Adaptive weighted scheduling method and system based on space-based distributed satellite cluster

Abstract

The present invention provides an adaptive weight scheduling system based on a space-based distributed satellite cluster, including a dynamic link simulation module, a distributed cluster resource status real-time perception module and a distributed cluster resource scheduling module. The present invention also provides an adaptive weight scheduling method based on a space-based distributed satellite cluster based on the above system, including a distributed cluster resource status monitoring sub-method, a distributed cluster optimal scheduling sub-method and a distributed cluster balanced scheduling sub-method. The scheduling strategy is extended and optimized for the satellite dynamic link status and node resource information, and the unscheduled Pod resource object is scheduled to the optimal satellite node in the space-based distributed cluster, so as to avoid excessive use of cluster resources and tilt toward a certain type of resources, thereby more reasonably and fully utilizing the resources of the cluster.

Description

Technical Field

[0001] The present invention relates to the field of container orchestration, and in particular to an adaptive weight scheduling method and system based on a space-based distributed satellite cluster. Background Art

[0002] As satellite Internet business scenarios become increasingly complex and diverse, users have stricter requirements on application service performance. However, the current computing and storage capabilities of a single satellite node are insufficient, and problems such as high dynamic changes in link topology and high network transmission latency have greatly limited the system's ability to respond quickly in the satellite Internet environment. First, the increasing number of business functions has put intersatellite links under great traffic pressure. Satellite nodes with high dynamics and high mobility may cause computing traffic to change over time, thereby triggering changes in application function requirements in the nodes; secondly, in terms of network environment, satellite networks have the characteristics of high dynamic changes in link topology and large transmission latency, which makes computing, storage and communication network resources very scarce when providing services to the outside world, and it is possible that the service will be unavailable due to the failure of a single satellite node. Finally, the satellite Internet system has increasing requirements for user service experience, and it is necessary to build a space-based satellite distributed cluster with high reliability and high availability to provide users with diversified and stable service support.

[0003] Kubernetes is the most commonly used container cloud orchestration management system and is widely used in distributed computing scenarios. Its resource scheduling strategy first filters out nodes that do not meet the requirements based on the minimum resource requirements declared by the user, and then uses the remaining CPU utilization and remaining memory utilization of the nodes as evaluation indicators, and uses the weighted sum method to score the candidate nodes, and selects the nodes with the highest scores for deployment.

[0004] However, the space-based distributed satellite cluster container orchestration and scheduling system is different from the ground-based distributed cluster. It is limited by the highly dynamic changes in link topology and high transmission latency in the space environment, and the scarcity of resources such as computing, storage and bandwidth. Directly using the Kubernetes native resource scheduling mechanism has the following problems: it does not consider the application's demand for resources such as bandwidth and disk, and cannot reasonably schedule bandwidth- and disk-oriented applications in distributed computing scenarios; and for applications with different requirements, Kubernetes uses a fixed sum weight to score nodes, which cannot meet the personalized resource requirements of applications; it does not consider the characteristics of space-based distributed satellite clusters. If the application is deployed on an inappropriate satellite node, the frequency of expansion and contraction will increase, affecting the performance of the application and causing waste of resources, making it difficult to ensure cluster stability and service reliability.

[0005] In order to solve the problem that the Kubernetes native resource scheduling mechanism is difficult to apply to the application scenarios of space-based distributed satellite clusters due to the above-mentioned problems, the present invention proposes an adaptive weight scheduling method and system based on space-based distributed satellite clusters, which expands and optimizes the scheduling strategy according to the satellite dynamic link status and node resource information, and schedules the unscheduled Pod resource objects to the optimal satellite node in the space-based distributed cluster, avoiding excessive use of cluster resources to a certain type of resources, thereby more reasonably and fully utilizing the resources of the cluster. Summary of the invention

[0006] In order to solve the problem that the current Kubernetes native resource scheduling mechanism is difficult to apply to the high dynamic change of link topology, high transmission delay and satellite node computing, storage, network and other resource-constrained scenarios of space-based distributed satellite clusters, container resources cannot be scheduled to suitable satellite nodes, increasing the frequency of expansion and contraction, affecting application performance and causing resource waste, the present invention proposes an adaptive weight scheduling method and system based on space-based distributed satellite clusters, the specific contents are as follows:

[0007] An adaptive weighted scheduling system based on a space-based distributed satellite cluster, comprising:

[0008] The dynamic link simulation module is used to simulate the link connection status of the transmission satellite nodes under the dynamic constellation topology according to the shortest path routing algorithm, so as to simulate and model the network environment of the distributed satellite cluster;

[0009] The distributed cluster resource status real-time perception module is used to monitor the computing, storage and communication resource status information of the nodes in the cluster in real time, and provide a basis for system resource scheduling for the distributed cluster resource scheduling module;

[0010] The distributed cluster resource scheduling module is used for scheduling the Pod resource objects of the entire cluster application service. It completes the scheduling decision based on the system resource scheduling basis, schedules the unscheduled application Pod resource objects to the optimal satellite node in the cluster, and realizes system load balancing.

[0011] Furthermore, the dynamic link simulation module can also control the delay, bandwidth, packet loss rate and bit error rate between satellite nodes in the cluster in the form of IP data packets.

[0012] Furthermore, the distributed cluster resource status real-time perception module deploys a monitoring process on each node in the cluster, uses a time series database to collect and store the node's resource monitoring data, and regularly provides each node's CPU, memory, disk usage and inter-node link network information to the distributed cluster resource scheduling module as a basis for system resource scheduling.

[0013] Furthermore, the system resource scheduling is based on real-time statistical information of four types of resources including CPU, memory, disk and network.

[0014] The present invention also provides an adaptive weight scheduling method based on a space-based distributed satellite cluster, based on the above system, including the following sub-methods:

[0015] A1: Distributed cluster resource status monitoring sub-method, which is used to monitor the actual usage of computing, storage and network resources of each node in the distributed cluster. The resource monitoring process is deployed on each node, and the time series database is used to collect and store the resource monitoring data of the node. The above data is regularly provided to the adaptive weight scheduler as a scheduling basis;

[0016] A2: Distributed cluster optimal scheduling sub-method, used to filter out nodes that do not meet the requirements according to the minimum resource requirements declared by the application Pod to be scheduled, and then use the node CPU, memory, disk availability and real-time network resource load obtained from the time series database described in sub-method A1 as evaluation indicators, use weighted summation to score candidate nodes, and select the node with the highest score for Pod deployment;

[0017] A3: Distributed cluster balanced scheduling sub-method, used to disperse the physical nodes according to the Pod deployment status of the available candidate node set;

[0018] Among them, sub-method A1 is based on the distributed cluster resource status real-time perception module, and sub-method A2 and sub-method A3 are based on the distributed cluster resource scheduling module.

[0019] Further, sub-method A1 comprises the following steps:

[0020] S110: using the dynamic link simulation module to simulate the link connection state of the space-based satellite node obtained according to the shortest path routing algorithm under the dynamic constellation topology, so as to set the network condition parameters of the distributed satellite cluster;

[0021] S120: The resource monitoring process of each node in the distributed cluster periodically collects the computing, storage and network resource load information of the current node;

[0022] S130: The master node server of the distributed cluster periodically queries each working node and collects the CPU, memory, disk and network resource load information of all nodes within the monitoring period, and stores the resource monitoring data in the time series database.

[0023] Further, sub-method A2 comprises the following steps:

[0024] S210: Perform node filtering detection, filter out nodes that do not meet the minimum resource requirements of the Pod resource object to be scheduled according to the set strategy, and obtain a set of available candidate nodes;

[0025] S220: The cluster monitoring module sends a query request to the time series database to obtain monitoring information of nodes and Pod resources to be scheduled and returns monitoring data in the current cycle;

[0026] S230: Calculate the adaptive weights of the four types of resources according to the node CPU, memory, disk and network bandwidth usage obtained in step S220;

[0027] S240: Counting the maximum and minimum values ​​of each network condition parameter in the distributed cluster, and calculating the network score of each node in the available candidate node set;

[0028] S250: weighted sum of the adaptive weights of the CPU, memory, disk and network resources obtained in step S230 and the corresponding resource scores to obtain the optimal scheduling score of the available candidate node set.

[0029] Further, sub-method A3 comprises the following steps:

[0030] S310: Count the number of Pod deployments on each node in the available node set, obtain the number of scheduled Pod resource objects on the node with the most Pod deployments, record it as the maximum number of deployed Pods, and set the maximum score of the balanced scheduling strategy;

[0031] S320: Calculate a balanced scheduling score list for each node according to the highest strategy score and the maximum number of deployed Pods obtained in step S310;

[0032] S330: Perform a weighted summation of the preferred scheduling scores and the balanced scheduling scores of the available candidate nodes to obtain a final score list of each node;

[0033] S340: Select the node with the highest score as the optimal node for this scheduling, bind the application Pod resource object to the target node, and complete the deployment of the container application.

[0034] Furthermore, the network condition parameters include network delay, packet loss rate, bit error rate, link delay, bandwidth and routing hop number parameters.

[0035] Furthermore, in step S210, the setting strategy includes port conflict, CPU, memory capacity detection and service occupancy and satellite partition visibility information.

[0036] The beneficial effects of the present invention are as follows: through the adaptive weight scheduling method and system based on the space-based distributed satellite cluster proposed by the present invention, the satellite dynamic link status and the usage of node computing and storage resources in the cluster are comprehensively considered to achieve the optimization of the cluster resource scheduling mechanism, avoid excessive resource usage to a certain type of resource, meet the personalized resource requirements of different applications, and schedule the Pod resource object to the optimal node of the distributed cluster, so as to more reasonably and fully utilize the cluster resources and improve the cluster stability and service reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The software architecture of the dynamic link simulation module in the present invention;

[0038] Figure 2 It is the distributed cluster resource status real-time perception module architecture in the present invention;

[0039] Figure 3 It is the distributed cluster resource scheduling module function in the present invention;

[0040] Figure 4 is a flow chart of the method of the present invention;

[0041] Figure 5 It is a flow chart of an embodiment of an adaptive weight scheduling method based on a space-based distributed satellite cluster of the present invention;

[0042] Figure 6 This is an execution flow chart of an embodiment of a distributed cluster resource status monitoring sub-method of the present invention;

[0043] Figure 7 This is an execution flow chart of an embodiment of a distributed cluster optimal scheduling sub-method of the present invention;

[0044] Figure 8 This is an execution flow chart of an embodiment of a distributed cluster balanced scheduling sub-method of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose and technical solution of the present invention more clear, the following is a further detailed description of an adaptive weight scheduling method and system based on a space-based distributed satellite cluster of the present invention in conjunction with the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] The functions of each module in the system of the present invention are described below in conjunction with the accompanying drawings:

[0047] The dynamic link simulation module mainly simulates the link connection status of the transmission satellite nodes under the dynamic constellation topology according to the shortest path routing algorithm, so as to simulate and model the network environment of the distributed satellite cluster, and can control the delay, bandwidth, packet loss rate and bit error rate between the satellite nodes in the cluster according to the IP data packet mode. Its software architecture is as follows Figure 1 shown.

[0048] The dynamic link simulation module is located above the driver layer. Its input is the network management control signaling frame, and its output is the NETEM processing signaling. The processing flow is as follows: after receiving the network management message, according to the command type in the signaling frame header, it enters the corresponding command processing function. The input of the function is the command content stripped from the signaling frame header; according to the processing flow of the corresponding command, the structured command content operates the corresponding NETEM instruction, and controls the network rate, delay, packet loss rate and other information according to the NETEM instruction.

[0049] The distributed cluster resource status real-time perception module is mainly responsible for real-time monitoring of the computing, storage and communication resource status information of the nodes in the cluster, using Prometheus [7] As a tool for acquiring node monitoring data. This module deploys a monitoring process on each node in the cluster, uses a time series database to collect and store node resource monitoring data, and regularly provides each node's CPU, memory, disk usage, and inter-node link network information (packet loss, latency, bandwidth, and routing hop count, etc.) to the distributed cluster resource scheduling module as a basis for system resource scheduling. The core architecture of this functional module is as follows: Figure 2 shown.

[0050] The distributed cluster resource status real-time perception module is integrated into the distributed cluster resource scheduling module in the form of microservices to complete the sending of HTTP requests and the processing of response data. First, the distributed cluster resource status real-time perception module sends a query request to obtain node resource perception information to the HTTP API interface provided by the Prometheus Server; then, after receiving the request, the master node Prometheus Server returns the corresponding resource data in JSON format to the resource status real-time perception module; finally, the resource status real-time perception module decodes the data, extracts the corresponding node monitoring information, and stores the corresponding data structure. At this point, a task of the distributed cluster resource status real-time perception module has been completed. This process will continue when the distributed cluster is powered on, providing support for cluster health status detection and resource object scheduling.

[0051] The distributed cluster resource scheduling module is responsible for the scheduling of the entire cluster application service Pod resource objects, such as Figure 3As shown in the figure, the real-time statistical information of four types of resources, namely CPU, memory, disk and network, obtained by the real-time perception module of the distributed cluster resource status is used as the basis for completing the scheduling decision, and scheduling the unscheduled application Pod resource objects to the optimal satellite node in the cluster, so as to avoid excessive use of distributed cluster resources to a certain type of resources, thereby more reasonably and fully utilizing the cluster resources, reducing the impact of network topology and high latency on the distributed cluster service performance, and achieving system load balancing.

[0052] Figure 4 It is a schematic diagram of the overall process of the present invention.

[0053] Before explaining the embodiment, we first define the basic concepts in the adaptive weight scheduling method as follows. This definition is only used to explain an embodiment of the present invention and is not limited to the present invention and this embodiment:

[0054] (1) Define resource type: Assume that there are k satellite nodes in the space-based distributed satellite cluster, denoted as node i , i = {1,2,…,k}; each node has four available resources: CPU, memory, disk and network resources, R = {cpu,mem,disk,net}; the Pod resource object to be scheduled is composed of j containers, P = {1,2,…,j}.

[0055] In order to solve the problem that the current Kubernetes native resource scheduling mechanism is difficult to apply to the high dynamic changes of link topology, high transmission delay and satellite node computing, storage, network and other resource-constrained scenarios of space-based distributed satellite clusters, container resources cannot be scheduled to suitable satellite nodes, increasing the frequency of expansion and contraction, affecting the performance of applications and causing waste of resources, the present invention proposes an adaptive weight scheduling method and system based on space-based distributed satellite clusters, designs a distributed cluster resource monitoring module to obtain real-time statistical information of four types of resources: CPU, memory, disk and network, optimizes the cluster resource scheduling mechanism according to the satellite dynamic link status and node resource information, and schedules the unscheduled Pod resource objects to the optimal satellite node in the cluster, thereby more reasonably and fully utilizing the cluster resources and improving the cluster stability and service reliability. It mainly includes three main sub-methods: distributed cluster resource status monitoring sub-method, distributed cluster optimal scheduling sub-method and distributed cluster balanced scheduling sub-method.

[0056] The so-called distributed cluster resource status monitoring sub-method refers to the monitoring function of the actual usage of computing, storage and network resources of each node in the distributed cluster, deploying resource monitoring processes on each node, and using a time series database to collect and store node resource monitoring data, and regularly providing the above data to the adaptive weight scheduling method as a scheduling basis. The specific steps are as follows:

[0057] Step S110: using a dynamic link simulator to simulate the link connection state of the space-based satellite node under the dynamic constellation topology according to the shortest path routing algorithm, so as to set the network environment delay delay(i,j), packet loss loss_rate(i,j) and link bandwidth bw(i,j) parameters of the distributed satellite cluster;

[0058] Step S120: The resource load monitor of each node in the distributed cluster periodically collects the computing, storage and network resource load information of the current node;

[0059] Step S130: The master node server of the distributed cluster periodically queries each working node and collects the CPU, memory, disk and network resource load information of all nodes within the monitoring period, and stores the resource monitoring data in the time series database.

[0060] The CPU, memory, disk and network resource load conditions of each node in the space-based distributed satellite cluster obtained in the above steps will be used as the input of the distributed cluster optimal scheduling sub-method.

[0061] The so-called distributed cluster optimal scheduling sub-method refers to the process of first filtering out nodes that do not meet the requirements based on the minimum resource requirements declared by the application Pod to be scheduled, and then using the node CPU, memory, disk availability and network resource (link delay, packet loss rate, routing hop count) real-time load obtained in step S130 as evaluation indicators, scoring the candidate nodes using a weighted summation method, and selecting the node with the highest score for Pod deployment. The specific steps are as follows:

[0062] Step S210: Perform node filtering detection, filter out nodes that do not meet the minimum resource requirements of the Pod resource object to be scheduled according to strategies such as port conflict, CPU, memory capacity detection, and service occupancy, and obtain the available candidate node set nodes avail ;

[0063] Step S220: The cluster monitoring module sends a query request for obtaining node and Pod resource monitoring information to be scheduled to the HTTP API interface provided by the distributed cluster resource status monitoring sub-method and returns the local resource occupancy rate containerLocalRate(i,c,r) of the container on each node of the monitoring data within the cycle;

[0064] Step S230: Calculate the adaptive weights w(p, r) of the four types of resources according to the node CPU, memory, disk and network bandwidth utilization rates obtained in step S220.

[0065] First, calculate the node of each node in the space-based distributed cluster iThe global proportion of the four types of resources r is nodeGlobalRate(i,r), which is calculated as follows:

[0066] nodeGlobalRate(i,r)=nodeTotal(i,r) / clusterTotal(r)

[0067] Among them, nodeTotal(i,r) represents the node node i clusterTotal(r) represents the total amount of resources r in the entire space-based distributed cluster.

[0068] Secondly, the node i Multiply the global proportion of the four types of resources r nodeGlobalRate(i,r) by the local occupancy rate of resource r of container c on the node containerLocalRate(i,c,r) to get the global occupancy rate of resource r of container c containerGlobalRate(i,c,r).

[0069] Then, the global occupancy rate containerGlobalRate(i,c,r) of the resource r in all containers of the Pod p to be scheduled is summed to obtain the global occupancy rate podGlobalRate(p,r)=∑containerGlobalRate(i,c,r) of the resource r of Pod p.

[0070] Finally, the adaptive weights w(p,r) of the four types of resources are obtained by calculating the ratio of the global occupancy rate of the resource r of the Pod p to be scheduled to the sum of the global occupancy rates of all resources in the Pod. The calculation formula is as follows:

[0071]

[0072] Step S240: Count the maximum and minimum values ​​of network delay, packet loss rate and routing hop count parameters in the distributed cluster, delay_MAX, delay_MIN, loss_rate_MAX, loss_rate_MIN, hop_count_MAX and hop_count_MIN, and calculate the network score net_score of each node in the available candidate node set i ;

[0073] First, calculate the nodes separately i The delay score delay_score i , packet loss rate score loss_rate_score i and routing hop count score hop_count_scorei , the calculation formula is as follows:

[0074] delay_score i =(delay_MAX–delay_avg i ) / (delay_MAX–delay_MIN)

[0075] loss_rate_score i =(loss_rate_MAX–loss_rate_avg i ) / (loss_rate–loss_rate

[0076] _MIN)

[0077] hop_count_score i =(hop_count_MAX–hop_count_avg i ) / (hop_count

[0078] –hop_count_MIN)

[0079] Among them, delay_avg i 、loss_rate_avg i and hop_count_avg i Respectively represent nodes i The average value of latency, packet loss rate, and routing hop count with other nodes in the distributed cluster.

[0080] Then, the latency, packet loss rate, and routing hop score are weighted and summed to obtain the network score of each available candidate node in the cluster.

[0081] net_score i =net_scoreMAX×(w delay ×delay_score i +w loss_rate ×loss_rate_score i +w hop_count ×hop_count_score i )

[0082] Generally, w delay 、w loss_rate 、w hop_count and the values ​​of net_scoreMAX are 0.4, 0.3, 0.3, and 100 respectively.

[0083] Step S250: The CPU, memory, disk and network resource adaptive weights w(p, r) and corresponding resource scores cpu_score obtained in step S230 are i 、mem_score i 、disk_score i and net_score i Weighted summation to obtain the priority scheduling score priority_score of the available candidate set i .

[0084]

[0085] The above steps obtain the set of available candidate nodes for Pod scheduling and their preferred scheduling scores, which will be used as the input of the distributed cluster balanced scheduling sub-method.

[0086] The so-called distributed cluster balanced scheduling sub-method refers to the process of distributing physical nodes according to the Pod deployment status of the available candidate node set to avoid all multi-copy Pod resource objects being scheduled to the same node with the highest preferred scheduling score. The specific steps are as follows:

[0087] Step S310: Count the number of Pod deployments on each node in the available node set (pod_count) i , get the number of scheduled Pod resource objects on the node with the most deployed Pods (podCountMAX), and then set the maximum score of the balanced scheduling strategy (spreadNodeScoreMAX);

[0088] Step S320: Calculate the balanced scheduling score list spreadScore of each node according to the highest strategy score spreadNodeScoreMAX obtained in step S310 and the maximum number of deployed Pods i ,i∈nodes avail , the calculation formula is as follows:

[0089] spreadScore i =spreadNodeScoreMAX×(podCountMAX-podCount i ) / podCountMAX

[0090] Step S330: Prioritize the available candidate nodes by scheduling priority_score i and the balance strategy score spreadScore i Perform weighted summation to obtain the final score list score for each node i .

[0091] scorei =w priority ×priority_score i +w spread ×spreadScore i

[0092] Step S340: Select the node with the highest score as the optimal node for this scheduling, bind the application Pod resource object to the target node, and complete the deployment of the container application.

[0093] The result obtained by executing the above steps is the optimal node for scheduling the application Pod. At this point, the cluster Pod resource object scheduling process is complete.

[0094] The beneficial effects of the present invention are as follows: through the adaptive weight scheduling method and system based on the space-based distributed satellite cluster proposed by the present invention, the satellite dynamic link status and the usage of node computing and storage resources in the cluster are comprehensively considered to achieve the optimization of the cluster resource scheduling mechanism, avoid excessive resource usage to a certain type of resource, meet the personalized resource requirements of different applications, and schedule the Pod resource object to the optimal node of the distributed cluster, so as to more reasonably and fully utilize the cluster resources and improve the cluster stability and service reliability.

[0095] The result obtained by executing the above steps is the optimal node for Pod scheduling of this application. At this point, the space-based distributed cluster resource scheduling process is complete.

[0096] Those skilled in the art may make various modifications to the above content without departing from the spirit and scope of the present invention as defined in the claims. Therefore, the scope of the present invention is not limited to the above description, but is determined by the scope of the claims.

Claims

1. A scheduling method for an adaptive weighted scheduling system based on a space-based distributed satellite cluster, characterized in that: The dispatch system includes: The dynamic link simulation module is used to simulate the link connection status of the transmission satellite nodes under the dynamic constellation topology according to the shortest path routing algorithm, so as to simulate and model the network environment of the distributed satellite cluster; The distributed cluster resource status real-time perception module is used to monitor the computing, storage and communication resource status information of the nodes in the cluster in real time, and provide a basis for system resource scheduling for the distributed cluster resource scheduling module; The distributed cluster resource scheduling module is used for scheduling the Pod resource objects of the entire cluster application service. It completes the scheduling decision based on the system resource scheduling basis, schedules the unscheduled application Pod resource objects to the optimal satellite node in the cluster, and realizes system load balancing. The schedule method includes the following sub-methods: A1: Distributed cluster resource status monitoring sub-method, which is used to monitor the actual usage of computing, storage and network resources of each node in the distributed cluster. The resource monitoring process is deployed on each node, and the time series database is used to collect and store the resource monitoring data of the node. The above data is regularly provided to the adaptive weight scheduler as a scheduling basis; A2: Distributed cluster optimal scheduling sub-method, used to filter out nodes that do not meet the requirements according to the minimum resource requirements declared by the application Pod to be scheduled, and then use the node CPU, memory, disk availability and real-time network resource load obtained from the time series database described in sub-method A1 as evaluation indicators, use weighted summation to score candidate nodes, and select the node with the highest score for Pod deployment; A3: Distributed cluster balanced scheduling sub-method, used to disperse the physical nodes according to the Pod deployment status of the available candidate node set; Among them, sub-method A1 is based on the distributed cluster resource status real-time perception module, and sub-method A2 and sub-method A3 are based on the distributed cluster resource scheduling module; Sub-method A3 comprises the following steps: S310: Count the number of Pod deployments on each node in the available node set, obtain the number of scheduled Pod resource objects on the node with the most Pod deployments, record it as the maximum number of deployed Pods, and set the maximum score of the balanced scheduling strategy; S320: Calculate a balanced scheduling score list for each node according to the highest strategy score and the maximum number of deployed Pods obtained in step S310; S330: Perform a weighted summation of the preferred scheduling scores and the balanced scheduling scores of the available candidate nodes to obtain a final score list of each node; S340: Select the node with the highest score as the optimal node for this scheduling, bind the application Pod resource object to the optimal node, and complete the deployment of the container application.

2. The method according to claim 1, characterized in that The dynamic link simulation module can also control the delay, bandwidth, packet loss rate and bit error rate between satellite nodes in the cluster in the form of IP data packets.

3. The method according to claim 1, characterized in that The distributed cluster resource status real-time perception module deploys a monitoring process on each node in the cluster, uses a time series database to collect and store the node's resource monitoring data, and regularly provides each node's CPU, memory, disk usage and inter-node link network information to the distributed cluster resource scheduling module as a basis for system resource scheduling.

4. The method according to claim 1, characterized in that: The system resource scheduling is based on real-time statistical information of four types of resources including CPU, memory, disk and network.

5. The method according to claim 1, characterized in that Sub-method A1 comprises the following steps: S110: using the dynamic link simulation module to simulate the link connection state of the space-based satellite node obtained according to the shortest path routing algorithm under the dynamic constellation topology, so as to set the network condition parameters of the distributed satellite cluster; S120: The resource monitoring process of each node in the distributed cluster periodically collects the computing, storage and network resource load information of the current node; S130: The master node server of the distributed cluster periodically queries each working node and collects the CPU, memory, disk and network resource load information of all nodes within the monitoring period, and stores the resource monitoring data in the time series database.

6. The method according to claim 1, characterized in that Sub-method A2 comprises the following steps: S210: Perform node filtering detection, filter out nodes that do not meet the minimum resource requirements of the Pod resource object to be scheduled according to the set strategy, and obtain a set of available candidate nodes; S220: The cluster monitoring module sends a query request to the time series database to obtain monitoring information of nodes and Pod resources to be scheduled and returns monitoring data in the current cycle; S230: Calculate the adaptive weights of the four types of resources according to the node CPU, memory, disk and network bandwidth usage obtained in step S220; S240: Counting the maximum and minimum values ​​of each network condition parameter in the distributed cluster, and calculating the network score of each node in the available candidate node set; S250: weighted sum of the adaptive weights of the CPU, memory, disk and network resources obtained in step S230 and the corresponding resource scores to obtain the optimal scheduling score of the available candidate node set.

7. The method according to any one of claims 5-6, characterized in that: The network condition parameters include network delay, packet loss rate, bit error rate, link delay, bandwidth and routing hop number parameters.

8. The method according to claim 6, characterized in that In step S210, the setting strategy includes port conflict, CPU, memory capacity detection and service occupancy and satellite partition visibility information.

Images