A method for node load balancing deployment in space network simulation

Abstract

The present invention discloses a method for node load balancing deployment in space network simulation, belonging to the technical field of network simulation. First, a deployment model for space network simulation nodes is established, and the input, output, and resource constraint model of the node load balancing deployment method are clarified; then, the objective function of the node load balancing deployment method is constructed from the deployment model, the concept of resource matching degree is introduced to balance the load of simulation nodes, and the pheromone mechanism and probability scheduling are used to generate the solution space of the deployment method; then, for the solution strategies in the solution space, verify their satisfaction with the resource constraint model, and compare and update the optimal solution and the optimal scheme accordingly; finally, use the updated solution strategy to refresh the pheromone of the deployment method, iterate until the termination condition is reached, and output the final optimal objective function value and the optimal deployment scheme. The present invention can balance the placement of space network simulation nodes and minimize the additional resource overhead as much as possible while ensuring the progress of simulation services.

Description

Technical Field

[0001] The present invention relates to the technical field of network simulation, and more particularly to a method for deploying node load balancing for space network simulation. Background Art

[0002] In recent years, large-scale satellite constellation deployment plans have been proposed from time to time, such as the OneWeb constellation, the Starlink constellation, etc.; combined with ground nodes, the total number of nodes in the entire scenario can reach thousands or even tens of thousands. Simulating large-scale scenarios is an inevitable problem in space network simulation. However, the currently commonly used single-machine network simulation method based on virtualization technology may encounter problems of insufficient hardware resources when facing target scenarios with numerous nodes due to computer resource limitations and cannot complete the simulation. Therefore, in space network simulation, multiple servers or clusters are generally used to expand hardware resources, and a container orchestration system is generally used to manage and allocate hardware resources.

[0003] Traditional node load balancing deployment methods generally only consider the relationship between the hardware resources required by simulation nodes and the remaining hardware resources of the servers on which the simulation nodes are deployed for node placement. The widely used container orchestration system kubernetes (k8s) adopts load balancing placement strategies such as MostRequestPriority and BalancedResouceAllocation during the node placement stage, and the key factors considered are the relationships between the resources such as CPU, memory, and disk required by the simulation node containers to be placed and the remaining CPU, memory, and disk resources that can be provided by the servers. This is applicable to cloud node placement scenarios whose goal is to provide services for requests outside the cluster. However, for space network simulation scenarios, its main business is to simulate space network functions. In addition to each simulation node requiring hardware resources such as CPU, memory, and disk to ensure the normal operation of its internal processes and services, another very important aspect in simulating space network functions is the simulation of space network links. In a space network, due to the continuous movement of satellites, the connection relationships, link characteristics, network topologies, etc. between satellite nodes change over time; however, each satellite node operates periodically according to a predetermined orbit, and the visibility relationships between satellite nodes are also determined over time. Therefore, according to the connection relationships between satellite nodes, the network topology is also determined within each time slot. In order to simulate the space network links that change over time, it is necessary to establish link connections between simulation nodes within each time slot.

[0004] In spatial network simulation, a multi-layer virtual switch (OVS) is usually used to establish links and forward data between simulation nodes. Since OVS relies on the CPU to process packet switching and forwarding, the resources occupied by OVS ultimately manifest as additional occupation of the server's CPU resources. Therefore, in spatial network simulation, in addition to the CPU, memory, disk and other resources required by simulation nodes to ensure the normal operation of their internal processes and services, a large amount of additional CPU resources are occupied due to the network data interaction on the flexible spatial network links established between simulation nodes. In addition, since there are a large number of simulation links between simulation nodes in spatial network simulation, and these simulation nodes are likely to be deployed on different hosting servers, a large amount of data interaction between different servers will occur at this time, occupying the bandwidth resources of the servers. Therefore, if the original cloud node placement load balancing method is used to place spatial network simulation nodes, only considering the relationship between the needs of simulation nodes and the remaining resources of the server, while ignoring the consideration of spatial network simulation links, will inevitably cause a series of unreasonable deployment problems. Even due to improper node placement, it may affect the preset bandwidth of the simulation link, resulting in insufficient remaining resources of the server to support the continuous progress of the simulation service, thus affecting the spatial network simulation effect.

[0005] Therefore, in order to achieve accurate spatial network simulation in each time slot, in the node load balancing deployment method, it is far from enough to only consider the hardware resources required by simulation nodes and the hardware resources that the server can provide. The consideration of the additional CPU resources occupied due to the establishment of simulation links in spatial network simulation and the communication bandwidth between servers should also be added. Summary of the Invention

[0006] Aiming at the problem that the current node load balancing deployment method lacks consideration of the large amount of additional resources occupied by network data interaction on spatial network simulation links in spatial network simulation, the present invention provides a node load balancing deployment method for spatial network simulation, aiming to add the consideration of additional resource requirements caused by spatial network simulation links, balance the placement of spatial network simulation nodes, and minimize the additional overhead of all servers as much as possible while ensuring the progress of the simulation service.

[0007] The above object is achieved by the following technical solutions:

[0008] The present invention provides a node load balancing deployment method for spatial network simulation, including the following steps:

[0009] S1. Establish a deployment model for spatial network simulation nodes: Analyze the particularity of the spatial network simulation scenario, construct the basic framework of the deployment model for spatial network simulation nodes, clarify the input and output content of the node load balancing deployment method, and clarify the hardware resource constraint model.

[0010] S2. According to the space network simulation node deployment model established in step S1, construct the objective function of the space network simulation node load balancing deployment method, introduce the concept of resource matching degree to balance the simulation node load, and generate the solution space of the space network simulation node deployment method by using the pheromone mechanism and the idea of probabilistic scheduling.

[0011] S3. For the solutions in the solution space generated in step S2, verify their satisfaction with the hardware resource constraint model, and compare and update the solutions in the solution space of step S2 with the recorded optimal solutions and optimal schemes according to the results.

[0012] S4. Use the solution strategy updated in step S3 to refresh the pheromone of the space network simulation node deployment model, continue to iterate to generate the solution space until the termination condition is reached, and output the final optimal objective function value and the optimal deployment scheme.

[0013] Furthermore, the specific steps in step S1 include the following sub-steps:

[0014] S1.1 Analyze the particularity of the space network simulation scenario according to the characteristics of the space network simulation. During the space network simulation process, in addition to the necessary hardware resource requirements to maintain the normal operation of its internal services, the simulation nodes also need to simulate the space network links.

[0015] S1.2 Construct the basic framework of the space network simulation node deployment model, and give the input and output content of the node load balancing deployment method.

[0016] The space network simulation node deployment model includes a set of simulation nodes with and includes a set of servers with .

[0017] ;

[0018] ;

[0019] Among them, respectively represent the simulation nodes , respectively represent the servers .

[0020] For the simulation link situation established based on OVS between simulation nodes in the space network simulation, a simulation node connection relationship matrix is used to represent it, and its meaning is to express the connection relationship between simulation nodes:

[0021] ;

[0022] Matrix in is expressed as the link connection relationship between simulation nodes and simulation nodes . If there is no link connection between simulation node and simulation node , then is 0. If there is a link connection between simulation node and simulation node , then is 1.

[0023] Define variable to indicate whether simulation node is deployed on server . If simulation node is deployed on server , then ; otherwise it is 0. Since a simulation node can only be deployed on one server, therefore, .

[0024] The input of the deployment model is a set of simulation nodes , a set of servers and the simulation node connection relationship matrix . The output of the deployment model is the minimum value of the objective function value of the deployment plan, denoted as the optimal objective function value, and the simulation node deployment plan when the objective function value takes the minimum value, denoted as the optimal deployment plan.

[0025] S1.3 Define the resource constraint model. Specify the hardware resources that each server can provide, which are respectively expressed as a three-dimensional vector :

[0026] ;

[0027] Among them, , and are the CPU, memory, and disk resources that server can provide for simulation nodes respectively.

[0028] Since some processes are still running on the server before deploying nodes, consuming some resources, it is necessary to quantify the resources already used by the server. The resource load situation of server is defined as a three-dimensional vector :

[0029] ;

[0030] Among them, , and Is the server Used CPU, memory, and disk resources.

[0031] For each simulation node In order to maintain the normal operation of its internal business, the demand for server resources is defined as the following three-dimensional vector :

[0032] ;

[0033] in, , and Represents the simulation nodes For CPU, memory, and disk resources.

[0034] In order to achieve the purpose of load balancing, the sum of all hardware requirements of the simulation nodes loaded by each server should not exceed the available hardware resources of the server. Therefore, the final resource constraint model is expressed as the sum of the resources used by the server and the resources used for the simulation nodes cannot exceed the resources it can provide. The resource constraint model is as follows:

[0035] ;

[0036] in Indicates the preset bandwidth of the simulated link; The additional CPU resource requirements on the server to meet the preset bandwidth of the emulated link; Representation Server The available bandwidth of the network interface.

[0037] Step S2 specifically includes the following sub-steps:

[0038] S2.1 According to the spatial network simulation node deployment model in step S1, construct the objective function of the node load balancing deployment method of spatial network simulation. In order to use as few hardware resources as possible to complete the placement of simulation nodes and the construction of simulation scenarios in spatial network simulation, the final objective function can be obtained as follows: Under the premise of not exceeding the available CPU, memory, disk and network port communication bandwidth of each server, the additional CPU resource usage caused by the simulation link between simulation nodes relying on OVS is minimized. , expressed as:

[0039] .

[0040] S2.2 introduces the concept of resource matching to balance the load of simulation nodes.

[0041] The concept of resource matching degree is divided into three parts, namely CPU resource matching degree, memory resource matching degree, and disk resource matching degree.

[0042] CPU resource matching degree It is expressed as the ratio of the CPU resources that the server can provide to the CPU resource requirements of the simulation node.

[0043] ;

[0044] Among them, is the CPU resources already used by the server; is the CPU resources that the server can provide; represents the CPU resource requirements of the simulation node.

[0045] Memory resource matching degree It is expressed as the ratio of the memory resources that the server can provide to the memory resource requirements of the simulation node.

[0046] ;

[0047] Among them, is the memory resources already used by the server; is the memory resources that the server can provide; represents the memory resource requirements of the simulation node.

[0048] Disk resource matching degree It is expressed as the ratio of the disk resources that the server can provide to the disk resource requirements of the simulation node.

[0049] ;

[0050] Among them, is the disk resources already used by the server; is the disk resources that the server can provide; represents the disk resource requirements of the simulation node.

[0051] If the CPU resource matching degree, memory resource matching degree, and disk resource matching degree are weighted according to their importance, the greater the final resource matching degree, the greater the potential benefits that can be obtained by deploying the simulation node to this server, and the more likely it is to be deployed to this server. Therefore, the final resource matching degree is expressed as the weighted average of these three.

[0052] S2.3 Generate the solution space of the deployment method by using the pheromone mechanism and the idea of probabilistic scheduling. The addition of pheromone is based on the solution found by the current search individual, and the pheromone concentration on the relevant path is updated according to the quality of the solution. The pheromone concentration on the path of the better solution increases faster, so as to guide more individuals to choose these paths in the next iteration. The process of constructing a solution is a probabilistic process, and each individual in the search population selects the next path according to the comprehensive influence of the pheromone concentration on the path and the heuristic factor. During the process of constructing a solution, the simulation nodes According to the probability are scheduled to the server :

[0053] ;

[0054] where and respectively represent the pheromone and resource matching degree when the search individual deploys the simulation node to the server at the iteration number 𝑡; and are the weight parameters of the pheromone and resource matching degree respectively. The higher the pheromone, the stronger the tendency to deploy the simulation node to the server . The higher the resource matching degree, the higher the potential benefit of allocating the simulation node to the server , and the stronger the tendency

[0055] For each individual in the search population, a solution strategy will be generated according to the pheromone mechanism and probabilistic scheduling. The solution strategy includes the objective function value and the corresponding simulation node deployment plan. The set of all solution strategies constitutes the solution space of this search

[0056] Step S3 specifically includes the following sub-steps

[0057] S3.1 Constraint condition verification. For the solution space obtained in S2, it is necessary to verify whether each solution strategy in it meets the restrictions of the resource constraint model in step S1

[0058] If the simulation node deployment plan of the solution strategy does not meet the restrictions of the resource constraint model in step S1, the objective function value of this solution strategy will be modified to a penalty value, and the simulation node deployment plan of this solution strategy will be discarded. Avoid that the final optimal deployment plan does not meet the constraint conditions

[0059] S3.2 Compare and update the optimal solution and the optimal plan. If it is verified that the constructed solution strategy meets the constraint conditions, then compare the objective function value obtained by this solution strategy with the current optimal solution. If its value is less than the optimal solution, replace the objective function value obtained by this solution strategy with the optimal solution, and record the current simulation node deployment plan as the optimal plan:

[0060] ;

[0061] where represents the simulation node deployed on the server and so on. This is the simulation node deployment plan under this solution strategy.

[0062] Step S4 is as follows:

[0063] S4.1 Refresh the pheromone of the deployment model using the updated solution strategy. The pheromone update process is divided into two steps: pheromone evaporation and pheromone addition. The mathematical expression of pheromone evaporation is as follows:

[0064] ;

[0065] where, represents the pheromone concentration when the simulation node is deployed to the server at the th iteration. is the pheromone evaporation factor, usually taking values between 0 and 1. Through this operation, the pheromone concentration will be multiplied by a factor less than 1 after each iteration, so it will gradually decrease.

[0066] The addition of pheromone is to update the pheromone concentration on the relevant paths according to the solution found by the current search individual, based on the quality of the solution. The pheromone concentration on the path of a better solution increases faster, thus guiding more search individuals to choose these paths. Each search individual increases the pheromone on the path it has passed through. The mathematical expression is as follows:

[0067] ;

[0068] where: is the pheromone update constant, usually denoted as 1, represents the cost of the solution found by the search individual , numInd represents the total number of search individuals in the search population. For a better solution, the lower the cost, the greater the increment of pheromone. Therefore, it is more appropriate to use the objective function value of the solution strategy constructed by this search individual as . A lower objective function value means that this solution strategy is better.

[0069] Combining pheromone volatilization and pheromone addition, the complete pheromone update is expressed as:

[0070] 。

[0071] S4.2 Continue to iterate to generate the solution space until the termination condition is reached, and output the final optimal objective function value and the optimal deployment plan. After step S4.1 ends, according to the pheromone of the refreshed deployment method, repeat steps S2, S3, and S4.1 until the preset iteration upper limit is reached and terminated. After reaching it, output the final optimal solution as the optimal objective function value , and the corresponding optimal plan is the optimal deployment plan :

[0072] ;

[0073] where represents that the simulation node is deployed on the server and so on.

[0074] The beneficial effects of the present invention compared with the prior art are:

[0075] The present invention provides a method for node load balancing deployment in space network simulation. This method can more reasonably deploy space network simulation nodes into appropriate servers, solve problems such as insufficient remaining resources of the server in space network simulation to support the simulation service, and affect the space network simulation effect. When deploying a certain number of simulation nodes under certain server resources, while ensuring the progress of the simulation service, it can minimize the additional overhead of all servers as much as possible, improve the efficiency and reliability of the simulation service, and has certain popularization significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flowchart of the method for node load balancing deployment in space network simulation

[0077] Figure 2 is a typical structure of space network simulation

[0078] Figure 3 is a connection relationship diagram of the Iridium satellite simulation scenario in the embodiment of the present invention DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] The embodiments of the present invention will be further described below in conjunction with the drawings and specific embodiments.

[0080] Refer to the attached Figure 1, introduce the process of the node load balancing deployment method for space network simulation. The whole process can be divided into the following four steps:

[0081] S1. Establish a space network simulation node deployment model: Analyze the particularity of the space network simulation scenario, construct the basic framework of the space network simulation node deployment model, clarify the input and output content of the node load balancing deployment method, and clarify the hardware resource constraint model.

[0082] S2. According to the space network simulation node deployment model established in step S1, construct the objective function of the space network simulation node load balancing deployment method, introduce the concept of resource matching degree to balance the simulation node load, and use the pheromone mechanism and the idea of probabilistic scheduling to generate the solution space of the space network simulation node deployment method.

[0083] S3. For the solutions in the solution space generated in step S2, verify their satisfaction with the hardware resource constraint model, and compare and update the solutions in the solution space of step S2 with the recorded optimal solutions and optimal schemes according to the results.

[0084] S4. Use the solution strategy updated in step S3 to refresh the pheromone of the space network simulation node deployment model, continue to iterate to generate the solution space until the termination condition is reached, and output the final optimal objective function value and the optimal deployment scheme.

[0085] Furthermore, the specific steps in step S1 include the following sub-steps:

[0086] S1.1 According to the characteristics of space network simulation, analyze the particularity of the space network simulation scenario. During the space network simulation process, in addition to the hardware resource requirements for the simulation nodes to maintain the normal operation of their internal services, they also need to simulate the space network links. The simulation of space network links is completed by the Open vSwitch (OVS). And the OVS relies on the CPU to process the exchange and forwarding of data packets, so the resources occupied by the OVS ultimately manifest as additional occupation of the server CPU resources. Therefore, in space network simulation, in addition to the CPU, memory, disk and other resources required by the simulation nodes to meet the normal operation of their internal processes and services, due to the network data interaction on the flexibly established space network links between the simulation nodes, a large amount of additional CPU resources are occupied.

[0087] Refer to the Figure 2 typical structure of space network simulation in the appendix. Different node deployment strategies will result in different additional hardware resource overheads caused by the simulation of space network links. Specifically, there are two cases:

[0088] If the two simulation nodes at both ends of a space network simulation link are deployed on the same server (such as in the appendix Figure 2In the case of simulation node 1 and simulation node 2), the additional hardware resources required for the simulation link formed by such two simulation nodes are only the CPU resource occupancy caused by OVS forwarding once, and it does not involve the communication bandwidth occupancy between servers. Therefore, the preset bandwidth of the simulation link between simulation nodes is set to , then in order to meet this preset bandwidth, OVS needs to occupy CPU core resources. Therefore, when two simulation nodes at both ends of a simulation link are deployed on the same server, the additional hardware resource requirements for the server are CPU resources.

[0089] If two simulation nodes at both ends of a space network simulation link are deployed on different servers (such as simulation node 1 and simulation node 4 in the appendix Figure 2 ), the OVS in each server needs to forward the communication traffic of the simulation link, and the physical link connecting the two servers also needs to forward the traffic. Therefore, in order to meet the preset bandwidth of this simulation link, not only CPU core resources need to be occupied in each server, but also the communication bandwidth between servers needs to be occupied. Therefore, when two simulation nodes at both ends of a simulation link are deployed on different servers, the additional hardware resource requirements for each of the two servers where the two simulation nodes are located are CPU resources and bandwidth resources.

[0090] Therefore, in addition to the resources required to maintain its internal services, the simulation nodes in the space network simulation will also require more additional resources to meet the requirements of its space network simulation link, which is its special feature.

[0091] Referring to Figure 3 the connection relationship diagram of the Iridium network simulation scenario, the space network simulation scenario of this embodiment is the Iridium communication scenario, which includes 66 simulation nodes and belongs to 6 satellite orbits. As shown in the figure, the 6 orbits are separated by dotted lines. On each orbit, 11 Iridium satellites are running, which are the simulation nodes in this embodiment. On each orbit, adjacent satellites are connected end to end. The satellites adjacent to the left and right on different orbits are connected left and right. It should be particularly noted that due to the existence of the reverse seam, the satellites adjacent to the left and right on the leftmost orbit and the rightmost orbit, that is, the 1st orbit and the 6th orbit in the figure, do not establish link connections.

[0092] S1.2 Construct the basic framework of the space network simulation node deployment model and give the input and output contents of the node load balancing deployment method.

[0093] The space network simulation node deployment model includes a set of simulation nodes with nodes and a set of servers with servers.

[0094] ;

[0095] ;

[0096] Among them, respectively represent the simulation nodes , respectively represent the servers .

[0097] In the Iridium network simulation scenario of this embodiment, the number of simulation nodes is 66, and the number of hosting servers is 3.

[0098] For the simulation link established based on OVS between simulation nodes in space network simulation, a simulation node connection relationship matrix is used to represent. Its meaning is to express the connection relationship between simulation nodes:

[0099] ;

[0100] In the matrix , represents the link connection relationship between the simulation node and the simulation node . If there is no link connection between the simulation node and the simulation node , then is 0. If there is a link connection between the simulation node and the simulation node , then is 1.

[0101] Define a variable to represent whether the simulation node is deployed on the server . If the simulation node is deployed on the server , then ; otherwise it is 0. Since a simulation node can only be deployed on one server, therefore, .

[0102] The input of the deployment model is the set of simulation nodes , the set of servers and the simulation node connection relationship matrix The output of the deployed model is the minimum value of the objective function of the deployment plan, denoted as the optimal objective function value, and the simulation node deployment plan when the objective function value takes the minimum value, denoted as the optimal deployment plan.

[0103] S1.3 Define the resource constraint model. To meet the requirements of server load balancing, the hardware resources that each server can provide are respectively represented as a three-dimensional vector :

[0104] ;

[0105] Among them, , and are the CPU, memory, and disk resources that the server can provide for the simulation nodes respectively.

[0106] Since some processes are still running on the server before deploying the nodes, consuming some resources, it is necessary to quantify the resources already used by the server. The resource load situation of the server is defined as a three-dimensional vector :

[0107] ;

[0108] Among them, , and are the CPU, memory, and disk resources already used by the server .

[0109] For each simulation node , in order to maintain the normal operation of its internal services, the demand for server resources is defined as the following three-dimensional vector :

[0110] ;

[0111] Among them, , and respectively represent the CPU, memory, and disk resources of the simulation node .

[0112] In this embodiment, each server has 36 CPU cores, and is respectively installed with 64G of memory and 512G of disk resources. Without running other processes, these resources can almost be fully provided for space network simulation. And the fixed demand for each simulation node is 1 core, 1.5G of memory, and 3G of disk resources. After testing, in this embodiment, the preset bandwidth between simulation nodes is 100 , to reserve this bandwidth, 0.1% of the CPU core resources in the server need to be occupied. The network card of each server is a 10 Gigabit Ethernet port, which can provide 10000 bandwidth.

[0113] To achieve the purpose of load balancing, the sum of all hardware requirements of the simulation nodes loaded by each server should not exceed the available hardware resources of the server. Therefore, the final resource constraint model is expressed as that the sum of the resources used by the server and the resources used for the simulation nodes cannot exceed the resources it can provide. The resource constraint model is as follows:

[0114] ;

[0115] Among them, is the number of simulation nodes; is the number of servers carrying simulation tasks; represents the preset bandwidth of the simulation link; is the additional CPU resource requirement for the server to meet the preset bandwidth of the simulation link; represents the server available bandwidth of the network interface.

[0116] Step S2 specifically includes the following sub-steps:

[0117] S2.1 According to the spatial network simulation node deployment model in step S1, construct the objective function of the node load balancing deployment method for spatial network simulation. In order to complete the placement of simulation nodes and the construction of simulation scenarios with as few hardware resources as possible in spatial network simulation, the final goal can be obtained: that is, without exceeding the available CPU, memory, disk, and inter-server network interface communication bandwidth of each server, minimize the additional CPU resource occupancy caused by the simulation links implemented by OVS between simulation nodes , expressed as:

[0118] .

[0119] S2.2 Introduce the concept of resource matching degree to balance the load of simulation nodes. During the placement of simulation nodes, for nodes that consume more resources, place them in servers with more remaining resources, which can avoid excessive resource occupancy of a certain server and achieve the purpose of load balancing. Thus, introduce the concept of resource matching degree to balance the load of simulation nodes.

[0120] The concept of resource matching degree is divided into three parts, namely CPU resource matching degree, memory resource matching degree, and disk resource matching degree.

[0121] CPU resource matching degree It is expressed as the ratio of the CPU resources that the server can provide to the CPU resource requirements of the simulation node:

[0122] ;

[0123] Among them, is the CPU resources already used by the server ; is the CPU resources that the server can provide; represents the CPU resource requirements of the simulation node .

[0124] Memory resource matching degree It is expressed as the ratio of the memory resources that the server can provide to the memory resource requirements of the simulation node:

[0125] ;

[0126] Among them, is the memory resources already used by the server ; is the memory resources that the server can provide; represents the memory resource requirements of the simulation node .

[0127] Disk resource matching degree It is expressed as the ratio of the disk resources that the server can provide to the disk resource requirements of the simulation node:

[0128] ;

[0129] Among them, is the disk resources already used by the server ; is the disk resources that the server can provide; represents the disk resource requirements of the simulation node .

[0130] Weight the three according to a certain degree of importance. If the final resource matching degree is larger, it proves that the greater the potential benefits that the simulation node can obtain by being deployed to this server, and the more likely it is to be deployed to this server. Such a deployment plan can avoid scheduling simulation nodes with large resource requirements to servers with tight remaining resources, and balance the server load. Finally, a better load balancing effect is achieved. Therefore, the final resource matching degree is expressed as the weighted average of these three. In spatial network simulation, we pay more attention to the guarantee situation of CPU resources, so its weight is set to the highest, and memory and hard disk resources are ranked behind respectively.

[0131] In this embodiment, more emphasis is placed on the server's guarantee of the CPU resources of the simulation nodes. Therefore, its weight is set to the highest, and the memory and hard disk resources are ranked behind respectively;

[0132] 。

[0133] S2.3 Generate the solution space of the deployment method using the pheromone mechanism and the idea of probabilistic scheduling. The addition of pheromone is based on the solution found by the current search individual, and the pheromone concentration on the relevant path is updated according to the quality of the solution. The pheromone concentration on the path of the better solution increases faster, so as to guide more individuals to choose these paths in the next iteration. The process of constructing a solution is a probabilistic process. Each individual in the search population selects the next path according to the comprehensive influence of the pheromone concentration on the path and the heuristic factor. During the process of solution construction, the simulation node According to the probability is scheduled to the server :

[0134] ;

[0135] where and respectively represent the pheromone and resource matching degree when the search individual deploys the simulation node to the server at the iteration number 𝑡; and are the weight parameters of the pheromone and the resource matching degree respectively. The higher the pheromone, the stronger the tendency to deploy the simulation node to the server . The higher the resource matching degree, the higher the potential benefit of allocating the simulation node to the server , and the stronger the tendency.

[0136] In this embodiment, the weight parameter of the pheromone is set to 1. Since more attention is paid to the balanced placement of the simulation node load, the weight parameter of the resource matching degree is set to a higher 5, thereby ensuring that all servers can evenly allocate and undertake all the simulation nodes.

[0137] For each individual in the search population, a solution strategy will be generated according to the pheromone mechanism and probabilistic scheduling. The solution strategy includes the objective function value and the corresponding simulation node deployment plan. The set of all solution strategies constitutes the solution space of this search.

[0138] Step S3 specifically includes the following sub-steps:

[0139] S3.1 Constraint condition verification. For the solution space obtained in S2, it is necessary to verify whether each solution strategy therein satisfies the restrictions of the resource constraint model in step S1.

[0140] If the simulation node deployment plan of the solution strategy does not meet the restrictions of the resource constraint model in step S1, modify the objective function value of this solution strategy to a very large penalty value, and discard the simulation node deployment plan of this solution strategy. This is to prevent the final optimal deployment plan from not meeting the constraint conditions.

[0141] In this embodiment, the penalty value when the simulation node deployment plan of this solution strategy does not meet the restrictions of the resource constraint model in step S1 is set to 100,000 to ensure that the simulation node deployment plan of this solution strategy is discarded.

[0142] S3.2 Compare and update the optimal solution and the optimal plan. If it is verified that the constructed solution strategy meets the constraint conditions, compare the objective function value obtained by this solution strategy with the current optimal solution. If its value is less than the optimal solution, replace the objective function value obtained by this solution strategy with the optimal solution, and record the simulation node deployment plan at this time as the optimal plan:

[0143] ;

[0144] where represents the simulation node deployed on the server and so on. That is the simulation node deployment plan under this solution strategy.

[0145] Step S4 is as follows:

[0146] S4.1 Refresh the pheromone of the deployment method using the updated solution strategy. The pheromone update process can be divided into two steps: pheromone evaporation and pheromone addition. The evaporation process ensures that the pheromone concentration of old and suboptimal path information gradually weakens and does not affect subsequent searches. The mathematical expression of pheromone evaporation is as follows:

[0147] ;

[0148] where: represents the pheromone concentration when the simulation node is deployed to the server at the th iteration. is the pheromone evaporation factor, usually taking values between 0 and 1. Through this operation, the pheromone concentration will be multiplied by a factor less than 1 after each iteration, so it will gradually decrease.

[0149] In this embodiment, the pheromone evaporation factor Set it to 0.1, thus avoiding sudden changes in the solution strategy.

[0150] The addition of pheromone is based on the solution found by the current search individual, and the pheromone concentration on the relevant path is updated according to the quality of the solution. The pheromone concentration on the path of a better solution increases faster, thus guiding more search individuals to choose these paths. Each search individual increases the pheromone on the path it has passed through. The mathematical representation is as follows:

[0151] ;

[0152] Where: is the pheromone update constant, usually denoted as 1, represents the search individual the cost of the solution found, numInd represents the total number of search individuals in the search population. For a better solution, the lower the cost, the greater the increment of pheromone. Therefore, it is more appropriate to use the objective function value of the solution strategy constructed by this search individual as A lower objective function value means that the solution strategy is better.

[0153] Combining the above two parts, the complete pheromone update is expressed as:

[0154] .

[0155] S4.2 Continue to iteratively generate the solution space until the termination condition is reached, and output the final optimal objective function value and the optimal deployment plan. After step S4.1 ends, repeat steps S2, S3, and S4.1 according to the pheromone of the refreshed deployment method until the pre-set iteration upper limit is reached and terminated. After reaching it, output the final optimal solution as the optimal objective function value , and the corresponding optimal plan is the optimal deployment plan :

[0156] ;

[0157] Where indicates that the simulation node is deployed on the server and so on.

[0158] Finally, the result data of this embodiment is shown in Table 1:

[0159] Table 1 Result display of this embodiment

[0160]

[0161] As can be obtained from Table 1, in this embodiment, compared with the traditional method of evenly distributing and deploying nodes, the deployment method of this solution can save the additional occupancy rate of CPU resources caused by the forwarding of simulation link traffic through OVS. By comparison, the additional occupancy is reduced by approximately 16.7%. At the same time, due to the more reasonable placement of simulation nodes, the network interface communication pressure between servers is also significantly reduced, and the demand for the total simulation bandwidth shows an obvious decline.

Claims

1. A node load balancing deployment method for space network simulation, characterized in that the steps include: S1. Establish a space network simulation node deployment model: build the basic framework of the space network simulation node deployment model, clarify the input and output content of the node load balancing deployment method, and clarify the hardware resource constraint model; S2. According to the space network simulation node deployment model established in step S1, the objective function of the space network simulation node load balancing deployment method is constructed, the concept of resource matching is introduced to balance the simulation node load, and the solution space of the space network simulation node deployment method is generated by using the pheromone mechanism and the idea of ​​probability scheduling; S3. Verify the satisfaction of the hardware resource constraint model with respect to the solution in the solution space generated in step S2, and compare and update the solution in the solution space of step S2 with the recorded optimal solution and optimal solution according to the result; S4. Use the solution strategy updated in step S3 to refresh the pheromone of the spatial network simulation node deployment model, continue to iterate and generate the solution space until the termination condition is reached, and output the final optimal objective function value and the optimal deployment plan; The specific implementation steps of step S1 include: S1.1 Analyze the particularity of space network simulation scenarios based on the characteristics of space network simulation; During space network simulation, in addition to the hardware resource requirements for maintaining the normal operation of its internal business, simulation nodes also need to simulate space network links; S1.2 Construct the basic framework of the space network simulation node deployment model and provide the input and output content of the node load balancing deployment method; The space network simulation node deployment model includes n v The set of simulation nodes NODE contains n nodes s The server set SERVER of servers; in, Represents the simulation nodes Represent servers 1...n s ; For the simulation links between simulation nodes in the space network simulation based on OVS, a simulation node connection relationship matrix A is used to express the connection relationship between simulation nodes: a in matrix A ij Represented as a simulation node node i and simulation node j The link connection relationship between the simulation nodes i and simulation node j If there is no link connection between ij 0, simulation node node i and simulation node j If there is a link connection between ij is 1; Define the variable x is Represents the simulation node node i Whether to deploy on the server s If the simulation node node i Deployed on server s , then x is =1; otherwise it is 0. Since a simulation node can only be deployed on one server, The input of the deployment model is the set of simulation nodes NODE, the set of servers SERVER, and the simulation node connection relationship matrix A. The output of the deployment model is the minimum value of the objective function value of the deployment plan, recorded as the optimal objective function value, and the simulation node deployment plan when the objective function value takes the minimum value, recorded as the optimal deployment plan; S1.3 defines the resource constraint model; it specifies the hardware resources that each server s can provide, which are represented as a three-dimensional vector Among them, H s,core , H s,mem and H s,disk They are server s Able to provide CPU, memory, and disk resources for simulation nodes; Because the server is still running some processes before deploying nodes, which consumes some resources, it is necessary to quantify the resources already used by the server. s The resource load is defined as a three-dimensional vector Among them, F s,core 、F s,mem and F s,disk They are server s CPU, memory, and disk resources used; For each simulation node i In order to maintain the normal operation of its internal business, the demand for server resources is defined as the following three-dimensional vector It is expressed as the following formula: Among them, S i,core ,S i,mem and S i,disk Represents the simulation node node i For CPU, memory, and disk resources; In order to achieve the purpose of load balancing, the sum of all hardware requirements of the simulation nodes loaded by each server should not exceed the available hardware resources of the server. Therefore, the final resource constraint model is expressed as the sum of the resources used by the server and the resources used for the simulation nodes cannot exceed the resources it can provide. The resource constraint model is as follows: Among them, n v is the number of simulation nodes; n s is the number of servers carrying the simulation task; B is the preset bandwidth of the simulation link; m is the additional CPU resource requirement of the server to meet the preset bandwidth of the simulation link; B s,total Indicates server s The available bandwidth of the network interface.

2. A node load balancing deployment method for space network simulation according to claim 1, characterized in that The specific implementation steps in step S2 include: S2.1 According to the spatial network simulation node deployment model in step S1, the objective function of the node load balancing deployment method of spatial network simulation is constructed. The objective function is to reduce the extra CPU resource usage caused by the simulation link between simulation nodes relying on OVS without exceeding the available CPU, memory, disk and network port communication bandwidth of each server. cost , expressed as: S2.2 introduces the concept of resource matching to balance the load of simulation nodes. The concept of resource matching is divided into three parts: CPU resource matching, memory resource matching, and disk resource matching; CPU resource matching degree η core (t) is the ratio of the CPU resources that the server can provide to the CPU resource requirements of the simulation node: Among them, F s,core is server s CPU resources used; H s,core is server s CPU resources that can be provided; S i,core Represents the simulation node node i The demand for CPU resources; Memory resource matching degree η mem (t) is the ratio of the memory resources that the server can provide to the memory resource requirements of the simulation node: Among them, F s,mem is server s Memory resources used; H s,mem is server s Memory resources that can be provided; S i,mem Represents the simulation node node i The need for memory resources; Disk resource matching degree η mem (t) is the ratio of the disk resources that the server can provide to the disk resources required by the simulation node: Among them, F s,mem is server s Disk resources used; H s,mem is server s The disk resources that can be provided; S i,mem Represents the simulation node node i The demand for disk resources; The CPU resource matching degree, memory resource matching degree, and disk resource matching degree are weighted according to their importance. If the final resource matching degree is larger, it proves that the potential benefit of deploying the simulation node to the server is greater, and there is a greater probability that it will be deployed to the server. Therefore, the final resource matching degree η is (t) is expressed as the weighted average of these three; S2.3 uses the pheromone mechanism and the idea of ​​probabilistic scheduling to generate the solution space of the deployment method. The addition of pheromones is based on the solution found by the current search individual. The pheromone concentration on the relevant path is updated according to the quality of the solution. The pheromone concentration on the path with better solution increases faster, thereby guiding more individuals to choose these paths in the next iteration. The solution construction process is a probabilistic process. Each individual in the search population chooses the next path based on the combined influence of the pheromone concentration and the heuristic factor on the path. In the process of solution construction, the simulation node node i According to probability Dispatched to server s : Among them, τ is (t) and η is (t) respectively represent the search individual deployment simulation node node at the iteration number t i To server s α and β are the weight parameters of pheromone and resource matching, respectively. The higher the pheromone, the more simulation nodes are deployed. i To server s The stronger the tendency, the higher the resource matching degree, and the simulation node node i Assigned to server s The higher the potential benefit, the stronger the tendency; For each individual in the search population, a solution strategy will be generated based on the pheromone mechanism and probability scheduling. The solution strategy includes the objective function value and the corresponding simulation node deployment plan. The collection of all solution strategies constitutes the solution space of this search.

3. A node load balancing deployment method for space network simulation according to claim 2, characterized in that The specific implementation steps of step S3 include: S3.1 Constraint verification: For the solution space obtained in S2, it is necessary to verify whether each solution strategy satisfies the constraints of the resource constraint model in step S1; If the simulation node deployment plan of the solution strategy does not meet the restrictions of the resource constraint model in step S1, the objective function value of the solution strategy is modified to a penalty value, and the simulation node deployment plan of the solution strategy is abandoned to avoid the final optimal deployment plan not meeting the constraint conditions; S3.2 The optimal solution and the optimal plan are compared and updated; if it is verified that the constructed solution strategy meets the constraints, the objective function value obtained by this solution strategy is compared with the optimal solution at this time. If its value is less than the optimal solution, the objective function value obtained by this solution strategy is replaced as the optimal solution, and the simulation node deployment plan at this time is recorded as the optimal plan: in, Indicates that the simulation node node1 is deployed on the server server x And so on, Plan is the simulation node deployment plan under this solution strategy.

4. A node load balancing deployment method for space network simulation according to claim 1, characterized in that The specific implementation steps of step S3 include: S4.1 uses the updated solution strategy to refresh the pheromone of the deployment model. The pheromone update process is divided into two steps: pheromone volatilization and pheromone addition. The mathematical expression of pheromone volatilization is as follows: t is (t+1)=(1-ρ)τ is (t) Among them, τ is (t+1) indicates the simulation node node at iteration t+1 i Deploy to server s The pheromone concentration on is (t) represents the simulation node node when iterating t times i Deploy to server s ρ is the pheromone concentration on the graph, and ρ is the pheromone volatility factor, which takes a value between 0 and 1. Through this operation, the pheromone concentration will be multiplied by a factor less than 1 after each iteration, so it will gradually decrease; The addition of pheromones is based on the solution found by the current search individual. The pheromone concentration on the relevant path is updated according to the quality of the solution. The pheromone concentration on the path with better solution increases faster, thereby guiding more search individuals to choose these paths. Each search individual adds pheromones to the path it passes through. The mathematical expression is as follows: Where: Q is the pheromone update constant, recorded as 1, cost k represents the cost of the solution found by search individual k, and numInd represents the total number of search individuals in the search population; Combining pheromone volatilization and pheromone addition, the complete pheromone update is expressed as: S4.2 continues to iterate and generate the solution space until the termination condition is reached, and outputs the final optimal objective function value and the optimal deployment plan. After step S4.1 is completed, according to the pheromone of the refreshed deployment method, steps S2, S3 and S4.1 are repeated until the preset iteration limit is reached. After reaching it, the final optimal solution is output as the optimal objective function value. And the corresponding optimal solution is the optimal deployment solution Solution best : in Indicates that in the optimal deployment solution, the simulation node node1 is deployed on the server server x And so on.

Images