Edge end load dynamic balancing method and edge computing system

Abstract

The invention relates to an edge end load dynamic balancing method and an edge computing system.The edge end load dynamic balancing method comprises the steps that when the load state of at least one edge computing node is high load, the load state of an edge computing cluster is judged to be high load, and otherwise, the load state of the edge computing cluster is judged to be low load; when the load state of the edge computing cluster is low load, allocating edge computing tasks to each edge computing node in turn; and when the load state of the edge computing cluster is a high load, preferentially allocating an edge computing task to the edge computing node with the highest resource score, calculating the resource data of each edge computing node through a non-integral rank sum ratio method to obtain the resource score of each edge computing node, and negatively correlating the resource score of each edge computing node with the load degree. The invention solves the problem that the current edge end load balancing method cannot dynamically allocate the calculation task for the edge node.

Description

Technical Field

[0001] The present application relates to the field of edge computing, and in particular to an edge load dynamic balancing method and an edge computing system. Background Art

[0002] With the rapid development of technologies such as cloud computing, big data, and the Internet of Things, edge computing, as an emerging computing paradigm, has attracted more and more attention. Edge computing reduces data transmission latency, saves bandwidth, and improves real-time performance and reliability by sinking computing and storage resources to the edge of the network so that they are closer to the data source. However, edge devices usually face the challenge of limited computing power and storage resources, which makes load balancing a key technology to ensure the efficient operation of edge computing.

[0003] In traditional load balancing strategies (such as round-robin scheduling, random balancing scheduling, etc.), tasks are mainly allocated based on fixed algorithms or pre-set rules. At the same time, many existing studies reduce latency by adjusting these static configurations. These methods are often difficult to cope with sudden load changes and resource bottlenecks in dynamic environments. For example, forwarding tasks to edge nodes that are already full of a resource according to static rules causes the edge nodes to continue to be overloaded, which not only fails to fully utilize the overall performance of the system, but also reduces the overall resource utilization of the system.

[0004] There is currently no effective solution to the problem that current edge load balancing methods cannot dynamically allocate computing tasks to edge nodes. Summary of the invention

[0005] The present invention provides an edge load dynamic balancing method and an edge computing system to solve the problem that the current edge load balancing method cannot dynamically allocate computing tasks to edge nodes.

[0006] In a first aspect, the present invention provides an edge end load dynamic balancing method, comprising:

[0007] Obtain resource data and load status of each edge computing node in the edge computing cluster, where the load status of each edge computing node is determined based on its own resource data;

[0008] When there is at least one edge computing node whose load state is high load, the load state of the edge computing cluster is determined to be high load; otherwise, the load state of the edge computing cluster is determined to be low load;

[0009] When the load state of the edge computing cluster is low load, allocating edge computing tasks to each edge computing node in turn;

[0010] When the load state of the edge computing cluster is high, edge computing tasks are preferentially assigned to edge computing nodes with the highest resource scores. The resource score of each edge computing node is calculated by calculating its own resource data through the non-integer rank sum ratio method. The resource score of the edge computing node is negatively correlated with the load degree.

[0011] In a second aspect, the present invention provides an edge computing system, including an edge computing cluster, a central control module and a plurality of resource monitoring modules, wherein the edge computing cluster includes a plurality of edge computing nodes;

[0012] Each of the resource monitoring modules corresponds to each of the edge computing nodes one by one, and each of the resource monitoring modules is configured to: obtain resource data of the corresponding edge computing node in real time and determine the load state of the corresponding edge computing node according to the resource data;

[0013] The central control module is configured to: obtain the resource data and load status of each edge computing node from each resource monitoring module; when there is at least one edge computing node with a high load, determine the load status of the edge computing cluster to be high load, otherwise determine the load status of the edge computing cluster to be low load; when the load status of the edge computing cluster is low load, assign edge computing tasks to each edge computing node in turn; when the load status of the edge computing cluster is high load, preferentially assign edge computing tasks to the edge computing node with the highest resource score, and the resource score of each edge computing node is obtained by calculating its own resource data through the non-integer rank sum ratio method, and the resource score of the edge computing node is negatively correlated with the load level.

[0014] Compared with the related art, the edge load dynamic balancing method and edge computing system provided by the present invention adopt a dynamic strategy to allocate edge computing tasks to each edge computing node. First, the load status of the edge computing cluster is judged; when the edge computing cluster is underloaded, edge computing tasks are allocated to each edge computing node in turn, and task allocation is faster at this time; when the load status of the edge computing cluster is underloaded, edge computing tasks are allocated to the edge computing nodes with the highest resource scores first, and task allocation is more reasonable at this time, which can improve the computing efficiency of the edge computing cluster as a whole. Therefore, the present invention can select the most appropriate task allocation method according to the load status of the edge computing cluster, ensure that the computing efficiency of the edge computing cluster is in the best state, and solve the problem that the current edge load balancing method cannot dynamically allocate computing tasks to edge nodes.

[0015] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a method for dynamic edge load balancing provided in an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of an edge computing system provided in an embodiment of the present invention;

[0018] Figure 3 This is a comparison chart of load balancing strategies;

[0019] Figure 4 This is a throughput comparison chart between the AG1 group and the AG2 group at different concurrency levels in the experiment;

[0020] Figure 5 This is a comparison chart of the average response time between the AG1 group and the AG2 group at different concurrency levels in the experiment;

[0021] Figure 6 This is a comparison chart of the throughput of the AG1 group and the BG1 group under different concurrency levels in the experiment;

[0022] Figure 7 This is a comparison chart of the average response time of the AG1 group and the BG1 group under different concurrency levels in the experiment;

[0023] Figure 8 This is a throughput comparison chart between the AG2 group and the BG2 group at different concurrency levels in the experiment;

[0024] Fig. 9 This is a comparison chart of the average response time of the AG2 group and the BG2 group under different concurrency levels in the experiment. DETAILED DESCRIPTION

[0025] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0026] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0027] In an embodiment of the present invention, a method for dynamic load balancing at an edge end is provided. Figure 1 is a flow chart of a method for dynamic load balancing at an edge end provided in an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0028] Step S110, obtaining resource data and load status of each edge computing node in the edge computing cluster, and the load status of each edge computing node is determined according to its own resource data.

[0029] In some embodiments, the resource data includes multiple resource cost parameters, and the multiple resource cost parameters include multiple of CPU occupancy, RAM occupancy, disk occupancy, and communication delay. The step of determining the load state of each edge computing node includes: when at least one resource cost parameter of the edge computing node exceeds a preset threshold, the load state of the edge computing node is determined to be high load, otherwise the load state of the edge computing node is determined to be low load.

[0030] The resource cost parameters of each edge computing node mainly reflect its own load. For example, when the computing load of an edge computing node is large, at least one resource cost parameter in the edge computing node will increase. Therefore, when at least one resource cost parameter in an edge computing node exceeds the preset threshold, the edge computing node is judged to be in a high load state, otherwise the edge computing node is judged to be in a low load state. Among them, the preset threshold can be set according to the actual situation, and different resource cost parameters correspond to different preset thresholds.

[0031] In an application scenario, a central control module and multiple resource monitoring modules can be set, and each resource monitoring module corresponds to each edge computing node one by one. Each resource monitoring module is configured to: obtain the resource data of the corresponding edge computing node in real time and determine the load status of the corresponding edge computing node based on the resource data. The central control module is configured to: execute the edge load dynamic balancing method provided by the present invention, and obtain the resource data and load status of each edge computing node from the resource monitoring module.

[0032] Step S120: When the load state of at least one edge computing node is high load, the load state of the edge computing cluster is determined to be high load; otherwise, the load state of the edge computing cluster is determined to be low load.

[0033] This step is a load status determination step for the edge computing cluster. The load status determination step for the edge computing cluster can be performed periodically or when the load status of the edge computing node is updated.

[0034] Step S130: When the load state of the edge computing cluster is low load, edge computing tasks are assigned to each edge computing node in turn.

[0035] This step is the task allocation step when the load state of the edge computing cluster is low load. In this step, when there are edge computing tasks, the edge computing tasks are allocated to each edge computing node in order. For example, when the previous edge computing task is allocated to the 5th edge computing node, the next edge computing task is allocated to the 6th edge computing node. This task allocation method has the advantage of high efficiency.

[0036] Step S140, when the load state of the edge computing cluster is high load, edge computing tasks are preferentially assigned to edge computing nodes with the highest resource scores. The resource score of each edge computing node is calculated by calculating its own resource data through the non-integer rank sum ratio method. The resource score of the edge computing node is negatively correlated with the load degree.

[0037] This step is a task allocation step when the load state of the edge computing cluster is high load. Since there is at least one high-load edge computing node in the edge computing cluster, if edge computing tasks continue to be assigned to each edge computing node in turn, the edge computing task may be assigned to the high-load edge computing node, causing the load of the edge computing node to further increase, thereby reducing the edge computing efficiency. In this step, the load level of each edge computing node is measured by the resource score. The higher the resource score of an edge computing node, the lower the load of the edge computing node. By preferentially assigning edge computing tasks to the edge computing nodes with the lowest load, the computing speed of the edge computing tasks can be increased, and the computing efficiency of the edge computing cluster can be improved as a whole.

[0038] In some of the embodiments, the resource data includes multiple resource parameters, which include multiple resource cost parameters and multiple resource efficiency parameters; illustratively, the multiple resource cost parameters include multiple CPU occupancy, RAM occupancy, disk occupancy and communication delay, and the multiple resource efficiency parameters include multiple CPU specifications and RAM specifications.

[0039] The calculation steps of the resource score of each edge computing node include:

[0040] The rank corresponding to each resource parameter of each edge computing node is calculated by the first formula and the second formula.

[0041] The first formula is:

[0042]

[0043] The second formula is:

[0044]

[0045] Among them, R ij It represents the rank corresponding to the jth resource parameter of the i-th edge computing node. When the jth resource parameter is the resource cost parameter, the first formula is used to calculate R ij When the jth resource parameter is the resource benefit parameter, the second formula is used to calculate R ij ; n represents the total number of edge computing nodes, X ij represents the jth resource parameter of the i-th edge computing node. min() and max() represent the minimum and maximum functions respectively. 1j ,X 2j ,···,X nj They represent the j-th resource parameters of the 1st, 2nd, up to the nth edge computing nodes respectively.

[0046] The above two formulas can be used to calculate the rank corresponding to each resource parameter of each edge computing node. The specific formula used is determined according to the type of resource parameter.

[0047] Calculate the rank sum ratio (RSR) value of each edge computing node:

[0048]

[0049] Among them, WRSR i It represents the rank sum ratio (RSR) value of the i-th edge computing node, and m represents the total number of resource parameters.

[0050] In summary, the edge load dynamic balancing method provided by the present invention adopts a dynamic strategy to allocate edge computing tasks to each edge computing node. First, the load status of the edge computing cluster is judged; when the edge computing cluster is underloaded, edge computing tasks are allocated to each edge computing node in turn, and task allocation is faster at this time; when the load status of the edge computing cluster is underloaded, edge computing tasks are allocated to the edge computing nodes with the highest resource scores first, and task allocation is more reasonable at this time, which can improve the computing efficiency of the edge computing cluster as a whole. Therefore, the present invention can select the most appropriate task allocation method according to the load status of the edge computing cluster, ensure that the computing efficiency of the edge computing cluster is in the best state, and solve the problem that the current edge load balancing method cannot dynamically allocate computing tasks to edge nodes.

[0051] An embodiment of the present invention further provides an edge computing system, including an edge computing cluster, a central control module and multiple resource monitoring modules, wherein the edge computing cluster includes multiple edge computing nodes.

[0052] Each resource monitoring module corresponds to each edge computing node one by one, and each resource monitoring module is configured to: obtain resource data of the corresponding edge computing node in real time and determine the load status of the corresponding edge computing node based on the resource data.

[0053] In some embodiments, the resource data includes multiple resource cost parameters, and the multiple resource cost parameters include multiple of CPU occupancy, RAM occupancy, disk occupancy, and communication delay. The step of determining the load state of each edge computing node includes: when at least one resource cost parameter of the edge computing node exceeds a preset threshold, the load state of the edge computing node is determined to be high load, otherwise the load state of the edge computing node is determined to be low load.

[0054] The resource cost parameters of each edge computing node mainly reflect its own load. For example, when the computing load of an edge computing node is large, at least one resource cost parameter in the edge computing node will increase. Therefore, when at least one resource cost parameter in an edge computing node exceeds the preset threshold, the edge computing node is judged to be in a high load state, otherwise the edge computing node is judged to be in a low load state. Among them, the preset threshold can be set according to the actual situation, and different resource cost parameters correspond to different preset thresholds.

[0055] The central control module is configured to: obtain the resource data and load status of each edge computing node from each resource monitoring module; when there is at least one edge computing node with a high load, the load status of the edge computing cluster is determined to be high load, otherwise the load status of the edge computing cluster is determined to be low load; when the load status of the edge computing cluster is low load, edge computing tasks are assigned to each edge computing node in turn; when the load status of the edge computing cluster is high load, edge computing tasks are assigned to the edge computing node with the highest resource score first, and the resource score of each edge computing node is calculated by the non-integer rank sum ratio method on its own resource data, and the resource score of the edge computing node is negatively correlated with the load degree.

[0056] Specifically, the central control module is configured to execute the edge load dynamic balancing method provided in the present invention, so the functional introduction of the central control module can refer to the above description of the edge load dynamic balancing method.

[0057] In summary, the edge computing system provided by the present invention first sets a central control module and a resource monitoring module, and the central control module uses a dynamic strategy to assign edge computing tasks to each edge computing node. First, the load state of the edge computing cluster is judged; when the edge computing cluster is underloaded, edge computing tasks are assigned to each edge computing node in turn, and task allocation is faster at this time; when the load state of the edge computing cluster is underloaded, edge computing tasks are assigned to the edge computing nodes with the highest resource scores first, and task allocation is more reasonable at this time, which can improve the computing efficiency of the edge computing cluster as a whole. Therefore, the present invention can select the most appropriate task allocation method according to the load state of the edge computing cluster to ensure that the computing efficiency of the edge computing cluster is in the best state.

[0058] As follows, the edge end load dynamic balancing method provided by the present invention is further described through a specific embodiment.

[0059] In a specific embodiment, the edge end load dynamic balancing method includes:

[0060] Build an edge computing system, ensure normal communication between edge computing nodes, and deploy control modules on each node, which include the load balancing central control module Center Controller and the edge node resource monitoring module Node Monitor. The structure is as follows: Figure 2 shown.

[0061] The edge node resource monitoring module continuously obtains its own resource status Resouce and pushes it to the load balancing central control module for the load balancing central control module to make load decisions. When the edge device's request reaches the periphery of the edge node, the load balancing central control module first receives the request, then the load balancing module selects the target node according to the current load balancing strategy, and then forwards the request to the target node. After the target node processes the request, it returns the data to the device.

[0062] The load balancing strategy of this embodiment is similar to the native load balancing strategy. Figure 3 (Native strategy (left), strategy of the present invention (right)). The strategy of the present invention is a dynamic strategy, which is divided into a polling processing strategy and an optimal node processing strategy. When the edge device requests are small and the edge cluster is in a low-load state, the load balancing central control module switches the load balancing strategy to a round-robin processing strategy, and each node takes turns to process edge device requests. When each edge node resource monitoring module reports resource data, it will be accompanied by a load status flag. This flag is calculated by the edge node resource monitoring module based on its own resource data at each moment. When the load balancing central control module detects that a node is in a high-load state, it switches the load balancing strategy to the optimal node processing strategy. At this time, the load balancing central control module will calculate the latest resource scores of all edge nodes based on the resource data reported by the edge nodes, and each request will be assigned to the node with the highest score. When the edge computing cluster is in a high-load state, the edge computing node with the strongest computing power is allowed to process requests each time to achieve load balancing of the edge computing cluster and maximize resource utilization.

[0063] In the optimal node processing strategy of load balancing, the load balancing central control module will calculate the score of the resource data of each node. The score calculation adopts the rank sum ratio comprehensive evaluation method (RSR method). The process is as follows:

[0064] According to the purpose of the evaluation, select appropriate evaluation indicators. Distinguish whether the indicator is high-quality or low-quality. Generally, benefit-type indicators, that is, the larger the value of the indicator, the better; cost-type indicators, that is, the smaller the value of the indicator, the better. Assuming that there are 3 evaluation objects, that is, 3 edge nodes, and 6 evaluation indicators, namely, the server's CPU specifications, CPU usage, RAM specifications, RAM usage, disk occupancy, communication delay, etc., construct a data matrix:

[0065]

[0066] Calculate the rank value, sort each specific evaluation index according to its index value, and obtain the rank R, and use the rank R to replace the original evaluation index value. Because each index has its own weight in actual situations, this embodiment adopts the non-integer rank sum ratio method to improve the shortcomings of the RSR method ranking method. There is a quantitative linear correspondence between the compiled rank and the original index value, which overcomes the disadvantage that the quantitative information of the original index value is easily lost when the RSR method is ranked.

[0067] For efficiency indicators, such as CPU specifications, RAM specifications, etc., the ranking is constructed as follows:

[0068]

[0069] For cost-based indicators, such as CPU usage, RAM usage, disk usage, communication latency, etc., the rank is constructed as follows:

[0070]

[0071] Calculate the rank sum ratio (RSR) value. The weight can be adjusted, either manually set by the user or continuously updated by the algorithm through self-learning. The following calculation is performed according to different weights:

[0072]

[0073] The RSR value is dimensionless. The evaluation objects are directly ranked according to the RSR value, and the edge computing node with the largest RSR value is selected. This node is the best edge computing node.

[0074] For example, several indicators are selected as examples, and a data matrix is ​​first constructed as shown in the following table:

[0075] Table 1 Resource parameters of each node

[0076]

[0077]

[0078] Here, we assume three nodes. For the fixed parameters of CPU core count and RAM specification, the parameters are different, with nodes 1 and 2 having similar parameters, and node 3 having the highest parameters. For the dynamic parameters of CPU usage, RAM usage, communication delay, CPU usage, and RAM usage, the parameters are increased, with node 3 having the highest parameters.

[0079] According to the above data matrix and a set of example weight values, the non-integer rank sum ratio method is used to process the following table:

[0080] Table 2 Rank table of each resource parameter of each node

[0081]

[0082] After calculation, it can be seen that the RSR value of node 3 is the highest at this time, and node 3 is judged to be the best edge computing node. Observing the node situation, although the CPU occupancy and RAM occupancy of node 3 are the highest, node 3 has higher specifications. According to the weight calculation, node 3 has more remaining computing power, so node 3 is selected as the best edge computing node.

[0083] When the edge computing system receives a device request, it first determines whether the edge cluster is in a high-load state. If not, it adopts a polling processing strategy to forward the request to each node in turn. If the edge cluster is in a high-load state, the load balancing central control module will calculate the resource score based on the resource data of each node, and select the node with the highest resource score to process the request to ensure that the request can be completed as quickly as possible. In this mode, when the load is low, the load balancing central control module does not need to frequently collect and calculate node resource data to ensure that the device request reaches the node for processing as quickly as possible. When the load is high, the load balancing central control module calculates the node resource score to ensure that the device request reaches the best node for processing, maximizing the computing power of the edge cluster and ensuring service quality.

[0084] In summary, the above embodiment builds a cluster load balancing central control module Center Controller in the edge environment, which is responsible for node load analysis, traffic distribution, etc. This module monitors the resource usage of each node in the cluster, dynamically adjusts the cluster's load balancing strategy, and forwards requests to the node with the strongest cluster computing power at the current moment, ultimately reducing service delays and improving system resource utilization.

[0085] In addition, we build an edge node resource monitoring module, Node Monitor, which is responsible for collecting various resource data of nodes in real time, such as CPU usage, RAM usage, disk occupancy, network latency, and other parameters. NodeMonitor on each node collects various resource data of the node and provides two data sharing strategies: one is for the Center Controller to read when the cluster load is low; the other is to actively push it to the Center Controller when the cluster load is high, avoiding the extra delay of the cluster obtaining all node resource data under high load conditions.

[0086] To monitor node parameters, the full-stack monitoring component Prometheus is used to collect and store the resource utilization and other indicators of the cluster, making the overall operation status of the cluster transparent and visible, and facilitating the load balancing of the cluster.

[0087] The load balancing central control module will determine whether the cluster is in a low-load state. If the cluster is in a low-load state, a polling processing strategy will be adopted for the request, and the request will be sent to all nodes in turn. Secondly, the control module will determine whether the cluster is in a high-load state. If the cluster is in a high-load state, the optimal node processing strategy will be adopted for the request. Based on the collected resource data of all nodes, the resource scores of all nodes are calculated, and the node with the highest resource score is selected to process the request.

[0088] The high load status of the cluster is determined by the status of all nodes. For each node, the NodeMonitor of the node collects various resource data of the node (such as resource 1 ,resource 2 ,resource 3 ,…), if it detects that a resource consumption exceeds the threshold, the node will be marked as high load state (node1_overload_flag=true)(limit 1 ,limit 2 etc. are high load judgment thresholds for various resource parameters):

[0089] node1_overload_flag

[0090] =resource 1 >limit 1 ||resource 2 >limit 2 ||resouce 3 >limit 3

[0091] When the load balancing central control module obtains node resource data or pushes node resource data, if the Center Controller detects a node in a high-load state, it marks the cluster as a high-load state (center_overload_flag=true) and changes the load balancing strategy to the optimal node processing strategy.

[0092] center_overload_flag=node1_overload_flag||node2_overload_flag

[0093] The edge node resource monitoring module is responsible for real-time monitoring of various resource data of the node. The cluster's load balancing central control module performs node status analysis, dynamic selection of load balancing strategies, and traffic distribution to achieve high-performance clusters.

[0094] The edge end load dynamic balancing method in the above embodiment has the following technical effects:

[0095] 1. When the cluster is highly loaded, the optimal node processing strategy is adopted, which reduces service delay and improves system performance compared with the native load balancing strategy.

[0096] 2. Monitor the resource usage of all nodes in real time, select the optimal node to process requests, avoid abnormal backlog of requests on overloaded nodes, and improve system resource utilization.

[0097] 3. Dynamic adjustment of the load balancing strategy can not only allow the system to not make preferential judgments under low load conditions to reduce response time, but also ensure that the system selects the optimal node to process requests under high load conditions to ensure service quality.

[0098] 4. Compared with the native solution, the overall load balancing strategy can make better use of resources and achieve better results in dynamic environments with low load and high load.

[0099] As follows, a set of comparative tests are performed to verify the edge load dynamic balancing method in the above specific embodiment.

[0100] First, we test the load balancing native strategy and prepare two groups of servers. One group is a node group with the same native strategy configuration, referred to as group AG1, and the other group is a node group with different native strategy configurations, referred to as group AG2. The configuration specifications are shown in Table 3 and Table 4:

[0101] Table 3 AG1 group configuration

[0102] Serial number CPU Specifications RAM specifications disk 1 2vCPUs 4GiB 1T 2 2vCPUs 4GiB 1T 3 2vCPUs 4GiB 1T

[0103] Table 4 AG2 group configuration

[0104] Serial number CPU Specifications RAM specifications disk 1 1vCPUs 2GiB 1T 2 2vCPUs 4GiB 1T 3 4vCPUs 8GiB 1T

[0105] The performance of the AG1 and AG2 groups was tested and the results were Figure 4 , Figure 5 The results show that under low load, the throughput of the two groups is equivalent. Under high load, the throughput of group AG1 is higher than that of group AG2, because the resource configurations of the server nodes in group AG2 are different and the computing capabilities of the nodes are different. When the load gradually increases and approaches the high load state, the low-configuration node reaches the full load state first, while the high-configuration node is not fully loaded at this time. According to the load balancing native strategy, the request at this time will still be forwarded to the end of the request queue of the low-configuration node, while the high-configuration node still has a lot of computing resources that are not used, which ultimately reduces the overall throughput of the cluster. The above shows that the different node resource data of the server nodes in the edge cluster due to resource specifications and other reasons will cause the overall throughput of the cluster to decrease under high load conditions, low resource utilization, and reduce server quality.

[0106] Next, the strategy of the present invention is tested. Two groups of servers are prepared. One group is a node group with the same configuration as the strategy of the present invention, hereinafter referred to as the BG1 group, which has the same configuration as the AG1 group, and the configuration specifications are shown in Table 3. The other group is a node group with different configurations of the strategy of the present invention, hereinafter referred to as the BG2 group, which has the same configuration as the AG2 group, and the configuration specifications are shown in Table 4.

[0107] The performance of the BG1 and BG2 groups was tested and compared with the results of the AG1 and AG2 groups. Figures 6 to 9 result. Figure 6 and Figure 7 It shows that when the edge cluster node configuration specifications are the same, that is, there is no difference in the initial node resource data, it can be seen that the throughput and average response time of the AG1 group and the BG1 group are very close at low loads, and when the number of concurrent requests reaches 16, the throughput of the BG1 group is slightly higher than that of the AG1 group, and the average response time of the BG1 group is also faster than that of the AG1 group. The throughput and average response time of the BG1 group are about 10% higher than those of the AG1 group. Although the server nodes have the same configuration specifications, different device requests consume different resources, and different node resource data will appear at different times. Therefore, the resource data of each server node will be different under high load conditions. Under the scheduling of the strategy of the present invention, the optimal node can be selected to process requests, maximize resource utilization, improve throughput, and reduce response time. From Figure 8 and Fig. 9 It can be seen that for edge clusters with different configuration specifications, when the number of concurrent requests is 8, the throughput of the BG2 group begins to lead the AG2 group, and the average response time of BG2 is also better than that of the AG2 group, with an overall improvement of about 30%. When there are large differences in the resource data of each node in the edge cluster, the request distribution performed by the strategy of the present invention based on the analysis of the node resource data can greatly improve the load balancing effect and improve the service quality.

[0108] In general, under low load conditions, the strategy of the present invention is equivalent to the native strategy. Under high load conditions, if the server nodes in the edge cluster are configured the same, the throughput of the strategy of the present invention is improved by about 10% compared with the native strategy. When the server nodes in the edge cluster are configured differently, there are obvious differences in node computing power. The throughput of the strategy of the present invention is improved by about 30% compared with the native strategy. The optimal node processing strategy adopted by the present invention can forward requests to the node with the strongest computing power at the current moment, ensuring that the request is processed as quickly as possible, maximizing the use of server resources, and improving service quality.

[0109] In an embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the edge load dynamic balancing method provided by the present invention are implemented.

[0110] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0111] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

Claims

1. A method for dynamic load balancing at an edge end, characterized in that: include: Obtain resource data and load status of each edge computing node in the edge computing cluster, where the load status of each edge computing node is determined based on its own resource data; When there is at least one edge computing node whose load state is high load, the load state of the edge computing cluster is determined to be high load; otherwise, the load state of the edge computing cluster is determined to be low load; When the load state of the edge computing cluster is low load, allocating edge computing tasks to each edge computing node in turn; When the load state of the edge computing cluster is high, edge computing tasks are preferentially assigned to edge computing nodes with the highest resource scores. The resource score of each edge computing node is calculated by calculating its own resource data through the non-integer rank sum ratio method. The resource score of the edge computing node is negatively correlated with the load degree.

2. The edge load dynamic balancing method according to claim 1, characterized in that: The resource data includes multiple resource cost parameters, and the multiple resource cost parameters include multiple of CPU occupancy rate, RAM occupancy rate, disk occupancy rate and communication delay; The step of determining the load status of each edge computing node includes: When at least one of the resource cost parameters of the edge computing node exceeds a preset threshold, the load state of the edge computing node is determined to be high load, otherwise the load state of the edge computing node is determined to be low load.

3. The edge load dynamic balancing method according to claim 1, characterized in that: The resource data includes multiple resource parameters, and the multiple resource parameters include multiple resource cost parameters and multiple resource benefit parameters; The steps of calculating the resource score of each edge computing node include: Calculate the rank corresponding to each resource parameter of each edge computing node by using the first formula and the second formula; The first formula is: The second formula is: Among them, R ij It represents the rank corresponding to the jth resource parameter of the i-th edge computing node. When the jth resource parameter is the resource cost parameter, the first formula is used to calculate R ij When the jth resource parameter is the resource benefit parameter, the second formula is used to calculate R ij ; n represents the total number of edge computing nodes, X ij represents the jth resource parameter of the i-th edge computing node; Calculate the rank sum ratio (RSR) value of each edge computing node: Among them, WRSR i represents the rank sum ratio (RSR) value of the i-th edge computing node, m represents the total number of resource parameters, and W j Represents the weight of the j-th resource parameter.

4. The edge end load dynamic balancing method according to claim 3 is characterized in that: The plurality of resource cost parameters include multiple items of CPU occupancy, RAM occupancy, disk occupancy and communication delay, and the plurality of resource efficiency parameters include multiple items of CPU specifications and RAM specifications.

5. An edge computing system, characterized in that: It includes an edge computing cluster, a central control module and multiple resource monitoring modules, wherein the edge computing cluster includes multiple edge computing nodes; Each of the resource monitoring modules corresponds to each of the edge computing nodes one by one, and each of the resource monitoring modules is configured to: obtain resource data of the corresponding edge computing node in real time and determine the load state of the corresponding edge computing node according to the resource data; The central control module is configured to: obtain the resource data and load status of each edge computing node from each resource monitoring module; when there is at least one edge computing node with a high load, determine the load status of the edge computing cluster to be high load, otherwise determine the load status of the edge computing cluster to be low load; when the load status of the edge computing cluster is low load, assign edge computing tasks to each edge computing node in turn; when the load status of the edge computing cluster is high load, preferentially assign edge computing tasks to the edge computing node with the highest resource score, and the resource score of each edge computing node is obtained by calculating its own resource data through the non-integer rank sum ratio method, and the resource score of the edge computing node is negatively correlated with the load level.

6. The edge computing system according to claim 5, characterized in that: The resource data includes multiple resource cost parameters, and the multiple resource cost parameters include multiple of CPU occupancy rate, RAM occupancy rate, disk occupancy rate and communication delay; The step of determining the load status of each edge computing node includes: When at least one of the resource cost parameters of the edge computing node exceeds a preset threshold, the load state of the edge computing node is determined to be high load, otherwise the load state of the edge computing node is determined to be low load.

7. The edge computing system according to claim 5, characterized in that: The resource data includes multiple resource parameters, and the multiple resource parameters include multiple resource cost parameters and multiple resource benefit parameters; The steps of calculating the resource score of each edge computing node include: Calculate the rank corresponding to each resource parameter of each edge computing node by using the first formula and the second formula; The first formula is: The second formula is: Among them, R ij It represents the rank corresponding to the jth resource parameter of the i-th edge computing node. When the jth resource parameter is the resource cost parameter, the first formula is used to calculate R ij When the jth resource parameter is the resource benefit parameter, the second formula is used to calculate R ij ; n represents the total number of edge computing nodes, X ij represents the jth resource parameter of the i-th edge computing node; Calculate the rank sum ratio (RSR) value of each edge computing node: Among them, WRSR i It represents the rank sum ratio (RSR) value of the i-th edge computing node, and m represents the total number of resource parameters.

8. The edge computing system according to claim 7, characterized in that: The plurality of resource cost parameters include multiple items of CPU occupancy, RAM occupancy, disk occupancy and communication delay, and the plurality of resource efficiency parameters include multiple items of CPU specifications and RAM specifications.

9. The edge computing system according to claim 5, characterized in that: The resource monitoring module is also configured to actively push the load status and resource data of the corresponding edge computing node to the central control module when the load status of the corresponding edge computing node is high load; otherwise, the load status and resource data of the corresponding edge computing node are pushed to the central control module in response to the data request of the central control module.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the edge load dynamic balancing method described in any one of claims 1 to 4 are implemented.

Images