guarantee is provided for various tasks.

[0095] Among them, the network layer refers to the use of various network technologies including Wifi, 5G, etc. to connect computing terminals and computing nodes distributed in various places, and use virtual private network and wide area network technologies to ensure the management platform's ability to schedule resources, distribute tasks, report results, and manage exceptions for each computing node.

[0096] Among them, the platform service layer (located in the management node) includes the computing power node management function module and the computing power node resource management function module. It is a control platform built based on virtualization technology and is divided into two parts: software and hardware. The hardware system is composed of various management servers, including various control servers, storage devices, virtualization software, etc., collectively referred to as management nodes; the software part mainly includes virtualization management software, network management software, task scheduling software, security management software, etc.

[0097] The virtualization management software mainly manages the computing power, network and storage resources on each computing node and distributed node. According to the actual situation, it splits and splices the resources of the physical nodes to meet the needs of different tasks.

[0098] Network management software is used to manage network traffic between virtual resources and implement network layer functions such as OVS;

[0099] Task scheduling software is mainly used for tasks creation, task distribution, status management, monitoring and other functions;

[0100] Security management software provides security functions such as identity authentication and logging.

[0101] The user interface includes an interface accessed using a browser, as well as an API or SDK that can be directly called through scripts and secondary development to operate and control the system through web pages and other means.

[0102] In addition, the system also fully considers security and monitoring capabilities. Since its functions are composed of related services in each system component, they will not be elaborated on.

[0103] (3) Virtual network module between computing nodes:

[0104] The virtual network module between computing nodes is mainly used to interconnect multiple management nodes based on the existing Internet or wide area network to ensure that data communication between management nodes is established on a reliable and secure network environment. The network supports the IPv6 protocol. Considering virtualization and container technology, tenant isolation is also performed at the network layer. Logical isolation between tenants is achieved by using OVS and SDN controller technology to meet actual business needs.

[0105] Figure 4 FIG. 1 shows a schematic diagram of a virtual network module. Figure 4 As shown in the figure, the network module is used to connect two management nodes through the IPv6 protocol. In a multi-tenant environment, the intranet network communication of different tenants can be divided into two types. One is that different tenants are on different hosts. When different tenants are on different hosts, if the controller determines that the source and destination IP of the message are not in the same tenant, it will discard the message before sending it to avoid unnecessary network overhead. The other is that different tenants are on the same host. When different tenants are on the same host, the intranet network card of tenant A and tenant B is eth0, and the internal access is on the same OVS. Among them, the intranet IP address of the public cloud host is provided by the platform and cannot be modified by the console, so the controller caches the tenant and IP information. When tenant A sends a message, when the message arrives at OVS for forwarding, the controller will determine whether its source and destination IP is in the same tenant. If it is not in the same tenant, the message will be directly blocked from being forwarded to tenant B, and user isolation is achieved in this way. This judgment only exists in the first packet, and the judgment information will be cached in the flow table, and no judgment will be required later.

[0106] (4) Computing task distribution module:

[0107] The computing task distribution module is a submodule inside the management node, which is used to push computing tasks to the computing power units under the management node. Therefore, the working scope of the computing task distribution module only covers the current management node (that is, the management node to which the computing task distribution module belongs). The task distribution and computing power scheduling across nodes are performed by a higher-level computing power scheduling module.

[0108] The computing task distribution module is used to match computing power rules and task characteristics. Computing power rules include computing power size, network conditions, stability, storage size, etc., and have the ability to measure computing power nodes. Task characteristics refer to the type of task, which can be simply divided into storage tasks (distributed storage, which has high storage requirements and low computing power requirements), data computing tasks (big data analysis, data mining, etc., which have high requirements for computing power), and resource usage tasks (such as network bandwidth usage, IP addresses, etc.). In subsequent use, computing power rules and task characteristics can be readjusted through rule definitions.

[0109] In order to improve reliability, the server-client mode (i.e., server-client mode) was first determined in system design. Figure 5 FIG. 2 shows a flow chart of the server-client mode. Figure 5 As shown in the figure, the task distribution module is responsible for issuing instructions according to the rules as the server. After receiving the task execution instruction, the computing node starts the local task process (container) and reports the running status regularly according to the rules. When the task is completed, the execution result is reported to the distribution module. The module records the result and then sends the status update record according to the situation. Secondly, a communication method based on grpc is adopted. During the task execution process, considering the possibility of communication and data exchange between processes, after evaluating the performance, the use of http and mq is abandoned and grpc is selected. In addition to improving reliability, it also has certain advantages in subsequent development and maintenance requirements. Figure 6 The schematic diagram of the backend management system composed of management nodes and network nodes is shown. Influx DB is a time series database. Figure 6 As shown, the background display module (i.e., Web) can display user information (including user login information, logs, equipment and node information, etc.) based on the MySQL database; can display usage collection information (including user ID, task ID, resource ID, timeline, etc.) based on the influx DB database; can display information related to network functions, such as node health status, network connection quality, etc., based on routing information and network status.

[0110] (5) Computing power scheduling module:

[0111] The computing power scheduling module is a global computing power management and task allocation module, also known as a scheduling system or task scheduling module, which is used to uniformly manage multiple management nodes. It mainly manages the computing power between management nodes and is also the interface for tasks to enter the system. Administrators can use scripts to package computing tasks into containers or encapsulate them into images, use K8S to issue tasks, distribute tasks to various management nodes, and then push tasks to computing power nodes through the task distribution module on the management node. In this example, the matching of tasks and computing power is based on computing power and task characteristics. In addition, algorithm models can be introduced to automatically match and adapt according to multiple factors such as region, computing power node characteristics, task priority, usage cost, etc., to achieve maximum optimization of system capabilities.

[0112] Figure 7 The block diagram of the scheduling system is shown. Among them, Etcd is a distributed key-value database, which includes three, namely Etcd01, Etcd02 and Etcd03, which saves the metadata of the entire cluster. The functions include: checking the health status of the cluster, endpoint status, alarm information, obtaining the current version, compressing all old versions, cleaning up excess fragmentation space, and canceling alarm information. Master is the control component (corresponding to Master cluster 1, K8S, and Master cluster 2 in the figure), which is mainly responsible for the management of the entire cluster, such as container scheduling, maintaining resource status, automatic expansion, and rolling updates. It mainly includes kube-apiserver, kube-scheduler, and kube-controller-manager components.

[0113] Among them, the k8s server includes: Scheduler, API Server, and Controller. The functions implemented by the computing power scheduling module mainly include task allocation and load balancing capabilities. Task allocation is to assign tasks to the most suitable device for processing based on task type, priority, and available resources of edge devices. Load balancing capability refers to the computing power scheduling module using a load balancing algorithm to evenly allocate tasks to edge devices to avoid excessive or low device loads, thereby achieving efficient operation of the system.

[0114] In addition, the functions include predictive analysis and offline computing. Predictive analysis refers to the prediction and analysis of the characteristics of tasks and the resources of equipment based on historical data and real-time data, so as to make scheduling decisions in advance and realize intelligent optimization of the system. It needs to maintain historical data and real-time data to realize data analysis and prediction of tasks and equipment. Offline computing means that due to network reasons, after the computing power node is offline, the task can still maintain its status, and after the communication is restored, the task can be automatically uploaded to avoid the waste of computing power due to unexpected failures.

[0115] (6) Computing task running status display module:

[0116] The computing task running status display module, also known as the monitoring module, is a service that monitors the virtualization resources and task running status. This module is an optional module. Figure 8 A block diagram of a computing task running status display module is shown.

[0117] Among them, the status of computing tasks running on all computing nodes is uniformly displayed through a Web page, including but not limited to: the running time of computing tasks, the use of computing node resources, and the multi-dimensional computing task deployment display (such as by computing task owner, by computing node and other different dimensions). Users can use the monitoring system to fully grasp key information such as resource usage and task running status, and can also see the use and operation of storage, database, network, cache and other resources in detail. Combined with operation and maintenance systems such as log management, an integrated functional system of collection, storage, display, calculation, aggregation and statistics can be gradually realized to provide a reference for matching tasks and resources. In this example, monitoring capabilities can include: Visual monitoring data: charts and statistical information of monitoring indicators, and quick viewing of historical trends; task monitoring: including the status of the process and the resources used when the task is running, such as the host's CPU, disk and network information.

[0118] This example establishes a distributed computing power scheduling and management system based on edge cloud. By designing management nodes, monitoring nodes, and network modules, it realizes the unified management and efficient utilization of edge resources, as well as the automated supervision and scheduling of edge computing tasks.

[0119] In addition, in some optional implementations, this example may further include the following multiple functional modules, which may be added on the basis of the 6 functional modules mentioned above, or may directly replace some of the 6 functional modules mentioned above:

[0120] (1) Equipment management module:

[0121] The device management module provides access and management capabilities for devices. Through a cross-platform agent, a unified management method is provided to uniformly manage access to devices that meet the requirements, such as X86 servers, desktop computers, low-power computing units, mobile terminals, and NAS systems. Based on the hardware characteristics of the device, a unique identifier is generated for each device as a unique node identifier for registration and joining the system and throughout the entire life cycle. Through node identity detection, security policy detection, and baseline testing, the security and reliability of access devices are ensured, and nodes that fail to pass registration or do not meet the requirements are bypassed and isolated to prevent illegal access. After entering the platform, access devices that meet the requirements use agents to centrally control device status information, including node status, task operation status, performance indicators, etc.

[0122] (2) Network access module:

[0123] The network access module is mainly used to manage the connection between nodes and resources, and to transmit the platform's control information to achieve centralized control and unified distribution capabilities. The network access module can be implemented based on the IPv6 VPN network of the wide area network to ensure that all access nodes have unique IP addresses and that the nodes accessing the network have undergone identity access verification. Data transmission between nodes is encrypted and cannot be monitored or stolen. Multiple access areas can be planned according to the topology, with multiple access points in one area. When an access point fails, the terminal device can automatically switch to other access points in the same area.

[0124] (3) User management module:

[0125] Used to manage relevant information of system users, including user login policies, operation logs, user permission management, access control settings, and login environment detection.

[0126] (4) Resource pool management module:

[0127] The resource pool management module is used to label and manage the resources under unified management. It can group devices according to different physical locations, resource attributes (computing power, intelligent computing, network, storage, etc.) or other attributes, so that the appropriate resources can be automatically matched according to the characteristics and attributes when scheduling tasks. The division rules of resource pools are required to be adjustable and can be divided according to administrative regions. The minimum granularity should be based on blocks or parks to classify and group resource pools.

[0128] (5) Performance monitoring module:

[0129] The performance monitoring module (which can be integrated into the management node) contains various tools to test and evaluate the managed resources, eliminate unqualified equipment, and perform SLA classification on the equipment that meets the requirements to meet the differentiated needs of the tasks. This module has built-in various tools that can detect and detect the node health status, network quality, storage space, utilization of key components such as CPU or memory, task running status, etc., and provide corresponding thresholds and alarm management. The collected information is uniformly stored in the back end, which is convenient for providing data support for the management of nodes and tasks in various dimensions, evaluating health, triggering automatic adjustments, displaying in the interface, and configuring alarms.

[0130] (6) Task management interface:

[0131] The task management interface (which can be integrated into the management node or into the computing power scheduling module) refers to the task management and scheduling module designed based on the container method, which is mainly used for task creation and deletion. Task creation supports the configuration of basic conditions required for tasks, including but not limited to deployment location, service coverage, requirements for occupying various resources, task start, stop and migration conditions, as well as distribution, scheduling, start and stop, migration, switching and termination.

[0132] Task management is completed by multiple programs working together in the background. First, the computing power node information that meets the requirements is obtained from the module of the resource pool. The information obtained includes the location, performance, current status, etc. of the computing power node. This information is compared with the information returned by the task template. The comparison results will be written to the corresponding database. After the comparison is completed, the system will obtain the computing power node information that meets the requirements from the database, and sort it according to priority to confirm the list of resources required for this task. After that, the task will be sent to the computing power node in the form of a container through the scheduling component, and the task name, start time, and resource usage information will be written to the corresponding database, and the monitoring component will be triggered to monitor the computing power and resource consumption. At the same time, the newly added task status will be added to the alarm monitoring system, combined with the resource consumption information, to detect the process status in real time to avoid abnormalities such as freezes or no loads.

[0133] (7) Task scheduling component:

[0134] The task scheduling component is the core component of the system, which is used to support the dispatch of tasks and the matching of resources. It specifically includes the node and task management cluster, real-time data storage cluster, database cluster, system and task mirroring, bastion host, load balancing component, monitoring alarm cluster, API service and other core systems.

[0135] (8)Task configuration module:

[0136] The task configuration module pre-configures tasks through scenarios and templates, including resource consumption estimates and recommended references required for the system to complete tasks, analyzes task characteristics, extracts key information, and combines resource consumption and key indicator monitoring during task operation to continuously self-correct and improve, thereby improving the accuracy of recommendations.

[0137] (9) User interface and console:

[0138] The system interface and console can be used to configure, manage and view the above systems in a graphical interface or command line mode, including completing basic functions such as task management, resource management, monitoring display, fault handling, and system settings.

[0139] (10) Security module:

[0140] The security module can cover multiple functions such as identity management, data protection, network security, log management, etc. Taking identity management as an example, corresponding deployments have been made in the following aspects:

[0141] User identity authentication: On the login interface, provide users with identity authentication methods, and obtain corresponding permissions through user name and password detection.

[0142] Access control: Based on groups and categories, establish sophisticated access control policies, and adjust access rights based on factors such as users, roles, resources, and operations. For example, different permissions can be assigned to different user groups to limit the operations that can be performed.

[0143] Audit and monitoring: The system has complete logging capabilities, can monitor user activities and record corresponding logs, so as to audit and investigate violations. The audit log should include user login information, access records, and operation details.

[0144] Data protection: It can provide data encryption function and support SM algorithm to ensure the security and reliability of user information. In addition, during the transmission process, it can also achieve end-to-end reliable links through communication encryption.

[0145] On the management node, the network is divided into multiple security zones, and network segmentation is achieved through technologies such as virtual private cloud (VPC) or virtual local area network (VLAN). Resources with different security levels can be placed in different network segments to achieve isolation and protection between resources.

[0146] Between management nodes, use ACL to control the inflow and outflow of network traffic and limit communication between different network areas. ACL can be configured based on factors such as IP address, port, protocol, etc. to enhance network security. Between devices, you can also deploy products such as firewalls to implement traffic monitoring and filtering to improve overall security.

[0147] In summary, on a dispatching platform that may need to connect to various external devices, in addition to common system capabilities, you can also combine the enterprise security strategy of Zhiwang to specify complete system protection measures to achieve comprehensive protection of the system and data.

[0148] In a second aspect, an embodiment of the present disclosure provides a management device for distributed fusion perception tasks. Fig. 9 A schematic diagram of a distributed fusion perception task management device provided by an embodiment of the present disclosure is shown, referring to Fig. 9 , the management device of the distributed fusion perception task includes:

[0149] An acquisition module 91 is adapted to acquire node reporting data of a plurality of management nodes; wherein each management node is used to manage a plurality of computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the plurality of computing nodes managed by the management node;

[0150] A state determination module 92, adapted to determine the node state of each management node according to the computing resource data and task state data of the multiple computing nodes managed by each management node;

[0151] The node determination module 93 is adapted to determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the management node in the first state has different remaining computing resources and / or the number of unprocessed tasks from the management node in the second state;

[0152] The management module 94 is adapted to send a first task management instruction to the first management node and send a second task management instruction to the second management node; wherein the instruction type of the first task management instruction is different from the instruction type of the second task management instruction.

[0153] In an optional implementation, the first state is used to characterize that the remaining computing resources of the management node are less than a first preset value and / or the number of unprocessed tasks is greater than a second preset value; the second state is used to characterize that the remaining computing resources of the management node are greater than a third preset value and / or the number of unprocessed tasks is less than a fourth preset value;

[0154] Furthermore, the instruction type of the first task management instruction includes: a removal type for removing the target task that has been allocated in the first management node; and the instruction type of the second task management instruction includes: a reallocation type for reallocating the target task.

[0155] In an optional implementation, the multiple management nodes include: a terminal management node for managing terminal computing nodes, and a cloud management node for managing cloud computing nodes;

[0156] Furthermore, in a case where the first management node includes the terminal management node, the second management node includes the cloud management node.

[0157] In an optional implementation, the second management node includes: a plurality of second management nodes respectively arranged at a plurality of different geographical locations;

[0158] Furthermore, when the distributed fusion perception task is used to perform calculations on a preset terminal computing power node, the management module is further used to:

[0159] If it is detected that the displacement of the preset terminal computing power node is greater than the preset displacement threshold, the second management node is re-determined from the multiple second management nodes, and a second task management instruction is sent to the re-determined second management node to reallocate the target task to the re-determined second management node.

[0160] In an optional implementation, the distributed fusion perception task includes: a vehicle-road cooperative computing task in a vehicle-road cooperative scenario, and the computing power node includes: a computing power device in a vehicle-mounted terminal and / or a mobile terminal, and a roadside sensor device;

[0161] In addition, the computing power node is configured with a preset program interface, and the preset program interface is used to send node resource data to the management node and receive node task data from the management node.

[0162] The functions or modules included in the device provided in the embodiments of the present disclosure can be used to execute the method described in the embodiments of the first aspect disclosed above. The specific implementation and technical effects thereof can be referred to the description of the method embodiments above, and will not be repeated here for the sake of brevity.

[0163] In addition, the embodiment of the present disclosure also provides a computing resource scheduling system based on distributed fusion perception tasks, the system comprising: Fig. 9 The computing resource scheduling device, multiple management nodes, and multiple computing nodes shown in the figure. The computing resource scheduling device can be specifically the task management device mentioned above, which can be specifically Figure 2The computing resource scheduling module in . It can be seen that the computing resource scheduling system based on distributed fusion perception tasks can be specifically Figure 2 Distributed computing power scheduling system in .

[0164] It should be noted that all modules involved in this embodiment are logic modules. In practical applications, a logic unit may be a physical unit, or a part of a physical unit, or may be implemented as a combination of multiple physical units. In addition, in order to highlight the innovative part of the present disclosure, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by the present disclosure, but this does not mean that there are no other units in this embodiment.

[0165] Reference Fig.10 , an embodiment of the present disclosure provides an electronic device, comprising:

[0166] One or more processors 601;

[0167] A memory 602, on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement any one of the above-mentioned methods for managing distributed fusion perception tasks;

[0168] One or more I / O interfaces 603 are connected between the processor and the memory and are configured to implement information exchange between the processor and the memory.

[0169] Among them, the processor 601 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 602 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH); the I / O interface (read-write interface) 603 is connected between the processor 601 and the memory 902, and can realize information interaction between the processor 601 and the memory 602, including but not limited to a data bus (Bus), etc.

[0170] In some embodiments, the processor 601 , the memory 602 , and the I / O interface 603 are connected to each other through a bus, and further connected to other components of the computing device.

[0171] This embodiment also provides a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the distributed fusion perception task management method provided in this embodiment is implemented. To avoid repeated description, the specific steps of the distributed fusion perception task management method are not repeated here.

[0172] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0173] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0174] Those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features from different embodiments is meant to be within the scope of the present embodiment and to form different embodiments.

[0175] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A method for managing distributed fusion perception tasks, characterized in that: The method comprises: Obtain node reporting data of multiple management nodes; wherein each management node is used to manage multiple computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the multiple computing nodes managed by the management node; Determine the node status of each management node according to the computing resource data and task status data of multiple computing nodes managed by each management node; Determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the management node in the first state has different remaining computing resources and / or the number of unprocessed tasks from the management node in the second state; A first task management instruction is sent to the first management node, and a second task management instruction is sent to the second management node; wherein an instruction type of the first task management instruction is different from an instruction type of the second task management instruction.

2. The method according to claim 1, characterized in that The first state is used to characterize that the remaining computing resources of the management node are less than a first preset value and / or the number of unprocessed tasks is greater than a second preset value; the second state is used to characterize that the remaining computing resources of the management node are greater than a third preset value and / or the number of unprocessed tasks is less than a fourth preset value; Furthermore, the instruction type of the first task management instruction includes: a removal type for removing the target task that has been allocated in the first management node; and the instruction type of the second task management instruction includes: a reallocation type for reallocating the target task.

3. The method according to claim 2, characterized in that The multiple management nodes include: a terminal management node for managing terminal computing nodes, and a cloud management node for managing cloud computing nodes; Furthermore, in a case where the first management node includes the terminal management node, the second management node includes the cloud management node.

4. The method according to claim 3, characterized in that The second management node includes: a plurality of second management nodes respectively arranged at a plurality of different geographical locations; Furthermore, when the distributed fusion perception task is used to perform calculations on a preset terminal computing node, after sending the second task management instruction to the second management node, the method further includes: If it is detected that the displacement of the preset terminal computing power node is greater than the preset displacement threshold, the second management node is re-determined from the multiple second management nodes, and a second task management instruction is sent to the re-determined second management node to reallocate the target task to the re-determined second management node.

5. The method according to any one of claims 1 to 4, characterized in that: The distributed fusion perception task includes: a vehicle-road collaborative computing task in a vehicle-road collaborative scenario, and the computing power nodes include: computing power equipment in a vehicle-mounted terminal and / or a mobile terminal, and a roadside sensor device; In addition, the computing power node is configured with a preset program interface, and the preset program interface is used to send node resource data to the management node and receive node task data from the management node.

6. A computing resource scheduling device based on distributed fusion perception tasks, characterized in that: The device comprises: An acquisition module, adapted to acquire node reporting data of a plurality of management nodes; wherein each management node is used to manage a plurality of computing nodes, and the node reporting data is used to characterize computing resource data and task status data of the plurality of computing nodes managed by the management node; A status determination module, adapted to determine the node status of each management node according to the computing resource data and task status data of the multiple computing nodes managed by each management node; a node determination module, adapted to determine a management node whose node status is in a first state as a first management node, and determine a management node whose node status is in a second state as a second management node; the management node in the first state and the management node in the second state have different remaining computing resources and / or the number of unprocessed tasks; The management module is adapted to send a first task management instruction to the first management node and send a second task management instruction to the second management node; wherein the instruction type of the first task management instruction is different from the instruction type of the second task management instruction.

7. A computing resource scheduling system based on distributed fusion perception tasks, characterized in that: The system includes: the computing resource scheduling device described in claim 6, multiple management nodes, and multiple computing nodes.

8. An electronic device, characterized in that: include: one or more processors; A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5; One or more I / O interfaces are connected between the processor and the memory and are configured to implement information interaction between the processor and the memory.

9. A computer readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method according to any one of claims 1 to 5.

10. A computer program product, characterized in that The invention comprises a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the method as claimed in any one of claims 1 to 5.

Images