A distributed multi-user computing power routing strategy making method for space-based computing power network

Abstract

This invention discloses a distributed multi-user computing power routing strategy formulation method for space-based computing power networks. The method involves: first, constructing a satellite element network composed of satellites and functional modules; then, modeling user services as a Directed Acyclic Graph (DAG) model, and determining the computing strategy by mapping the DAG to the corresponding functions in the space-based computing power network; next, modeling the satellite computing power resource allocation problem when different users initiate services on different access satellites as a distributed potential game process, enabling the formulation of multi-user computing power routing strategies with minimal user losses; finally, using a distributed self-learning iterative algorithm, solving the mapping strategy of the Nash equilibrium point, and using this strategy as the optimal computing power routing strategy in the final space-based computing power network. This invention can connect satellite computing power, realize the scheduling of satellite computing power resources, and solve the problem of computing power resource preemption and conflict in multi-user, multi-task access satellite computing power networks.

Description

Technical Field

[0001] This invention relates to the field of distributed computing technology in communication technology, and in particular to a method for formulating distributed multi-user computing power routing strategies for space-based computing power networks. Background Technology

[0002] With the continuous development of information technology, people's demand for IoT services is constantly expanding; from traditional urban scenarios to rural scenarios, and further to areas that traditional communication modes cannot cover. This means that even in scenarios such as oceans, deserts, Gobi, and outer space, users expect to receive real-time, seamless, and efficient IoT services. To meet the ever-growing business needs of users, satellite communication, as a communication technology that can provide efficient coverage, has received considerable attention in recent years and is considered one of the core technologies of 6G communication. However, in traditional satellite communication scenarios, satellites are only used for transparent data forwarding. Therefore, users' IoT services need to go through multiple steps: uploading to the satellite—forwarding to the ground cloud center—returning the result to the satellite—and the satellite sending the result back to the user. This greatly increases the service latency of users' services and makes it difficult to meet their business needs.

[0003] To address this issue, the concept of "space computers" was proposed, attempting to provide computing services directly to users using satellites, significantly reducing business processing latency. However, satellites typically possess limited computing and storage capabilities, are susceptible to errors due to the space environment, and the computing and storage capacity of a single satellite is relatively weak. Therefore, the concept of space-based computing networks was further proposed, attempting to unite multiple satellites in a satellite constellation to achieve collaborative computing. Computing power routing technology is the core of this network, enabling the interconnection of satellite computing power and the scheduling of satellite computing resources. However, due to the multiple satellites initiating services, the large number of satellites, and the wide coverage area, it is difficult to find a control center to formulate computing power routing strategies, posing a significant challenge to implementing computing power routing in space-based computing networks. Traditional routing strategies and satellite collaborative computing strategies are unsuitable for decentralized, multi-user satellite scenarios. Summary of the Invention

[0004] The purpose of this invention is to provide a distributed multi-user computing power routing strategy formulation method for space-based computing power networks that can connect satellite computing power and realize the scheduling of satellite computing power resources.

[0005] The technical solution to achieve the purpose of this invention is: a method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks, comprising the following steps:

[0006] Step 1: Construct a satellite element network consisting of satellites and functional modules;

[0007] Step 2: Model the user service as a DAG model, and determine the computing strategy by mapping the DAG to the corresponding function in the space-based computing network;

[0008] Step 3: Model the satellite computing power resource allocation problem when different users initiate services on different access satellites as a decentralized potential game process, which can realize the formulation of multi-user computing power routing strategy with minimal user loss;

[0009] Step 4: Solve the mapping strategy of the Nash equilibrium point through a distributed self-learning iterative algorithm, and use the strategy at this time as the optimal computing power routing strategy in the final space-based computing power network.

[0010] Compared with the prior art, the significant advantages of this invention are: (1) In order to accurately define the specific functions of the satellite, a satellite element network that abstracts the functional modules is proposed to achieve an abstract representation of satellite computing resources, laying the groundwork for subsequent research on user service mapping; (2) This invention adopts the game theory method to realize the decentralized allocation of resources among multiple game participants, solving the problem of computing resource contention conflict in multi-user, multi-task access to the satellite computing network; (3) In order to realize the decentralized solution of the game model, a decentralized, centerless algorithm is adopted, which can effectively adapt to the centerless characteristics of the satellite network and realize the formulation of computing power routing strategy. Attached Figure Description

[0011] Figure 1 This is a flowchart of the distributed multi-user computing power routing strategy formulation method for space-based computing power networks according to the present invention.

[0012] Figure 2 This is a schematic diagram of the architecture of the space-based satellite computing network in this invention.

[0013] Figure 3 This is a topological diagram of the satellite element network in this invention.

[0014] Figure 4 This is a schematic diagram of the task model used in an embodiment of the present invention.

[0015] Figure 5 This is a schematic diagram of the time spread diagram of the low-Earth orbit satellite network within three time slots in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Combination Figure 1 This invention discloses a method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks, comprising the following steps:

[0018] Step 1: Construct a satellite element network consisting of satellites and functional modules;

[0019] Step 2: Model the user service as a DAG model, and determine the computing strategy by mapping the DAG to the corresponding function in the space-based computing network;

[0020] Step 3: Model the satellite computing power resource allocation problem when different users initiate services on different access satellites as a decentralized potential game process, which can realize the formulation of multi-user computing power routing strategy with minimal user loss;

[0021] Step 4: Solve the mapping strategy of the Nash equilibrium point through a distributed self-learning iterative algorithm, and use the strategy at this time as the optimal computing power routing strategy in the final space-based computing power network.

[0022] As a specific example, step 1, which involves constructing a satellite element network consisting of satellites and functional modules, is as follows:

[0023] Step 1.1: Determine the low-Earth orbit satellite network topology by analyzing the connectivity between satellites;

[0024] Step 1.2: Determine the satellite element network topology by analyzing the distribution of functional modules on the satellite.

[0025] As a specific example, step 1.1 describes determining the low-Earth orbit satellite network topology by analyzing the connectivity relationships between satellites, combined with... Figure 2 The details are as follows:

[0026] Based on the time-varying coordinates of each satellite, the distance between satellites is calculated, and the signal-to-noise ratio (SNR) of the channel between two satellites is calculated based on the free propagation loss of the inter-satellite link. The channel capacity between any two satellites is then obtained using Shannon's formula, thereby determining the connectivity of the inter-satellite link.

[0027] A three-dimensional coordinate system is established with the Earth's center as the origin. The coordinates of any satellite can be expressed using Kepler's six parameters P. i ={a i ,e i ,δ i ,Ω i ,ω i ,f i Determine (t)}, where a i e is the semi-major axis i For the eccentricity, δ i Ω is the tilt angle. i For the right ascension of the ascending node, ω i f is the argument of perigee. i (t) represents the true perigee angle; for satellite nodes V = {v1, v2, ..., v} in a space-based computing network, V = {v1, v2, ..., v} m,...,v M} position All of these can be obtained through three-dimensional coordinates;

[0028] According to free space loss and Shannon's formula, (v i ,v j The capacity of inter-satellite links between satellites can be expressed as:

[0029] C ij (t)=Blog2(1+EN(t))

[0030] Where B is the inter-satellite link bandwidth, and EN(t) is the signal-to-noise ratio that varies with time and is related to the distance between satellites. Let the minimum capacity for inter-satellite communication be C0, when C... ij When ≥C0, (v i ,v j The inter-satellite links between satellites must be connected and have a link capacity of C0; otherwise, the links must be disconnected.

[0031] Because satellite network topology often changes within a short period of time, I use the satellite topology at a fixed time point t to represent the network topology over a period of time. We use a 0-1 adjacency matrix L to represent this, and the element L in the matrix can be represented as:

[0032]

[0033]

[0034] Where C0 represents the capacity threshold for inter-satellite link connectivity.

[0035] As a specific example, step 1.2 describes determining the satellite element network topology by analyzing the distribution of functional modules on the satellite, combined with... Figure 3 The details are as follows:

[0036] The nodes in the spatiotemporal expansion graph of the elements are divided into satellites and functional modules. Functional modules are abstracted as nodes directly connected to satellites. Since the functional modules equipped on satellite nodes do not change over time, time slot variations do not need to be considered when discussing elements. Assume the satellite network can provide N types of functional modules, and all functional module types of the 1 ≤ x ≤ N can be represented as an array. in 1≤j≤M represents satellite v j The xth type of functional module of the equipment, This represents the satellite v occupied by the functional module of type x. j The proportion of computing resources; therefore, all functional module nodes in the satellite network can be represented by F = {f}. 1 ,f 2 ,...,fN}express;

[0037] Based on the above functional module definitions, we construct the element satellite network adjacency matrix.

[0038]

[0039] in The latency is the time required to transmit a unit of task, where To calculate the latency per unit of task, For function x in node v i The proportion of computing resources, c i For node v i Computational power; Where 0 represents that there is no time delay from function to satellite.

[0040] As a specific example, step 2 involves modeling the user service as a DAG model. The computation strategy is determined by mapping the DAG to the corresponding function in the space-based computing network, as detailed below:

[0041] User services are modeled as DAG models. By mapping the DAG to the corresponding functions in the space-based computing network, the computing strategy is determined, forming a path that connects computing resources, which is a computing power route.

[0042] The user's business is modeled as a DAG model Φ = (Ψ, Y), where In the Directed Acyclic Graph (DAG), each node represents a subtask, and Y represents the edges of the DAG, indicating the logical relationships between different subtasks. Since the path in the Tiancomputing Power Network requires only one source satellite and one destination satellite, the DAG model used always uses a single source node. and a destination node

[0043] Let the workload of Φ = (Ψ, Y) be D. The number of subtasks is D j ;make express The parent node, therefore Task quantity to be calculated It can be expressed by the following formula:

[0044]

[0045] Where χ i It is the data scaling factor, with a value of χ. i =1;

[0046] Let B:Ψ→{V,G} denote the mapping from Ψ to {V,G}, and The default mapping to the function f1 of the mission initiating satellite 1 Right now Other tasks are mapped to other nodes on the transmission path, and the shortest path between different functional nodes can be obtained through GM calculation.

[0047] As a specific example, step 3, which models the satellite computing resource allocation problem when different users initiate services on different access satellites, as a decentralized potential game process, can realize the formulation of multi-user computing power routing strategies with minimal user losses, as detailed below:

[0048] Step 3.1: Perform latency and energy consumption modeling;

[0049] Step 3.2: Establish the utility function;

[0050] Step 3.2: Constructing the game theory model and proving it.

[0051] As a specific example, step 3.1, which involves modeling latency and energy consumption, is as follows:

[0052] For node v i Access Task Φ i Then its computation delay can be expressed as

[0053]

[0054] in for The shortest path between them can be obtained through the satellite feature adjacency matrix GM. for arrive The cumulative delay, For subtasks The workload.

[0055] And for node v i The energy consumption makes χ i For node v i The set of functions participating in the calculation can be represented as

[0056]

[0057] in For node v i Idle energy consumption For node v i Energy consumption during operation.

[0058] As a specific example, the establishment of the utility function described in step 3.2 is as follows:

[0059] Set node v iThe loss function consists of time delay and energy consumption, and this loss function converts time delay and energy consumption into unitless values; let the loss function U be...

[0060] U = λ1Latency + λ2Cost

[0061] Where λ1 and λ2 are conversion factors, and Cost = Energy is related to energy consumption. Given the node mapping result X... i ={χ i ,χ -i In the case of task Φ, i , χ i For node v i The functional usage can be considered as a set of functions, χ. -i For node v i Other node function usage, including node v i The latency generated by the task computation it accesses can be expressed as

[0062]

[0063] Energy cost can be expressed as

[0064]

[0065] in, For node idle energy consumption The dynamic energy consumption of node operation, node v i The loss function can be expressed as

[0066] U i (χ i ,χ -i )=λ1T i (χ i ,χ -i )+λ2O i (χ i )

[0067] Where λ1 and λ2 are the normalized loss unit prices; for a given node v i Its goal is to minimize U. i ,Right now

[0068]

[0069] As a specific example, the construction and proof of the game theory model described in step 3.3 are as follows:

[0070] The mapping process of tasks accessed from different nodes is modeled as a game process.

[0071] G =<M,X,U>

[0072] The three elements are the user set, the strategy set, and the cost set;

[0073] Definition 1, Nash Equilibrium: For G =<M,X,U> We call A Nash equilibrium point is reached if and only if no satellite nodes change their strategies to achieve lower losses.

[0074]

[0075] Definition 2, Potential Game: In the game process G =<M,X,U> There exists a potential function P that satisfies the following condition

[0076]

[0077] Let the potential function

[0078]

[0079] Proof: First

[0080]

[0081]

[0082] Because policy changes on each node do not affect the mapping of other tasks, therefore

[0083]

[0084] In order to make full use of the computing resources on the satellite nodes, In other words, starting a function on a node will fully utilize the node's computing resources, while not starting a function will not consume energy. Therefore, the above equation can be rewritten as follows:

[0085]

[0086] Right now Established.

[0087] As a specific example, step 4 describes solving the mapping strategy for the Nash equilibrium point using a distributed self-learning iterative algorithm, and using this strategy as the optimal computing power routing strategy in the final space-based computing power network, as follows:

[0088] The above Nash equilibrium solution is solved using a self-learning-based distributed mapping algorithm. The specific algorithm steps are as follows:

[0089] (1) Initialize the empty decision matrix X, U and the satellite feature adjacency matrix GM.

[0090] (2) For iteration number k = 1, 2, 3, ...

[0091] (3) For each user i

[0092] (4) Calculate U based on X i,k

[0093] (5) If U i,k >U i,k-1

[0094] (6) Keep X constant;

[0095] (7)Else

[0096] (8) Update η, update GM;

[0097] (9) Send RTU information;

[0098] (10)Endif

[0099] (11)Endfor

[0100] (12) If no RTU information can be sent

[0101] (13)Break;

[0102] (14)Endif

[0103] (15) Endfor

[0104] (16) Output the decision matrix X at this time and calculate the computation delay of each user task.

[0105] In the above algorithm, each participating node in the game tries to achieve the lowest possible cost U. When a user provides a better response, it sends a Request to Update (RTU) message, which includes the user ID and the latest decision.

[0106] Example 1

[0107] The simulation platform used in this embodiment is MATLAB, and the simulation parameters are all based on the following references: "Xue R, Yu H, Cheng Q. Adaptive Coded Modulation based on Continuous Phase Modulation for Inter-satellite Links of Global Navigation Satellite Systems[J]. IEEE Access, 2018:1-1.", "Xu Shuang Wang Xingwei Huang Min. (2015). Capacity analysis method for MLSN based on improved DGA.10.1109 / ICNC.2015.7377993.", and "Carl EF and Richard A R. An Overview of the Iridium Low Earth Orbit."

[0108] (LEO)Satellite System. Proceedings of the IEEE National Aerospace and Electronics Conference, 1998: 152-158. This paper describes the selection of 20 satellites on Starlink to build a satellite network. Satellite v1 is the source node, and v8 is the target node. The time used is 8:00 AM. The satellite's orbital data comes from the SpaceX website. We set the satellite's computing power to be evenly distributed between 5000 MHz and 1 GHz.

[0109] The NORAD Catalog Numbers of the selected satellites in the Starlink constellation are shown in Table 1.

[0110] Table 1 Satellite Numbering Information

[0111]

[0112] The cloud processor at the AGO ground station in the United States has a computing power of 12 GHz, such as Figure 4 As shown, this paper adopts a serial DAG. Each satellite has ten functional modules, from Function 1 to Function 10, in the initial stage. Figure 5 As shown, compared with cloud computing, the distributed computing power routing strategy of the space-based computing power network proposed in this paper can achieve a lower latency, and the advantages of distributed computing power routing become more obvious as the workload increases. Therefore, the distributed computing power routing strategy of the space-based computing power network is more suitable for the low latency requirements of user services.

[0113] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for formulating distributed multi-user computing power routing strategies for space-based computing power networks, characterized in that, Includes the following steps: Step 1: Construct a satellite element network consisting of satellites and functional modules; Step 2: Model the user service as a DAG model, and determine the computing strategy by mapping the DAG to the corresponding function in the space-based computing network; Step 3: Model the satellite computing resource allocation problem when different users initiate services on different access satellites as a decentralized potential game process. This enables the formulation of multi-user computing power routing strategies with minimal user losses, as detailed below: Step 3.1: Perform latency and energy consumption modeling; Step 3.2: Establish the utility function; Step 3.2: Constructing and proving the game theory model; Step 4: Solve the mapping strategy of the Nash equilibrium point through a distributed self-learning iterative algorithm, and use the strategy at this time as the optimal computing power routing strategy in the final space-based computing power network. Step 3.1, which involves modeling latency and energy consumption, is detailed below: For nodes Access Task The computational delay is expressed as: ； in for The shortest path between them, through the satellite feature adjacency matrix get; for arrive The cumulative delay, For subtasks The workload; And for nodes The energy consumption makes For nodes The set of functions participating in the calculation is represented as ； in For nodes Idle energy consumption For nodes Energy consumption during operation; Step 3.2, which involves establishing the utility function, is as follows: Set Node The loss function consists of time delay and energy consumption, and this loss function converts time delay and energy consumption into unitless values; let the loss function... for ； in and As a transformation factor, Related to energy consumption; mapping results at known nodes In the case of the task , For nodes The usage of functions, considered as a set of functions. For nodes Other node function usage, including for nodes The latency caused by the computation of the access task is expressed as ； Energy consumption Represented as ； in, For node idle energy consumption For the dynamic energy consumption of node operation, node The loss function is expressed as ； in and The normalized loss unit price; for a given node Its goal is to be as small as possible ,Right now ； The construction and proof of the game theory model described in step 3.3 are as follows: The mapping process of tasks accessed from different nodes is modeled as a game process. ； The three elements are the user set, the strategy set, and the cost set; Definition 1, Nash Equilibrium: For We call A Nash equilibrium point is reached if and only if no satellite nodes change their strategies to achieve lower losses. ； Definition 2, Potential Game: In the process of a game... There exists a potential function in The following conditions must be met ， ； Let the potential function ； Proof: First ； ； Because policy changes on each node do not affect the mapping of other tasks, therefore ； In order to make full use of the computing resources on the satellite nodes, That is, starting a function on a node will fully utilize the node's computing resources, while not starting a function will not consume energy. Therefore, the above equation can be rewritten as follows: ； Right now Established.

2. The method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks according to claim 1, characterized in that, Step 1 describes the construction of a satellite element network consisting of satellites and functional modules, as follows: Step 1.1: Determine the low-Earth orbit satellite network topology by analyzing the connectivity between satellites; Step 1.2: Determine the satellite element network topology by analyzing the distribution of functional modules on the satellite.

3. The method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks according to claim 2, characterized in that, Step 1.1, which involves analyzing the connectivity between satellites to determine the low-Earth orbit satellite network topology, is detailed below: The distance between satellites is calculated based on the time-varying coordinates of each satellite, and the signal-to-noise ratio of the channel between the two satellites is calculated based on the free propagation loss of the inter-satellite link. The channel capacity between any two satellites can be obtained using Shannon's formula, thus determining the connectivity of the inter-satellite link: A three-dimensional coordinate system is established with the Earth's center as the origin. The coordinates of any satellite can be determined using the six Kepler parameters. To determine, among which For the semi-major axis, For eccentricity, It is the angle of inclination. Right ascension of the ascending node, The argument of perigee. This is the true perigee angle; for satellite nodes in a space-based computing network. Location All of these can be obtained through three-dimensional coordinates; Based on free space loss and Shannon's formula The inter-satellite link capacity is expressed as: ； in This refers to the inter-satellite link bandwidth. It is the signal-to-noise ratio that varies over time. The value is related to the distance between satellites; let the minimum capacity for inter-satellite communication be... ,when hour, The inter-satellite links between them are in a connected state, and the link capacity is [missing information]. Otherwise, the link will be disconnected. Because satellite network topology often changes within a short period of time, a fixed time point is used. The satellite topology is used to represent the network topology over a period of time, using a 0-1 adjacency matrix. To represent, the elements in the matrix Represented as: ； ； in These are the capacity thresholds for inter-satellite link connectivity.

4. The method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks according to claim 2, characterized in that, Step 1.2, which involves analyzing the distribution of functional modules on the satellite to determine the satellite element network topology, is detailed below: The nodes in the spatiotemporal extension graph of elements are divided into satellites and functional modules. Functional modules are abstracted as nodes directly connected to satellites. Since the functional modules equipped on satellite nodes do not change over time, time slot variations do not need to be considered when discussing elements. Assume the satellite network provides... Type of functional module, all of the first The functional module type is represented as an array. ,in Representative satellite The equipment Type of functional module, Indicates the first The type of functional module occupies the satellite The proportion of computing resources is such that all functional module nodes in the satellite network use... express; Based on the above functional module definitions, construct the element satellite network adjacency matrix. ； in , The latency is the time required to transmit a unit of task, where , To calculate the latency per unit of task, For function x in node The proportion of computing resources. For nodes Computational power; , where 0 represents that no time delay is required from function to satellite.

5. The method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks according to claim 1, characterized in that, Step 2 involves modeling user services as a DAG model. The computation strategy is determined by mapping the DAG to the corresponding functions in the space-based computing network, as detailed below: User services are modeled as DAG models. By mapping the DAG to the corresponding functions in the space-based computing network, the computing strategy is determined, forming a path that connects computing resources, which is a computing power route. Model the user's business as a DAG model ,in The nodes in the DAG represent each subtask. The edges of a DAG represent the logical relationships between different subtasks; Since the path in the TianSurvey network only requires one source satellite and one destination satellite, the DAG model used always has one source node. and a destination node ; make The workload is , The number of subtasks is ; make express The parent node, therefore Task quantity to be calculated This can be expressed by the following formula: ； in It is the data scaling factor, with a value of ; make express arrive The mapping, and The default mapping is to the function of the satellite that initiates the mission. Right now This maps other tasks to other nodes on the transmission path, and the shortest path between different functional nodes is achieved through... Obtained through calculation.

6. The method for formulating a distributed multi-user computing power routing strategy for space-based computing power networks according to claim 1, characterized in that, Step 4 describes solving the mapping strategy for the Nash equilibrium point using a distributed self-learning iterative algorithm. This strategy is then used as the optimal computing power routing strategy in the final space-based computing power network, as detailed below: The above Nash equilibrium solution is solved using a self-learning-based distributed mapping algorithm. The specific algorithm steps are as follows: (1) Initialize the empty decision matrix , Adjacency matrix of satellite elements ; (2) For iteration count ; (3) For each user i; (4) According to calculate ; (5)If ; (6) Maintain constant; (7) Else; (8) Update ,renew ; (9) Send RTU information; (10) Endif (11) Endfor (12) If no RTU information can be sent (13) Break; (14)Endif (15)Endfor (16) Output the decision matrix at this time. Calculate the computation latency of each user task. In the algorithm described above, each participating node in the game strives to minimize losses. When a user provides a better response, a request for update information, or RTU information, is sent. The RTU information includes the user ID and the latest decision.

Images