Multi-hop routing and load distribution methods, systems and storage media for space-based computing constellations

By establishing a matrix-vector multiplication computation task model on a space-based computing constellation, employing encoding splitting and distributed computing, and combining block coordinate descent and dynamic programming algorithms to optimize routing and load distribution, the problems of insufficient computing power of a single satellite and extended multi-hop communication time were solved, resulting in a significant reduction in task execution latency and an improvement in robustness.

CN119996292BActive Publication Date: 2025-12-02HARBIN INSTITUTE OF TECHNOLOGY (SHENZHEN) (INSTITUTE OF SCIENCE AND TECHNOLOGY INNOVATION HARBIN INSTITUTE OF TECHNOLOGY SHENZHEN)
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510089878.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-01-21
Publication Date
2025-12-02
Estimated Expiration
2045-01-21

AI Technical Summary

Technical Problem

Existing space-based computing constellations suffer from insufficient computing power per satellite, extended multi-hop communication time, and straggler issues when handling computationally intensive and latency-sensitive tasks. Existing routing strategies fail to fully utilize relay satellite computing resources, and load allocation strategies lack distributed computing methods to address heterogeneous constellations.

Method used

By establishing a model for performing matrix-vector multiplication computation tasks using a space-based computing constellation, and employing encoding splitting and distributed computing, combined with block coordinate descent and dynamic programming algorithms, a multi-hop routing and load distribution method is designed to optimize routing and load distribution to achieve communication and computation load balancing and resist the effects of straggler.

Benefits of technology

Significantly reduce task execution latency, improve the robustness of space-based computing constellations, and achieve a balance between computing and communication loads through on-orbit processing, thereby enhancing the execution efficiency of computationally intensive and latency-sensitive tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119996292B_ABST
    Figure CN119996292B_ABST
Patent Text Reader

Abstract

This invention provides a multi-hop routing and load distribution method, system, and storage medium for space-based computing constellations. The method includes: Step 1: Establishing a model for matrix-vector multiplication computation tasks performed by the space-based computing constellation, defining source nodes that encode and split tasks, relay nodes that transmit tasks hop-by-hop, and relay nodes and destination nodes performing distributed computation on tasks during end-to-end routing; Step 2: Establishing an optimization problem to minimize task execution latency, proposing a computing power awareness and load distribution algorithm based on block coordinate descent to decompose the original problem into routing and load distribution problems; Step 3: Designing a computing power awareness routing algorithm based on dynamic programming to select the optimal relay node, and designing a multi-hop encoding load distribution algorithm based on interior point method to achieve communication and computation load balancing. The beneficial effects of this invention are: avoiding the transmission of large amounts of raw data to ground centers, distributing task processing during end-to-end routing, and reducing task execution latency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and in particular to a multi-hop routing and load distribution method, system and storage medium for space-based computing constellations. Background Technology

[0002] Space-based computing constellations, comprising numerous low-Earth orbit (LEO) satellites, have garnered significant attention from industry and academia due to their high bandwidth, ubiquitous coverage, ubiquitous connectivity, and ubiquitous computing capabilities. For example, my country's GW constellation plan deploys 12,992 satellites at orbital altitudes of 590km, 600km, and 1145km. With the development of integrated circuits, satellites possess high on-orbit processing speeds; for instance, the Xilinx Xiphos Q8S processor can achieve a computing speed of 1800 GFLOPS. However, with the application of deep learning and convolutional neural networks in image processing, the computationally intensive tasks of space-based computing networks are becoming increasingly massive. In traditional satellite networks, satellites merely act as relays, forwarding large amounts of raw data to ground computing centers for processing. This results in communication delays of up to 16 seconds between satellites and ground stations when task arrival rates are high, making it difficult to support latency-sensitive tasks. Furthermore, space-based computing constellations suffer from problems such as insufficient computing power of individual satellites, the straggler problem caused by the heterogeneity of large-scale satellite computing power, and extended multi-hop communication delays. These issues limit the efficiency of space-based computing constellations in handling computationally intensive and latency-sensitive tasks, making it impossible to directly apply existing routing and load balancing strategies to space-based computing constellations.

[0003] Currently, many routing strategies for satellite constellations ignore the impact of computation latency on task execution latency, and load balancing strategies lack methods to address the resistance of heterogeneous space-based computing constellations to straggler in distributed computing.

[0004] Currently, there are shortcomings in the routing and load distribution methods for space-based computing constellations performing computationally intensive and latency-sensitive tasks. Regarding routing strategies, existing methods focus solely on reducing communication latency, finding only the shortest path from the source node to the node used for computation. This ignores the impact of computation latency on task execution efficiency and fails to fully utilize the idle computing resources of relay satellites, easily leading to bottlenecks due to the limited computing resources of individual satellites. As for load distribution strategies, distributed computing, through multi-satellite cooperation, can overcome the bottleneck of single-satellite computing resources, but it can easily result in straggler nodes whose computation time is significantly slower than most other nodes. Introducing computational redundancy through coding can mitigate the impact of stragglers, but it is still necessary to design a balanced coding load based on the heterogeneous computing and communication capabilities of the satellites to reduce task execution latency through redundancy. Summary of the Invention

[0005] To address the problems in the prior art, this invention provides a multi-hop routing and load distribution method for space-based computing constellations, comprising the following steps:

[0006] Step 1: Establish a model for the space-based computing constellation to perform matrix-vector multiplication calculation tasks. The source node encodes and splits the task, and the relay node transmits the task hop by hop. During the end-to-end routing process, the relay node and the destination node perform distributed computation on the task.

[0007] Step 2: Establish an optimization problem to minimize task execution latency, and propose a computing power awareness and load allocation algorithm based on block coordinate descent to decompose the original problem into routing and load allocation problems;

[0008] Step 3: Select the best relay node based on the dynamic programming-based computing power-aware routing algorithm, and design a multi-hop encoded load distribution algorithm based on the interior point method to achieve communication and computing load balancing.

[0009] As a further improvement of the present invention, step one includes:

[0010] Step S1: The source node S first selects the optimal path p. i ,use The code encodes the task matrix A into... It is then divided into H sub-matrices, where path p is assigned. i The jth u Submatrix of skipping working node u It is a matrix composed of the code combinations of each row in task matrix A. It is a positive integer. A random linear combination of L rows from task matrix A;

[0011] Step S2: The source node encodes the matrix The data is transmitted hop-by-hop through H-1 relay nodes to the destination node D. Each relay node receives the submatrix assigned to it. For submatrices not assigned to it, the relay node does not store them but sends them directly to the next hop node, i.e., a transparent transmission mechanism is adopted.

[0012] Step S3: The destination node and each relay node complete the calculation of the submatrix assigned to them until the destination node and each relay node have completed at least L results.

[0013] As a further improvement of the present invention, in step S3, the communication delay of each node receiving the allocated submatrix consists of the transmission delay and the propagation delay of the multi-hop ISL.

[0014] The sum of the computation delay, propagation delay, and transmission delay of node u is given by the following formula:

[0015]

[0016] in, Indicates the selection of path p i Time j u Total delay for hopping to satellite node u to complete the task Indicates the selection of path p i Time j u The computation time of the hopping satellite node u. Indicates the selection of path p i Time task transmission to j u The propagation delay of the hopping satellite node u, Indicates the selection of path p i The task is transmitted to the j-th node on the path. u Transmission time of hop satellite node u;

[0017] The cumulative probability distribution function of the total delay for satellite node u to complete the task is as follows:

[0018]

[0019] Where, positive number μ u The straggling parameter, a, is determined by the computing power of satellite u. u The shift time parameter required for satellite u to complete a single inner product operation. Let l represent the number of hops in the same orbit inter-satellite link and the number of hops in different orbit inter-satellite links required for transmission from source node S to node u, respectively. i,v This indicates that the path p is assigned to it. i The number of rows in the encoding matrix of the v-th hop node. Represents path p i The jth u Jump transmission rate, considering mesh satellite network topology, d v d represents the distance between inter-satellite links of adjacent satellites in the same orbit. h b represents the distance between inter-satellite links of adjacent satellites in different orbits, and b represents the encoding matrix. The number of bits in a line, where c is the speed of light;

[0020] T exe CDF is composed of each The product of is expressed by the following formula:

[0021]

[0022] T exe This represents the total latency for the system to complete the task. Indicates as of T exe The set of nodes that complete the submatrix calculation at any time;

[0023] Because of T exeSince it is a positive number, its expectation can be obtained by integrating the CDF, as shown in the following formula:

[0024]

[0025] As a further improvement of the present invention, step two further includes:

[0026] Step y1: Find the optimal path p i and optimal load distribution Minimize average total latency The problem is defined as follows:

[0027]

[0028] in It is the set of natural numbers;

[0029] Step y2: Replace the objective function with a constraint containing the expectation, as follows:

[0030]

[0031] in, Indicates the expected deadline.

[0032] Step y3: Based on the block coordinate descent method, solve the total problem of step y2. Break it down into two sub-problems, problem This is an optimal path problem, given an initial feasible total task execution delay. and load distribution strategy i(0) Find the path p that maximizes the inner product number to complete the calculation. i ,question For the load distribution problem, as a constrained optimization problem, given path p i Then, find the minimum task execution delay that guarantees task completion. and specific load balancing strategies

[0033]

[0034] By continuously executing the above steps, the algorithm will gradually converge, yielding a routing and load distribution scheme that minimizes task execution latency.

[0035] As a further improvement of the present invention, in step three, the computationally aware routing algorithm based on dynamic programming design performs the following steps:

[0036] Step 1: Input source node S, destination node D, Total inner product L, d of the task v ,dh ,R v ,R h The total delay of the result of the (r-1)th iteration Load distribution vector l i(r-1) ;

[0037] Step 2: Initial State

[0038] Step 3: According to equation (19) Calculation state And according to equation (20)

[0039] Record the optimal previous hop node for node u.

[0040] Step 4: Output the optimal path

[0041]

[0042] As a further improvement of the present invention, in step three, the multi-hop coding load allocation algorithm based on the interior point method is implemented as follows:

[0043] Step a1: Read the route obtained by the routing algorithm as the path, and read the load distribution and task execution latency of the previous iteration as the initial feasible solution;

[0044] Step a2: Initialize the number of iterations and enter the loop;

[0045] Step a3: Construct the barrier function

[0046]

[0047] Step a4: With X (k-1) ={t exe*(k-1) ,l i (k-1) The unconstrained minimum problem of minimizing the obstacle function is solved using} as the initial point, and the formula is as follows:

[0048]

[0049] Where r k It is a barrier factor;

[0050] Step a5: Determine if the solution converges. If it converges, exit the loop; otherwise, decrease the obstacle factor r. k The value is incremented by 1, and the iteration count is returned to step a3.

[0051] Step a6: Output the optimal latency and optimal load distribution.

[0052] As a further improvement of the present invention, in step one, distributed matrix multiplication is performed on the space-based computing constellation. It is assumed that H hops are required for data transmission from source node S to destination node D. The computational task y = Ax generated by the source node is distributedly computed through destination node D and H-1 relay nodes.

[0053] This invention also discloses a multi-hop routing and load distribution system for space-based computing constellations, comprising:

[0054] Module 1: This module is used to establish a model for performing matrix-vector multiplication calculations on a space-based computing constellation. It provides a model where the source node encodes and splits the task, the relay node transmits the task hop by hop, and the relay node and destination node perform distributed computation on the task during the end-to-end routing process.

[0055] Module 2: This module is used to establish an optimization problem that minimizes task execution latency. It proposes a computational power awareness and load allocation algorithm based on block coordinate descent to decompose the original problem into routing and load allocation problems.

[0056] Module 3: Used for designing computationally aware routing algorithms based on dynamic programming to select the best relay node, and for designing multi-hop encoded load distribution algorithms based on the interior point method to achieve communication and computation load balancing.

[0057] As a further improvement of the present invention, the multi-hop routing and load distribution system operates through the following steps:

[0058] Step 1: Input the space-based computing constellation parameters, including inter-satellite distance, inter-satellite transmission rate, displacement time parameters related to satellite computing power, straggler parameters, task source node, destination node, and number of task rows;

[0059] Step 2: Initialize any feasible routes and load balancing;

[0060] Step 3: Invoke the computational power-aware routing algorithm based on dynamic programming, with a fixed load distribution scheme, and search for computational power-aware routes by maximizing the number of expected encoded inner product calculations on the transmission path;

[0061] Step 4: Invoke the encoding load distribution algorithm based on the interior point method, and calculate the computation and communication load of each hop on the path according to the route obtained in Step 3;

[0062] Step 5: Calculate the task execution latency under the routing in Step 3 and the load distribution in Step 4;

[0063] Step 6: Determine whether the task execution delay has decreased compared to the previous iteration result. If it has decreased, use the result of this iteration as the starting point for the next iteration and return to step 3; otherwise, jump to step 7.

[0064] Step 7: Output computing power routing and computing communication load distribution, then end.

[0065] The present invention also discloses a computer-readable storage medium storing a computer program configured to implement the steps of the multi-hop routing and load distribution method described in the present invention when invoked by a processor.

[0066] The beneficial effects of this invention are: 1. It proposes embedding coded distributed computing into a space-based computing constellation, which avoids transmitting large amounts of raw data to the ground center and processes tasks in a distributed manner during end-to-end routing, rather than on a single satellite, thereby significantly reducing task execution latency; 2. In order to select satellites with higher computing power to participate in task computation, the computing power-aware multi-hop routing scheme simultaneously evaluates the communication and computing capabilities of relay satellites, integrating the communication and computing processes; 3. The designed load allocation scheme can resist the influence of straggler by introducing coding technology, thereby improving the robustness of the space-based computing constellation and achieving a balance between computing and communication loads through on-orbit processing of tasks. Attached Figure Description

[0067] Figure 1 This is a flowchart of the multi-hop routing and load distribution system of the present invention;

[0068] Figure 2 This is a schematic diagram of the constellation topology for space-based computing power in this invention;

[0069] Figure 3 This is a flowchart of the workflow of the space-based computing power constellation-coded distributed computing system of the present invention;

[0070] Figure 4 This is a timing diagram of the task execution delay of this invention;

[0071] Figure 5 This is a schematic diagram comparing the average task execution latency of the present invention; (a) H = 14, Scenario 1; (b) L = 2000, Scenario 1; (c) H = 14, Scenario 2; (d) L = 2000, Scenario 2. Detailed Implementation

[0072] This invention addresses the characteristics of space-based computing constellations by analyzing the latency of computing tasks performed by these constellations. It designs a distributed computing method to solve the problem of insufficient computing power of a single satellite, an encoding load distribution method to solve the straggler problem, and a computing power-aware routing method to solve the problem of multi-hop communication latency, thus laying a technical foundation for the planning of space-based computing constellations.

[0073] The core objective of this invention is to embed coded distributed computing into a space-based computing constellation, avoiding the transmission of large amounts of raw data to ground centers. Simultaneously, it encodes and breaks down computing tasks, processing them distributed throughout the end-to-end routing process, rather than on a single satellite. This improves the efficiency of the space-based computing constellation in handling computationally intensive and latency-sensitive services. The method can identify the most suitable satellite nodes for relay transmission and computation between the source and destination nodes within a given satellite network, and design specific communication and computing loads for each node, reducing the task execution latency of the space-based computing constellation.

[0074] This invention discloses a multi-hop routing and load distribution method for space-based computing constellations, comprising the following steps:

[0075] Step 1: Establish a model for the space-based computing constellation to perform matrix-vector multiplication calculation tasks. The source node encodes and splits the task, and the relay node transmits the task hop by hop. During the end-to-end routing process, the relay node and the destination node perform distributed computation on the task.

[0076] Step 2: Establish an optimization problem to minimize task execution latency, and propose a computing power awareness and load allocation algorithm based on block coordinate descent to decompose the original problem into routing and load allocation problems;

[0077] Step 3: Design a computing power-aware routing algorithm based on dynamic programming to select the best relay node, and design a multi-hop encoded load distribution algorithm based on the interior point method to achieve communication and computing load balancing.

[0078] The specific solution of this invention is as follows:

[0079] (1) System workflow diagram as follows Figure 1 As shown, the workflow is as follows:

[0080] Step 1: Input the space-based computing constellation parameters, including inter-satellite distance, inter-satellite transmission rate, displacement time parameters related to satellite computing power, straggler parameters, task source node, destination node, and number of task rows;

[0081] Step 2: Initialize any feasible routes and load balancing;

[0082] Step 3: Invoke the computational power-aware routing algorithm based on dynamic programming, with a fixed load distribution scheme, and search for computational power-aware routes by maximizing the number of expected encoded inner product calculations on the transmission path;

[0083] Step 4: Invoke the encoding load distribution algorithm based on the interior point method, and calculate the computation and communication load of each hop on the path according to the route obtained in Step 3;

[0084] Step 5: Calculate the task execution latency under the routing in Step 3 and the load distribution in Step 4;

[0085] Step 6: Determine whether the task execution delay has decreased compared to the previous iteration result. If it has decreased, use the result of this iteration as the starting point for the next iteration and return to step 3; otherwise, jump to step 7.

[0086] Step 7: Output computing power routing and computing communication load distribution.

[0087] (2) Task execution delay analysis

[0088] Consider performing distributed matrix multiplication on a space-based computing constellation. Assume that the transmission from source node S to destination node D requires H hops. The computational task y = Ax generated by the source node is distributed and computed through destination node D and H-1 relay nodes.

[0089] like Figure 2 As shown, the multi-hop network topology of the space-based computing constellation is defined as graph G = {U, E, D, R}, where U represents the endpoints of the graph, i.e., satellite nodes. For each u ∈ U, the following is defined: Let j represent the number of in-orbit inter-satellite link (ISL) hops and the number of out-of-orbit ISL hops required for transmission from source node S to node u, respectively. If the number of hops from S to node u is j... u ,but E represents an edge of the graph, i.e., ISL. Where e vu ∈E represents the ISL connecting satellite nodes v and u. Each satellite is equipped with 4 inter-satellite links: two ISLs connecting adjacent satellites in the same orbit and two ISLs connecting adjacent satellites in different orbits, forming a network as shown below. Figure 1 The topology shown. For each e vu ∈E has two attributes, namely the distance d of ISL. vu ∈D and emission rate R vu ∈R, if e vu It is an inter-satellite link in the same orbit, d vu =d v R vu =R v Otherwise d vu =d h R vu =R h .

[0090] There are a total of Q reachable paths from source node S to destination node D, and the path set is defined as follows:

[0091] P = {p1, p2, ..., p} i ,...,p Q} (1)

[0092] Clearly, each path has the same number of hops H, therefore each path has the same number of nodes involved in the computation. The set of nodes contained in the path set P is denoted as Hi.

[0093] System workflow as follows Figure 3 As shown, the specific steps are as follows:

[0094] Step S1: The source node S first selects the optimal path p. i ,use The code encodes the task matrix A into... It is then divided into H sub-matrices, where path p is assigned. i The jth u Submatrix of skipping working node u It is a matrix composed of the encoding combinations of each row in A. It is a positive integer. To assign computational tasks to each worker node, A random linear combination of L rows from matrix A is used to recover the result Ax after all nodes have completed at least L rows of inner product.

[0095] Step S2: During the task distribution phase, the source node will encode the matrix. The data is transmitted hop-by-hop through H-1 relay nodes to the destination node D. Each relay node receives the submatrix assigned to it. For submatrices not assigned to it, the relay node does not store them but sends them directly to the next-hop node, i.e., a transparent transmission mechanism is adopted.

[0096] Step S3: Calculation phase. Each destination node and each relay node completes the calculation of the submatrix assigned to them until the destination node and each relay node have completed at least L results.

[0097] The communication delay for each node to receive its assigned submatrix consists of the transmission delay and the propagation delay of the multi-hop ISL. Define variables. Indicates the selection of path p i The task is transmitted to the j-th node on the path. u Transmission time of hop satellite node u:

[0098]

[0099] Among them, R i,j Represents path p i The jth u The hop transfer rate, b represents the number of bits in a single line of data.

[0100] Define variables Package selection path p i Time task transmission to j u Propagation delay of hopping satellite node u

[0101]

[0102] Where c is the speed of light.

[0103] Define random variables Indicates the selection of path p i Time j u The computation time of the hopping satellite node u, and H random variables. They are independent of each other. Consider that the computation time of each node follows a two-parameter shift exponential distribution, i.e., the computation... The working node u of the row vector dot product, computation time The CDF is as follows:

[0104]

[0105] in positive number a u The shift time parameter required for satellite u to complete a single inner product operation, a positive number μ. u The straggling parameters are determined by the computing power of satellite u.

[0106] Define variables Indicates the selection of path p i Time j u The total delay for hopping satellite node u to complete the task can be obtained from the above model. The sum of the computation delay, transmission delay, and propagation delay of node u.

[0107]

[0108] The cumulative probability distribution function (CDF) of the total delay for satellite node u to complete the task is as follows:

[0109]

[0110] Define variable X i,j (t) represents the path p. i Time j u The number of computations completed by satellite node u by time t is considered. For a given time t, if node u has not completed its assigned computational tasks, then the computational results of node u are discarded. Therefore, X... i,j The expectation of (t) can be expressed as:

[0111]

[0112] Define random variable T exe This represents the total latency for the system to complete the task. To ensure that the source node can decode the result, T is required. exeThe total number of inner product calculations completed by the nodes that have previously completed calculations exceeds L. (Definition) A set of nodes that satisfy the following conditions

[0113]

[0114] in Represents path p i The set of nodes participating in the computation. Then T exe It can be represented as

[0115]

[0116] exist Figure 4 In the example shown, after relay node 3 completes its computation task, the cumulative L-row inner product result has been completed.

[0117] T exe CDF can be generated by each The product is represented as:

[0118]

[0119] Because of T exe Since it is a positive number, its expectation can be obtained by integrating the CDF:

[0120]

[0121] (3) Establish an optimization problem

[0122] The main problem is finding the optimal path p. i and optimal load distribution Minimize average total latency The problem is defined as follows:

[0123]

[0124] in It is the set of natural numbers.

[0125] because They are not independent and identically distributed, making it difficult to obtain... The elements in (10) and (11) are difficult to obtain. therefore, A direct solution is impossible. We consider... The approximation is achieved by replacing the objective function with a constraint containing the expectation, as follows:

[0126]

[0127] To ensure the recovery of result Ax, the expected deadline t is...exe* At least L rows of inner products must be computed by then, hence the constraint (15). Meanwhile, for large-scale matrix multiplication, A typically has a large number of rows, therefore load distribution... The value is also large, and rounding has a relatively small impact on the result. Therefore, equation (16) further... The constraints are relaxed from positive integers to positive numbers.

[0128] The number of reachable paths Q increases with the number of hops H. If load distribution l is further considered... i This will generate more feasible solutions and a larger search space. Therefore, based on the block coordinate descent method, the total problem can be transformed into a more efficient solution. Break it down into two sub-problems, problem This is an optimal path problem, given an initial feasible total task execution delay. and load distribution strategy i(0) Find the path p that maximizes the inner product number to complete the calculation. i .question For the load distribution problem, as a constrained optimization problem, given path p i Then, find the minimum task execution delay t that guarantees task completion (i.e., the source node receives at least L inner products). exe* and specific load balancing strategies

[0129]

[0130] By continuously executing the above steps, the algorithm will gradually converge, yielding a routing and load distribution scheme that minimizes task execution latency.

[0131] (4) Problem Solving

[0132] This is a multi-stage decision problem. If an exhaustive search algorithm is used, the time complexity is O(n log n). This invention proposes a dynamic programming-based algorithm that decomposes the original problem into a series of subproblems and gradually obtains the solution to the original problem by solving the subproblems. The algorithm's complexity is O(n log n).

[0133] For each relay node, there are only two candidate transmission directions: transmission via the same-track ISL or transmission via a different-track ISL. Therefore, we define all nodes on the path from source node S to node u under a given load distribution l. i Next deadline t exe* The total number of inner products that can be completed is Where node u is the j-th node on its path. u Skip the node. Then The recurrence relation is

[0134]

[0135] Where nodes v and u are paths p i The adjacent nodes are connected via inter-satellite links on the same orbit, and nodes w and u are the paths p. o Adjacent nodes connected via inter-satellite links are defined. For the optimal preceding node of node u, we have:

[0136]

[0137] Algorithm 1 traverses all Repeat equations (19) and (20) until the result is obtained. Thus, the problem is obtained. The solution and process are shown in Table 1.

[0138] Table 1. Computational Power Awareness Routing Algorithm Based on Dynamic Programming

[0139]

[0140] The load distribution methods under a given route are shown in Table 2, with a complexity of O(n). The main steps of Algorithm 2 are as follows:

[0141] Step a1: Read the route obtained from Algorithm 1 as the path, and read the load distribution and task execution latency from the previous iteration as the initial feasible solution (line 1).

[0142] Step a2: Initialize the number of iterations and enter the loop (line 2-3).

[0143] Step a3: Construct the barrier function (line 4)

[0144]

[0145] Step a4: The result of the previous iteration Solving an unconstrained minimum problem of minimizing a barrier function using the given starting point (line 5)

[0146]

[0147] Where r k It is a barrier factor.

[0148] Step a5: Determine if the solution converges. If it converges, exit the loop; otherwise, decrease the obstacle factor r. k The value is incremented by 1, and the iteration count is returned to step a3 (line 6-9).

[0149] Step a6: Output the optimal latency and optimal load distribution.

[0150] Table 2. Coding load distribution algorithm based on interior point method

[0151]

[0152] (5) Simulation experiment verification

[0153] To verify the algorithm's performance, we consider using the first phase of Starlink satellites to perform matrix multiplication operations with L rows. A total of 1584 satellites are deployed in 22 orbits, with 72 satellites in each orbit. The inter-satellite distance d between satellites in the same orbit is given. v =600km, the inter-satellite distance d of satellites with different orbits h =1577km. Inter-satellite link transmission rate set to R v =2Mb / s, R h =0.8Mb / s, encoding matrix Each row can be represented by 512 bits.

[0154] Considering two scenarios with different heterogeneous performances, the parameter settings are shown in Table 3.

[0155] Table 3 Simulation Parameter Settings

[0156]

[0157] In scenario one, the parameters are drawn with equal probability from a given finite set, while in scenario two, the parameters are generated from a uniformly distributed random source within a given range.

[0158] The comparison scheme is as follows:

[0159] 1) Random path - evenly distributed without encoding

[0160] A reachable path from a given source node to a target node is randomly selected; the task matrix A is not encoded, and the matrix size assigned to each satellite is the same, i.e.

[0161] 2) Random Path - HCMM

[0162] A reachable path from a given source node to a target node is randomly selected; the load distribution strategy is calculated by the HCMM algorithm for coded distributed computing in heterogeneous environments without considering communication.

[0163] 3) Shortest path - evenly distributed without encoding

[0164] The shortest path is calculated using Dijkstra's algorithm to ensure the lowest possible transmission and propagation delay; the task matrix A is not encoded, and the matrix size allocated to each satellite is the same, i.e.

[0165] 4) Shortest Path - HCMM

[0166] The shortest path is calculated using Dijkstra's algorithm to ensure the lowest transmission and propagation latency; the load distribution strategy is calculated using the HCMM algorithm for coded distributed computing in heterogeneous environments without considering communication.

[0167] 5) CRLA

[0168] The optimal routing and load distribution strategies obtained by executing the algorithm proposed in this invention are the optimized results derived by comprehensively considering the satellite's computing power, communication capabilities, and inter-satellite distance.

[0169] from Figure 5 (a) It can be seen that the algorithm proposed in this invention has the lowest average task execution latency compared to the other four algorithms. For example... Figure 5 As shown in (b), fixing the number of task rows and increasing the number of satellite hops allows more satellites to participate in task execution, thereby reducing the average task execution latency. This demonstrates that coded distributed computing can improve the on-orbit computing speed of space-based computing constellations. With increasing hop count, the effect of CRLA in reducing average task execution latency compared to Shortest Path-HCMM increases from 12.44% to 16.88%, then decreases again to 5.57%. This is because Shortest Path-HCMM maintains good performance when communication or computation latency accounts for a large proportion. However, since Shortest Path-HCMM does not integrate routing and load balancing, its performance is inferior to CRLA.

[0170] To compare the algorithm's performance in different scenarios, we conducted repeated experiments in the more heterogeneous scenario two. When H=14 is fixed, we can see that... Figure 5 (c) In this study, the average task execution latency of CRLA was reduced by 24.16% to 26.06% compared to the shortest path-HCMM. Figure 5 In (d), when L = 2000, the latency is reduced by 7.58% to 25.51%, peaking at 14 hops. Compared to scenario 1, the Shortest Path-HCMM algorithm shows reduced effectiveness in reducing task execution latency, indicating that CRLA has a more significant advantage in more heterogeneous scenarios. CRLA avoids selecting relay nodes with weaker computing power by jointly optimizing routing and load distribution. Meanwhile, the average hop count of the low-Earth orbit (LEO) constellation is around 14, suggesting that our algorithm can significantly reduce average task execution latency in most cases.

[0171] The computing power routing and load allocation method for space-based computing constellations proposed in this invention has the following characteristics:

[0172] 1. Joint Optimization: First, suitable satellites are identified through routing to participate in the mission. After determining the routes, specific load allocation is performed for each satellite. To address the problem of excessively large feasible regions caused by the coupling of routing and load allocation, a Computational Awareness and Load Allocation (CRLA) algorithm based on block coordinate descent is proposed, ensuring low algorithm complexity.

[0173] 2. Computational Power Awareness: By maximizing the number of coded inner product calculations expected to be completed on the transmission path, a computational power-aware routing strategy is designed. This fully considers the impact of computational latency on task execution latency, and can evaluate and select satellites with good computational and communication conditions to participate in task execution while maintaining low algorithm complexity.

[0174] 3. Load balancing: The load distribution strategy takes into account both the randomness of computation latency and the multi-hop nature of communication latency. Redundancy is introduced through coding to resist the impact of straggler slowing down distributed computing. It achieves load balancing for computing and communication in a constellation with heterogeneous computing power, and allocates more tasks to nodes with strong capabilities, thereby reducing task execution latency.

[0175] This invention addresses the need for large-scale on-orbit computing by proposing a model that embeds coded distributed computing into a space-based computing constellation. It theoretically analyzes the task execution latency of on-orbit distributed computing, establishes a latency minimization problem, and proposes a load balancing algorithm (CRLA) to solve it. Experiments verify that the proposed algorithm outperforms existing algorithms in reducing average task execution latency. Specifically, this algorithm provides a multi-hop routing and load balancing scheme for on-orbit distributed processing of computing tasks on the space-based computing constellation, and has the following innovative contributions:

[0176] 1. It is proposed to embed distributed coding computing into a space-based computing constellation, which can avoid transmitting a large amount of raw data to the ground center, and at the same time, the task is processed in a distributed manner during the end-to-end routing process, rather than on a single satellite, thereby significantly reducing the task execution latency.

[0177] 2. In order to select satellites with higher computing power to participate in mission calculations, the computing power-aware multi-hop routing scheme simultaneously evaluates the communication and computing capabilities of relay satellites, integrating the communication and computing processes.

[0178] 3. The designed load distribution scheme can resist the influence of straggler by introducing coding technology, thereby improving the robustness of the space-based computing constellation. Through on-orbit processing of tasks, the computing and communication loads are balanced.

[0179] This invention also discloses a multi-hop routing and load distribution system for space-based computing constellations, comprising:

[0180] Module 1: This module is used to establish a model for performing matrix-vector multiplication calculations on a space-based computing constellation. It provides a model where the source node encodes and splits the task, the relay node transmits the task hop by hop, and the relay node and destination node perform distributed computation on the task during the end-to-end routing process.

[0181] Module 2: This module is used to establish an optimization problem that minimizes task execution latency. It proposes a Computational Awareness and Load Allocation (CRLA) algorithm based on block coordinate descent to decompose the original problem into a routing and load allocation problem.

[0182] Module 3: Used for designing computationally aware routing algorithms based on dynamic programming to select the best relay node, and for designing multi-hop encoded load distribution algorithms based on the interior point method to achieve communication and computation load balancing.

[0183] The multi-hop routing and load balancing system operates as follows:

[0184] Step 1: Input the space-based computing constellation parameters, including inter-satellite distance, inter-satellite transmission rate, displacement time parameters related to satellite computing power, straggler parameters, task source node, destination node, and number of task rows;

[0185] Step 2: Initialize any feasible routes and load balancing;

[0186] Step 3: Invoke the computational power-aware routing algorithm based on dynamic programming, with a fixed load distribution scheme, and search for computational power-aware routes by maximizing the number of expected encoded inner product calculations on the transmission path;

[0187] Step 4: Invoke the encoding load distribution algorithm based on the interior point method, and calculate the computation and communication load of each hop on the path according to the route obtained in Step 3;

[0188] Step 5: Calculate the task execution latency under the routing in Step 3 and the load distribution in Step 4;

[0189] Step 6: Determine whether the task execution delay has decreased compared to the previous iteration result. If it has decreased, use the result of this iteration as the starting point for the next iteration and return to step 3; otherwise, jump to step 7.

[0190] Step 7: Output computing power routing and computing communication load distribution.

[0191] The present invention also discloses a computer-readable storage medium storing a computer program configured to implement the steps of the multi-hop routing and load distribution method described in the present invention when invoked by a processor.

[0192] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multi-hop routing and load distribution method for space-based computing constellations, characterized in that, Includes the following steps: Step 1: Establish a model for the space-based computing constellation to perform matrix-vector multiplication calculation tasks. The source node encodes and splits the task, and the relay node transmits the task hop by hop. During the end-to-end routing process, the relay node and the destination node perform distributed computation on the task. Step 2: Establish an optimization problem to minimize task execution latency, and propose a computing power awareness and load allocation algorithm based on block coordinate descent to decompose the original problem into routing and load allocation problems; Step 3: Select the best relay node based on the dynamic programming-based computing power-aware routing algorithm, and design a multi-hop coded load distribution algorithm based on the interior point method to achieve communication and computing load balancing. Step one includes: Step S1: The source node S first selects the optimal path p. i ,use The code encodes the task matrix A into... It is then divided into H sub-matrices, where path p is assigned. i The jth u Submatrix of skipping working node u It is a matrix composed of the code combinations of each row in task matrix A. It is a positive integer. A random linear combination of L rows from task matrix A; Step S2: The source node encodes the matrix It is sent hop-by-hop to the destination node D via H-1 relay nodes; Each relay node receives the submatrix assigned to it. For submatrixes not assigned to it, the relay node does not store them but sends them directly to the next hop node, i.e., a transparent transmission mechanism is adopted. Step S3: The destination node and each relay node complete the calculation of the submatrix assigned to them until the destination node and each relay node have completed at least L results.

2. The multi-hop routing and load balancing method according to claim 1, characterized in that, In step S3, the communication delay for each node to receive the assigned submatrix consists of the transmission delay and the propagation delay of the multi-hop ISL. The sum of the computation delay, propagation delay, and transmission delay of node u is given by the following formula: in, Indicates the selection of path p i Time j u Total delay for hopping to satellite node u to complete the task Indicates the selection of path p i Time j u The computation time of the hopping satellite node u. Indicates the selection of path p i Time task transmission to j u The propagation delay of the hopping satellite node u, Indicates the selection of path p i The task is transmitted to the j-th node on the path. u Transmission time of hop satellite node u; The CDF of the total delay for satellite node u to complete the task is as follows: Where, positive number μ u The straggling parameter, a, is determined by the computing power of satellite u. u The shift time parameter required for satellite u to complete a single inner product operation. Let l represent the number of hops in the same orbit inter-satellite link and the number of hops in different orbit inter-satellite links required for transmission from source node S to node u, respectively. i,v This indicates that the path p is assigned to it. i The number of rows in the encoding matrix of the v-th hop node. Represents path p i The jth u Jump transmission rate, considering mesh satellite network topology, d v d represents the distance between inter-satellite links of adjacent satellites in the same orbit. h b represents the distance between inter-satellite links of adjacent satellites in different orbits, and b represents the encoding matrix. The number of bits in a line, where c is the speed of light; T exe CDF is composed of each The product of is expressed by the following formula: T exe This represents the total latency for the system to complete the task. Indicates as of T exe The set of nodes that complete the submatrix calculation at any time; Because of T exe Since it is a positive number, its expectation can be obtained by integrating the CDF, as shown in the following formula:

3. The multi-hop routing and load balancing method according to claim 2, characterized in that, Step two also includes: Step y1: Find the optimal path p i and optimal load distribution Minimize average total latency The problem is defined as follows: in It is the set of natural numbers; Step y2: Replace the objective function with a constraint containing the expectation, as follows: Among them, t exe* Indicates the expected deadline; Step y3: Based on the block coordinate descent method, solve the total problem of step y2. Break it down into two sub-problems, problem This is an optimal path problem, given an initial feasible total task execution delay. and load distribution strategy i(0) Find the path p that maximizes the inner product number to complete the calculation. i ,question For the load distribution problem, as a constrained optimization problem, given path p i Then, find the minimum task execution delay t that guarantees task completion. exe* and specific load balancing strategies By continuously executing the above steps, the algorithm will gradually converge, yielding a routing and load distribution scheme that minimizes task execution latency.

4. The multi-hop routing and load balancing method according to claim 3, characterized in that, In step three, the computationally aware routing algorithm based on dynamic programming performs the following steps: Step 1: Input source node S, destination node D, Total inner product L, d of the task v ,d h ,R v ,R h The total delay of the result of the (r-1)th iteration Load distribution vector l i(r-1) ; Step 2: Initial State Step 3: According to equation (19) Calculation state And according to equation (20) Record the optimal previous hop node for node u. Step 4: Output the optimal path 5. The multi-hop routing and load balancing method according to claim 4, characterized in that, In step three, the multi-hop coding load distribution algorithm based on the interior-point method is implemented as follows: Step a1: Read the route obtained by the routing algorithm as the path, and read the load distribution and task execution latency of the previous iteration as the initial feasible solution; Step a2: Initialize the number of iterations and enter the loop; Step a3: Construct the barrier function Step a4: With X (k-1) ={t exe*(k-1) ,l i (k-1) The unconstrained minimum problem of minimizing the obstacle function is solved using} as the initial point, and the formula is as follows: Where r k It is a barrier factor; Step a5: Determine if the solution converges. If it converges, exit the loop; otherwise, decrease the obstacle factor r. k The value is incremented by 1, and the iteration count is returned to step a3. Step a6: Output the optimal latency and optimal load distribution.

6. The multi-hop routing and load balancing method according to claim 1, characterized in that, In step one, distributed matrix multiplication is performed on the space-based computing constellation. Assuming the transmission from source node S to destination node D requires H hops, the computational task y = Ax generated by the source node is distributed and computed via destination node D and H-1 relay nodes.

7. A multi-hop routing and load distribution system for space-based computing constellations, characterized in that, Includes: Module 1: used to establish a model for performing matrix-vector multiplication calculation tasks on a space-based computing constellation, providing a model for source nodes to encode and split tasks, relay nodes to transmit tasks hop by hop, and relay nodes and destination nodes to perform distributed computation on tasks during end-to-end routing. Module 2: This module is used to establish an optimization problem that minimizes task execution latency. It proposes a computational power awareness and load allocation algorithm based on block coordinate descent to decompose the original problem into routing and load allocation problems. Module 3: Used for designing computationally aware routing algorithms based on dynamic programming to select the best relay node, and for designing multi-hop encoded load distribution algorithms based on the interior point method to achieve communication and computation load balancing.

8. The multi-hop routing and load distribution system according to claim 7, characterized in that, The multi-hop routing and load balancing system operates as follows: Step 1: Input the space-based computing constellation parameters, including inter-satellite distance, inter-satellite transmission rate, displacement time parameters related to satellite computing power, straggler parameters, task source node, destination node, and number of task rows; Step 2: Initialize any feasible routes and load balancing; Step 3: Invoke the computational power-aware routing algorithm based on dynamic programming, with a fixed load distribution scheme, and search for computational power-aware routes by maximizing the number of expected encoded inner product calculations on the transmission path; Step 4: Invoke the encoding load distribution algorithm based on the interior point method, and calculate the computation and communication load of each hop on the path according to the route obtained in Step 3; Step 5: Calculate the task execution latency under the routing in Step 3 and the load distribution in Step 4; Step 6: Determine whether the task execution delay has decreased compared to the previous iteration result. If it has decreased, use the result of this iteration as the starting point for the next iteration and return to step 3; otherwise, jump to step 7. Step 7: Output computing power routing and computing communication load distribution, then end.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to implement the steps of the multi-hop routing and load distribution method according to any one of claims 1-6 when invoked by a processor.

Citation Information

Patent Citations

  • Coding calculation load balanced distribution method and system under heterogeneous multi-hop network topology

    CN115119266A

  • Satellite multi-hop task unloading optimization method and system based on edge calculation

    CN119212003A