Moon base hybrid computing power center scheduling method and device

Abstract

The invention discloses a hybrid computing power center scheduling method and device for a lunar base. The method comprises the following steps: acquiring a computing task of the lunar base; clustering the calculation tasks to obtain a calculation task category cluster; the calculation task category cluster comprises a calculation-intensive task, a data-intensive task and a communication-intensive task; pooling computing resources of the lunar base hybrid computing power center to obtain M computing resource pools; and matching the calculation task category cluster with the M calculation resource pools to obtain a scheduling result of the lunar base hybrid computing power center. According to the method, the utilization rate of computing power resources is improved, the advantages of various computing resources are fully played through reasonable allocation and scheduling of different types of tasks, and idle and waste of the resources are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and particularly to a scheduling method and device for a hybrid computing power center of a lunar base. Background Art

[0002] With the continuous deepening of human exploration and development of the moon, building a base on the moon has become an important direction for future development. In the operation and scientific research activities of the lunar base, a large amount of data and complex computing tasks need to be processed, which poses high requirements for computing power resources. However, due to the special nature of the lunar environment, such as limited resources, high communication latency, unstable energy supply, etc., traditional computing power center scheduling methods are difficult to meet the requirements.

[0003] Currently, the computing power centers of lunar bases usually adopt a single computing architecture, such as a computing platform based on a central processing unit (CPU) or a graphics processing unit (GPU). This single-architecture computing power center is inefficient in processing certain specific types of tasks and cannot fully utilize the advantages of various computing resources. In addition, existing scheduling methods often do not consider the special environmental factors of the lunar base, resulting in low utilization rate of computing power resources, long task execution time, and affecting the normal operation of the lunar base and the progress of scientific research. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a scheduling method and device for a hybrid computing power center of a lunar base, which can reasonably allocate and schedule different types of computing resources according to the type and requirements of tasks, and improve the utilization rate of computing power resources and the task execution efficiency.

[0005] To solve the above technical problem, in the first aspect of an embodiment of the present invention, a scheduling method for a hybrid computing power center of a lunar base is disclosed, and the method includes:

[0006] S1, obtaining the computing tasks of the lunar base;

[0007] S2, clustering the computing tasks to obtain computing task category clusters; the computing task category clusters include compute-intensive tasks, data-intensive tasks, and communication-intensive tasks;

[0008] S3, pooling the computing resources of the hybrid computing power center of the lunar base to obtain M computing resource pools;

[0009] S4, matching the computing task category clusters with the M computing resource pools to obtain the scheduling result of the hybrid computing power center of the lunar base.

[0010] As an optional implementation manner, in the first aspect of an embodiment of the present invention, the clustering the computing tasks to obtain computing task category clusters includes:

[0011] S21. Perform EMD decomposition on the computing task to obtain IMF components \(c_i(t)\), where \(i = 1, 2, \cdots, N\) and \(N\) is the number of IMF components; the IMF components are computing task characteristic parameters. i (t), \(i = 1, 2, \cdots, N\), where \(N\) is the number of IMF components; the IMF components are characteristic parameters of the computing task;

[0012] S22. Use the correlation coefficient calculation model to perform correlation calculation on the IMF component \(c_i(t)\) and the computing task \(x(t)\) to obtain the correlation coefficient Corr(\(c_i\), \(x\)). i (t) and the computing task \(x(t)\) to obtain the correlation coefficient Corr(\(c_i\), \(x\));

[0013] The correlation coefficient calculation model is:

[0014]

[0015] where \(x_j\) i (t) is the \(j\)-th component in \(x(t)\), is the average value of the IMF component, is the mean value of \(x(t)\);

[0016] S23. Eliminate the IMF component with the smallest correlation coefficient to obtain the first IMF component.

[0017] S24. Use the sample entropy calculation model to process the first IMF component to obtain the sample entropy En.

[0018] S25. Select 3 clustering centers according to the sample entropy En; the 3 clustering centers are:

[0019] The IMF component with the largest En is the first clustering center \(K_1\) 1 , the IMF component with the smallest En is the second clustering center \(K_2\) 2 , and the IMF component with the median En is the third clustering center \(K_3\) 3 ;

[0020] S26. Calculate the distances between the first IMF component and the 3 clustering centers; according to the distances, cluster the first IMF component into the clusters where each clustering center is located to obtain the computing task category clusters.

[0021] The computing task category clusters include: the compute-intensive tasks where the clustering center \(K_1\) 1 is located, the data-intensive tasks where the clustering center \(K_2\) 2 is located, and the communication-intensive tasks where the clustering center \(K_3\) 3 is located.

[0022] As an optional implementation manner, in the first aspect of the embodiments of the present invention, the pooling of the computing resources of the lunar base hybrid computing center to obtain \(M\) computing resource pools includes:

[0023] S31. Obtain the computing task requirement information; the computing task requirement information is a weighted undirected graph G V ={N V , L V , C v}, where N V represents the resource utilization rate, L V represents the task execution time, and C v represents the task energy consumption;

[0024] S32. Construct a computing resource network; the computing resource network is a weighted undirected graph where N S represents the set of computing resource nodes, L S represents the set of computing resource links, represents the set of computing resource node performance metrics, represents the set of computing resource link performance metrics;

[0025] S33. According to the computing task requirement information, partition the computing resource network to obtain M computing resource pools.

[0026] As an optional implementation manner, in the first aspect of the embodiments of the present invention, the matching the computing task category clusters and the M computing resource pools to obtain the scheduling result of the lunar base hybrid computing power center includes:

[0027] S41. Process the computing tasks in the computing task category clusters to obtain the priority level index of the computing tasks;

[0028] S42. Monitor the M computing resource pools to obtain the status information of each computing resource; the status information of the computing resource includes CPU information, GPU information, available resources information of dedicated accelerators, load information, and energy consumption information;

[0029] S43. According to the priority level index of the computing tasks, allocate computing resources to each computing task and execute the computing tasks;

[0030] S44. During the execution of the computing tasks, when the status information of the computing resources exceeds the preset status information threshold, trigger a feedback mechanism to calculate the penalty coefficient;

[0031] S45. According to the penalty coefficient, re-allocate computing resources to each computing task to obtain the scheduling result of the lunar base hybrid computing power center.

[0032] As an optional implementation manner, in the first aspect of the embodiments of the present invention, the calculation formula of the priority level index of the computing tasks is:

[0033] rk = α 1 A 1k + α 2 A 2k + α 3 A 3k

[0034] where r k is the priority level index of the k-th computing task, A 1k is the urgency of the k-th computing task, A 2k is the resource requirement of the k-th computing task, A 3k is the task execution time of the k-th computing task, α 1 is the weight of A 1k is the weight of A 2 is the weight of A 2k is the weight of A 3 is the weight of A 3k is the weight.

[0035] As an optional implementation manner, in the first aspect of the embodiments of the present invention, allocating computing resources for each computing task according to the priority level index of the computing task and executing the computing task includes:

[0036] S431, adding l computing tasks to the queue according to the priority level index of the computing task;

[0037] S432, when the number l of computing tasks is less than or equal to a preset threshold L, re-establishing a task queue and adding other tasks to the queue;

[0038] S433, when the number of computing tasks to be scheduled on a node of a certain computing resource network is greater than or equal to the preset threshold L, another node of the computing resource network needs to be selected to assist in processing the computing task;

[0039] S434, when the number of computing tasks on a node of a certain computing resource network is less than the preset threshold L, allocating computing resources for each computing task according to the priority level index of the computing tasks in the queue and executing the computing task.

[0040] As an optional implementation manner, in the first aspect of the embodiments of the present invention, the expression of the penalty coefficient is:

[0041]

[0042] where y(t) is the penalty coefficient at time t, C t is the CPU usage rate at time t, G t is the GPU usage rate at time t, M t is the memory usage rate at time t, S tLet the memory utilization rate at time t be, the penalty coefficient gain be ε, the preset state information threshold be T, and max(C t ,G t ,M t ,S t ) be the maximum resource utilization rate at time t.

[0043] In the second aspect of the embodiments of the present invention, a scheduling device for a lunar base hybrid computing power center is disclosed. The device includes:

[0044] A computing information acquisition module, configured to acquire computing tasks of the lunar base;

[0045] A computing task clustering module, configured to cluster the computing tasks to obtain computing task category clusters; the computing task category clusters include compute-intensive tasks, data-intensive tasks, and communication-intensive tasks;

[0046] A computing resource pooling module, configured to pool the computing resources of the lunar base hybrid computing power center to obtain M computing resource pools;

[0047] A scheduling module, configured to match the computing task category clusters and the M computing resource pools to obtain the scheduling result of the lunar base hybrid computing power center.

[0048] As an optional implementation manner, in the second aspect of the embodiments of the present invention, the clustering of the computing tasks to obtain computing task category clusters includes:

[0049] S21, perform EMD decomposition on the computing tasks to obtain IMF components c i (t), i = 1, 2, L, N, where N is the number of IMF components; the IMF components are computing task characteristic parameters;

[0050] S22, use the correlation coefficient calculation model to perform correlation calculation on the IMF component c i (t) and the computing task x(t) to obtain the correlation coefficient Corr(c, x);

[0051] The correlation coefficient calculation model is:

[0052]

[0053] where x i (t) is the i-th component in x(t), is the average value of the IMF component, is the mean value of x(t);

[0054] S23, remove the IMF component with the smallest correlation coefficient to obtain the first IMF component;

[0055] S24. Use the sample entropy calculation model to process the first IMF component to obtain the sample entropy En.

[0056] S25. Select three clustering centers according to the sample entropy En. The three clustering centers are:

[0057] The IMF component with the largest En is the first clustering center K 1 , the IMF component with the smallest En is the second clustering center K 2 , and the IMF component with the median En is the third clustering center K 3 ;

[0058] S26. Calculate the distances between the first IMF component and the three clustering centers. According to the distances, cluster the first IMF component into the clusters where each clustering center is located to obtain the calculation task category clusters.

[0059] The calculation task category clusters include: the compute-intensive tasks where the clustering center K 1 is located, the data-intensive tasks where the clustering center K 2 is located, and the communication-intensive tasks where the clustering center K 3 is located.

[0060] As an optional implementation manner, in the second aspect of the embodiments of the present invention, the pooling of the computing resources of the lunar base hybrid computing power center to obtain M computing resource pools includes:

[0061] S31. Obtain the computing task requirement information. The computing task requirement information is the weighted undirected graph G V = {N V , L V , C v}, where N V represents the resource utilization rate, L V represents the task execution time, and C v represents the task energy consumption;

[0062] S32. Construct a computing resource network. The computing resource network is a weighted undirected graph where N S represents the set of computing resource nodes, L S represents the set of computing resource links, represents the set of computing resource node performance metrics, represents the set of computing resource link performance metrics;

[0063] S33. Divide the computing resource network according to the computing task requirement information to obtain M computing resource pools.

[0064] As an alternative implementation, in the second aspect of the embodiments of the present invention, the matching of the computing task category clusters and the M computing resource pools to obtain the scheduling result of the lunar base hybrid computing power center includes:

[0065] S41. Process the computing tasks in the computing task category clusters to obtain the priority level index of the computing tasks;

[0066] S42. Monitor the M computing resource pools to obtain the status information of each computing resource; the status information of the computing resource includes CPU information, GPU information, available resources information of dedicated accelerators, load information, and energy consumption information;

[0067] S43. Allocate computing resources to each computing task according to the priority level index of the computing task and execute the computing task;

[0068] S44. During the execution of the computing task, when the status information of the computing resource exceeds the preset status information threshold, trigger a feedback mechanism to calculate the penalty coefficient;

[0069] S45. Re-allocate computing resources to each computing task according to the penalty coefficient to obtain the scheduling result of the lunar base hybrid computing power center.

[0070] As an alternative implementation, in the second aspect of the embodiments of the present invention, the formula for calculating the priority level index of the computing task is:

[0071] r k =α 1 A 1k +α 2 A 2k +α 3 A 3k

[0072] where r k is the priority level index of the kth computing task, A 1k is the urgency of the kth computing task, A 2k is the resource demand of the kth computing task, A 3k is the task execution time of the kth computing task, α 1 is the weight of A 1k is the weight of A 2 is the weight of A 2k is the weight of A 3 is the weight of A 3k is the weight of A.

[0073] As an alternative implementation, in the second aspect of the embodiments of the present invention, the allocating computing resources to each computing task according to the priority level index of the computing task and executing the computing task includes:

[0074] S431. Add l computing tasks to the queue according to the priority index of the computing tasks.

[0075] S432. When the number l of computing tasks is less than or equal to the preset threshold L, re - establish the task queue and add other tasks to the queue.

[0076] S433. When the number of computing tasks to be scheduled on a node of a certain computing resource network is greater than or equal to the preset threshold L, it is necessary to select a node of another computing resource network to assist in processing the computing tasks.

[0077] S434. When the number of computing tasks on a node of a certain computing resource network is less than the preset threshold L, allocate computing resources for each computing task according to the priority index of the computing tasks in the queue and execute the computing tasks.

[0078] As an alternative implementation manner, in the second aspect of the embodiments of the present invention, the expression of the penalty coefficient is:

[0079]

[0080] where y(t) is the penalty coefficient at time t, C t is the CPU usage rate at time t, G t is the GPU usage rate at time t, M t is the memory usage rate at time t, S t is the storage usage rate at time t, ε is the penalty coefficient gain, T is the preset state information threshold, and max(C t , G t , M t , S t ) is the maximum resource usage rate at time t.

[0081] The third aspect of the present invention discloses another lunar base hybrid computing power center scheduling device, and the device includes:

[0082] A memory storing executable program code;

[0083] A processor coupled to the memory;

[0084] The processor calls the executable program code stored in the memory and executes some or all of the steps of the lunar base hybrid computing power center scheduling method disclosed in the first aspect of the embodiments of the present invention.

[0085] A fourth aspect of the present invention discloses a computer - storable medium. The computer - storable medium stores computer instructions which, when invoked, are used to execute some or all of the steps in the lunar base hybrid computing power center scheduling method disclosed in the first aspect of the embodiments of the present invention.

[0086] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0087] 1. Improve the utilization rate of computing power resources: By reasonably allocating and scheduling different types of tasks, the advantages of various computing resources are fully utilized, avoiding the idleness and waste of resources.

[0088] 2. Shorten the task execution time: Precise scheduling according to the requirements of tasks and the characteristics of computing resources can effectively improve the execution efficiency of tasks and shorten the completion time of tasks.

[0089] 3. Reduce energy consumption: Considering energy consumption factors during the scheduling process and reasonably arranging the use of computing resources contribute to reducing the energy consumption of the lunar base.

[0090] 4. Adapt to the special lunar environment: Fully considering special factors such as limited resources, high communication latency, and unstable energy supply in the lunar base, the reliability and stability of the computing power center in the lunar environment are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0092] Figure 1 is a schematic flowchart of a lunar base hybrid computing power center scheduling method disclosed in the embodiments of the present invention;

[0093] Figure 2 is a schematic flowchart of another lunar base hybrid computing power center scheduling method disclosed in the embodiments of the present invention;

[0094] Figure 3 is a schematic structural diagram of a lunar base hybrid computing power center scheduling system disclosed in the embodiments of the present invention;

[0095] Figure 4 is a schematic structural diagram of a lunar base hybrid computing power center scheduling device disclosed in the embodiments of the present invention;

[0096] Figure 5 is a schematic structural diagram of another lunar base hybrid computing power center scheduling device disclosed in the embodiments of the present invention. Detailed implementation manners

[0097] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

[0098] The terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or equipment.

[0099] Referring to "embodiment" herein means that a specific feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0100] The present invention discloses a scheduling method and device for a lunar base hybrid computing power center. The method includes: obtaining a computing task of the lunar base; clustering the computing task to obtain a computing task category cluster; the computing task category cluster includes compute-intensive tasks, data-intensive tasks, and communication-intensive tasks; pooling the computing resources of the lunar base hybrid computing power center to obtain M computing resource pools; matching the computing task category cluster and the M computing resource pools to obtain a scheduling result of the lunar base hybrid computing power center. The method of the present invention improves the utilization rate of computing power resources: by reasonably allocating and scheduling different types of tasks, the advantages of various computing resources are fully utilized, and the idle and waste of resources are avoided. The following will be described in detail respectively.

[0101] Embodiment 1

[0102] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a scheduling method for a lunar base hybrid computing power center disclosed in an embodiment of the present invention. Among them, Figure 1The described lunar base hybrid computing power center scheduling method is applied to the field of resource scheduling technology, and the embodiments of the present invention are not limited. As Figure 1 shown, the lunar base hybrid computing power center scheduling method may include the following operations:

[0103] S1. Obtain the computing tasks of the lunar base;

[0104] S2. Cluster the computing tasks to obtain computing task category clusters; the computing task category clusters include compute-intensive tasks, data-intensive tasks, and communication-intensive tasks;

[0105] S3. Pool the computing resources of the lunar base hybrid computing power center to obtain M computing resource pools;

[0106] S4. Match the computing task category clusters and the M computing resource pools to obtain the scheduling result of the lunar base hybrid computing power center.

[0107] Optionally, the clustering of the computing tasks to obtain computing task category clusters includes:

[0108] S21. Perform EMD decomposition on the computing tasks to obtain IMF components c i (t), i = 1, 2, L, N, where N is the number of IMF components; the IMF components are computing task characteristic parameters;

[0109] Each computing task includes parameters such as the type, scale, priority, and deadline of the task. Normalize these parameters to obtain the expression x(t) of the computing task, and perform EMD decomposition on the expression x(t) of the computing task to obtain IMF components c i (t);

[0110] S22. Use the correlation coefficient calculation model to perform correlation calculations on the IMF components c i (t) and the computing task x(t) to obtain the correlation coefficient Corr(c, x);

[0111] The correlation coefficient calculation model is:

[0112]

[0113] where x i (t) is the i-th component in x(t), is the average value of the IMF component, is the mean value of x(t);

[0114] S23. Eliminate the IMF component with the smallest correlation coefficient to obtain the first IMF component;

[0115] S24. Using the sample entropy calculation model, process the first IMF component to obtain the sample entropy En.

[0116] The sample entropy estimation formula is where m is the dimension, taking 1 or 2, r ∈ [0.1std, 0.25std] is the similarity tolerance, std is the standard deviation, N is the length of the sequence, and are the probabilities of two sequences matching m and m + 1 points respectively under the similarity tolerance r.

[0117] S25. According to the sample entropy En, select 3 clustering centers; the 3 clustering centers are:

[0118] The IMF component with the largest En is the first clustering center K 1 , the IMF component with the smallest En is the second clustering center K 2 , and the IMF component with the median En is the third clustering center K 3 ;

[0119] S26. Calculate the distances between the first IMF component and the 3 clustering centers; according to the distances, cluster the first IMF component into the clusters where each clustering center is located to obtain the calculation task category clusters.

[0120] The calculation task category clusters include: the compute-intensive tasks where the clustering center K 1 is located, the data-intensive tasks where the clustering center K 2 is located, and the communication-intensive tasks where the clustering center K 3 is located.

[0121] The distance formula is where X and Y are the covariance matrices of the samples respectively.

[0122] Optionally, pooling the computing resources of the lunar base hybrid computing center to obtain M computing resource pools includes:

[0123] S31. Obtain the computing task requirement information; the computing task requirement information is the weighted undirected graph G V ={N V , L V , C v}, where N V represents the resource utilization rate, L V represents the task execution time, and C v represents the task energy consumption;

[0124] S32. Construct a computing resource network; the computing resource network is a weighted undirected graph where N SDenote the set of computing resource nodes, L S Denote the set of computing resource links, Denote the set of computing resource node performance metrics, Denote the set of computing resource link performance metrics;

[0125] S33. According to the computing task requirement information, partition the computing resource network to obtain M computing resource pools. Specifically, use the breadth - first search algorithm to partition the computing resource network according to the computing task requirement information to obtain M computing resource pools.

[0126] Optionally, the matching of the computing task category clusters and the M computing resource pools to obtain the scheduling result of the lunar base hybrid computing power center includes:

[0127] S41. Process the computing tasks in the computing task category clusters to obtain the priority level index of the computing tasks;

[0128] S42. Monitor the M computing resource pools to obtain the status information of each computing resource; the status information of the computing resource includes CPU information, GPU information, available resources information of dedicated accelerators, load information, and energy consumption information;

[0129] S43. According to the priority level index of the computing tasks, allocate computing resources to each computing task and execute the computing tasks;

[0130] S44. During the execution of the computing tasks, when the status information of the computing resources exceeds the preset status information threshold, trigger a feedback mechanism and calculate the penalty coefficient;

[0131] When one of the CPU information, GPU information, available resources information of dedicated accelerators, load information, and energy consumption information exceeds the preset threshold T, trigger a feedback mechanism;

[0132] S45. According to the penalty coefficient, re - allocate computing resources to each computing task to obtain the scheduling result of the lunar base hybrid computing power center, including:

[0133] When triggering the feedback mechanism, use the dynamic hierarchical network resource optimization algorithm to divide the computing resources into three levels and dynamically adjust the weights of each resource node. According to the task priority, dependency relationship, and system load conditions, use the penalty coefficient and the dynamic weight adjustment factor to adjust the resource allocation in real - time;

[0134] Monitor the success rate and total resource consumption of each task at time t, and calculate the dynamic weight adjustment factor;

[0135]

[0136] Among them, ε is the penalty coefficient gain, y(t) is the penalty coefficient at time t, and T1(t) is the success rate of task T at time t i and T2(t) is the total resource consumption of task T at time t i .

[0137] Construct the task dependency matrix D ik and temporarily update the weight w ij (t) of the resource node;

[0138]

[0139] In the formula, w ij (t) is the weight of task T at time t i at node j, R(t) is the available resource at time t, p i (t) is the priority index of task T i , θ i (t) is the dynamic weight adjustment factor of task T at time t i , ∑ k D ik is the sum of dependencies of task T i , ∑ l R kj (t) is the sum of resource requirements on resource node j at time t, p k (t) is the priority index of task T at time t k , θ k (t) is the dynamic weight adjustment factor of task T at time t k , ∑ l D lk is the sum of dependencies of task T k .

[0140] The construction method of the dependency matrix D ik is as follows:

[0141] List all tasks to be executed and record the dependency of each task;

[0142] Assign weights according to the importance of the dependencies. The weight is a value in the range [0,1], indicating the strength of the dependency;

[0143] Create an n×n matrix D, where n is the total number of tasks. The initial values of the matrix are all 0. Fill the matrix according to the dependencies between tasks and the assigned weights. The element D ik of the matrix represents the degree to which task i depends on task k.

[0144] Perform resource allocation in real time according to the updated resource node weights;

[0145] Optionally, the calculation formula for the priority level index of the computing task is:

[0146] r k = α 1 A 1k + α 2 A 2k + α 3 A 3k

[0147] where r k is the priority level index of the kth computing task, A 1k is the urgency of the kth computing task, A 2k is the resource requirement of the kth computing task, A 3k is the task execution time of the kth computing task, α 1 is the weight of A 1k α 2 is the weight of A 2k α 3 is the weight of A 3k .

[0148] Optionally, allocating computing resources to each computing task according to the priority level index of the computing task and executing the computing task includes:

[0149] S431. Adding l computing tasks to the queue according to the priority level index of the computing task;

[0150] S432. When the number l of computing tasks is less than or equal to the preset threshold L, re-establishing the task queue and adding other tasks to the queue;

[0151] S433. When the number of computing tasks to be scheduled on the node of a certain computing resource network is greater than or equal to the preset threshold L, it is necessary to select a node of another computing resource network to assist in processing the computing task;

[0152] S434. When the number of computing tasks on the node of a certain computing resource network is less than the preset threshold L, allocating computing resources to each computing task according to the priority level index of the computing tasks in the queue and executing the computing task.

[0153] Optionally, the expression of the penalty coefficient is:

[0154]

[0155] where y(t) is the penalty coefficient at time t, C t is the CPU usage rate at time t, G t is the GPU usage rate at time t, M t is the memory usage rate at time t, S tLet \(U\) be the memory utilization rate at time \(t\), \(\varepsilon\) be the penalty coefficient gain, \(T\) be the preset state information threshold, and \(\max(C t ,G t ,M t ,S t ) be the maximum resource utilization rate at time \(t\).

[0156] It can be seen that the method of the present invention improves the utilization rate of computing resources: by reasonably allocating and scheduling different types of tasks, it gives full play to the advantages of various computing resources and avoids the idle and waste of resources. It shortens the task execution time: by accurately scheduling according to the requirements of the task and the characteristics of the computing resources, it can effectively improve the execution efficiency of the task and shorten the completion time of the task. It reduces energy consumption: by considering the energy consumption factor in the scheduling process and reasonably arranging the use of computing resources, it helps to reduce the energy consumption of the lunar base. It adapts to the special lunar environment: by fully considering special factors such as limited resources, high communication delay, and unstable energy supply in the lunar base, it improves the reliability and stability of the computing center in the lunar environment.

[0157] Embodiment 2

[0158] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of another lunar base hybrid computing center scheduling method disclosed in the embodiments of the present invention. Among them, Figure 2 the described lunar base hybrid computing center scheduling method is applied to the field of resource scheduling technology, and the embodiments of the present invention do not make limitations. As Figure 2 shown, the lunar base hybrid computing center scheduling method may include the following operations:

[0159] 1. Task reception and analysis: Receive computing tasks from various application systems in the lunar base and analyze parameters such as the type, scale, priority, and deadline of the tasks.

[0160] 2. Resource monitoring: Real-time monitor the status of various computing resources in the hybrid computing center, including information such as the available resource amount, load condition, and energy consumption of the CPU, GPU, dedicated accelerators (such as FPGA), etc.

[0161] 3. Task classification and matching: According to the results of task analysis, classify the tasks into different types, such as compute-intensive, data-intensive, communication-intensive, etc., and match them with the corresponding types of computing resources in the hybrid computing center.

[0162] 4. Resource scheduling strategy formulation: Based on the results of task classification and matching, combined with the information of resource monitoring, formulate a reasonable resource scheduling strategy. The factors considered in the scheduling strategy include maximizing resource utilization, minimizing task execution time, and minimizing energy consumption.

[0163] 5. Task Allocation and Execution: According to the formulated scheduling strategy, tasks are allocated to the corresponding computing resources for execution, and the execution progress of tasks and the usage of resources are monitored in real time.

[0164] 6. Feedback and Optimization: Based on the results of task execution and resource usage, the scheduling strategy is fed back and optimized to continuously improve the efficiency and effectiveness of scheduling.

[0165] The scheduling device of the lunar base hybrid computing power center of the present invention includes a task receiving module, a resource monitoring module, a task classification module, a scheduling strategy formulation module, a task allocation module, and a feedback optimization module.

[0166] Step S101, Task Receiving and Analysis. When a new computing task is submitted to the hybrid computing power center of the lunar base, the task receiving module first receives the task and analyzes the relevant parameters of the task. For example, for an astronomical observation data processing task, analyze its data volume, computational complexity, priority, and deadline, etc.

[0167] Step S102, Resource Monitoring. The resource monitoring module obtains the status information of various computing resources in the hybrid computing power center in real time, including the number of cores and utilization rate of the CPU, the video memory usage and computing power of the GPU, the configuration status and the number of available logic units of the FPGA, etc. At the same time, it also monitors the energy supply situation and communication bandwidth.

[0168] Step S103, Task Classification and Matching. The task classification module classifies tasks into different types such as compute-intensive, data-intensive, and communication-intensive according to the results of task analysis. For example, an astronomical observation data processing task usually belongs to a data-intensive task, while an orbit calculation task may be a compute-intensive task. Then, different types of tasks are matched with the corresponding types of computing resources in the hybrid computing power center.

[0169] Step S104, Resource Scheduling Strategy Formulation. The scheduling strategy formulation module comprehensively considers the results of task classification and matching, the information of resource monitoring, and scheduling objectives (such as maximizing resource utilization, minimizing task execution time, minimizing energy consumption, etc.) to formulate a specific resource scheduling strategy. For example, if there is an urgent compute-intensive task currently, and the CPU load is high while the GPU has more idle resources, the scheduling strategy can be to allocate this task to the GPU for execution.

[0170] Step S105, Task Allocation and Execution. The task allocation module allocates tasks to the corresponding computing resources for execution according to the formulated scheduling strategy. During the task execution process, the execution progress of the task and the usage of resources are monitored in real time.

[0171] Step S106, Feedback and Optimization. The feedback and optimization module evaluates and provides feedback on the scheduling policy based on the results of task execution and resource usage. If any deficiencies or irrationalities are found in the scheduling policy, it is promptly optimized and adjusted to improve the efficiency and effectiveness of subsequent task scheduling.

[0172] As Figure 3 shown, the lunar base hybrid computing power center scheduling system of the present invention includes the following modules:

[0173] The task receiving module is used to receive computing tasks from various application systems of the lunar base and transfer the tasks to the task analysis module.

[0174] The resource monitoring module is used to monitor the status information of various computing resources in the hybrid computing power center in real time, including the available resource quantity, load condition, energy consumption, etc., and transfer the monitoring results to the scheduling policy formulation module.

[0175] The task classification module is used to classify the received tasks, determine the types of tasks, such as compute-intensive, data-intensive, communication-intensive, etc., and transfer the classification results to the scheduling policy formulation module.

[0176] The scheduling policy formulation module is used to formulate a reasonable resource scheduling policy based on the task classification results, resource monitoring information, and scheduling objectives, and transfer the policy to the task allocation module.

[0177] The task allocation module is used to allocate tasks to the corresponding computing resources for execution according to the scheduling policy, and monitor the task execution progress and resource usage.

[0178] The feedback and optimization module is used to provide feedback and optimize the scheduling policy based on the task execution results and resource usage to improve the scheduling efficiency and effectiveness.

[0179] The technologies for scheduling include:

[0180] Fast startup technology: Container image loading, significantly reducing the cold startup latency;

[0181] Seamless migration technology: xPU state saving and restoration, seamless business migration;

[0182] Logical computing power mapping: Computing power abstraction agent, decoupling AI services from xPU devices;

[0183] Barrier and reconstruction chain: Implementing the global-barrier semantics, realizing the state synchronization, link release, and reconstruction of supernode xPUs.

[0184] It can be seen that the method of the present invention improves the utilization rate of computing power resources: by reasonably allocating and scheduling different types of tasks, the advantages of various computing resources are fully utilized, and the idle and waste of resources are avoided. It shortens the task execution time: by accurately scheduling according to the requirements of the task and the characteristics of the computing resources, the execution efficiency of the task can be effectively improved, and the completion time of the task can be shortened. It reduces energy consumption: by considering the energy consumption factor during the scheduling process and reasonably arranging the use of computing resources, it helps to reduce the energy consumption of the lunar base. It adapts to the special lunar environment: fully considering special factors such as limited resources, high communication latency, and unstable energy supply in the lunar base, it improves the reliability and stability of the computing power center in the lunar environment.

[0185] Embodiment III

[0186] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a scheduling device for a hybrid computing power center in a lunar base disclosed in an embodiment of the present invention. Among them, Figure 4 the described scheduling device for the hybrid computing power center in the lunar base is applied to the field of resource scheduling, and the embodiments of the present invention do not make limitations. As Figure 4 shown, the scheduling device for the hybrid computing power center in the lunar base may include the following operations:

[0187] S301, a computing information acquisition module, for acquiring the computing task information of the lunar base;

[0188] S302, a computing task clustering module, for clustering the computing tasks to obtain computing task category clusters; the computing task category clusters include compute-intensive tasks, data-intensive tasks, and communication-intensive tasks;

[0189] S303, a computing resource pooling module, for pooling the computing resources of the hybrid computing power center in the lunar base to obtain M computing resource pools;

[0190] S304, a scheduling module, for matching the computing task category clusters and the M computing resource pools to obtain the scheduling result of the hybrid computing power center in the lunar base.

[0191] Embodiment IV

[0192] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of another scheduling device for a hybrid computing power center in a lunar base disclosed in an embodiment of the present invention. Among them, Figure 5 the described scheduling device for the hybrid computing power center in the lunar base is applied to the field of resource scheduling, and the embodiments of the present invention do not make limitations. As Figure 5 shown, the scheduling device for the hybrid computing power center in the lunar base may include the following operations:

[0193] A memory 401 storing executable program code;

[0194] A processor 402 coupled to the memory 401;

[0195] The processor 402 invokes the executable program code stored in the memory 401 to execute the steps in the lunar base hybrid computing power center scheduling method described in Embodiment 1 and Embodiment 2.

[0196] Embodiment 5

[0197] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute the steps in the lunar base hybrid computing power center scheduling method described in Embodiment 1 and Embodiment 2.

[0198] The device embodiments described above are merely illustrative. Modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0199] Through the specific descriptions of the above embodiments, those skilled in the art can clearly understand that each implementation mode can be realized by means of software plus a necessary general hardware platform, and of course, it can also be realized by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, and the storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0200] Finally, it should be noted that: The lunar base hybrid computing power center scheduling method and device disclosed in the embodiments of the present invention only disclose the preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling a hybrid computing center at a lunar base, characterized in that: The method comprises: S1, obtain the computing task of the lunar base; S2, clustering the computing tasks to obtain computing task category clusters; the computing task category clusters include computing-intensive tasks, data-intensive tasks, and communication-intensive tasks; S3, pooling the computing resources of the hybrid computing center of the lunar base to obtain M computing resource pools; S4, matching the computing task category cluster and the M computing resource pools to obtain the scheduling result of the hybrid computing power center of the lunar base.

2. The method for scheduling a hybrid computing center at a lunar base according to claim 1, characterized in that: The computing tasks are clustered to obtain computing task category clusters, including: S21, performing EMD decomposition on the computing task to obtain IMF component c i (t), i = 1, 2, L, N, N is the number of IMF components; the IMF components are characteristic parameters of the calculation task; S22, using the correlation coefficient calculation model, the IMF component c i (t) and the calculation task x(t) are correlated to obtain the correlation coefficient Corr(c,x); The correlation coefficient calculation model is: Among them, x i (t) is the i-th component in x(t), is the average value of the IMF component, is the mean of x(t); S23, eliminating the IMF component with the smallest correlation coefficient to obtain the first IMF component; S24, using a sample entropy calculation model to process the first IMF component to obtain a sample entropy En; S25, selecting three cluster centers according to the sample entropy En; the three cluster centers are: The largest IMF component of En is the first cluster center K1, the smallest IMF component of En is the second cluster center K2, and the median IMF component of En is the third cluster center K3; S26, calculating the distance between the first IMF component and the three cluster centers; clustering the first IMF component into clusters where the cluster centers are located according to the distance, to obtain computing task category clusters; The computing task category cluster includes: computing intensive tasks where the cluster center K1 is located, data intensive tasks where the cluster center K2 is located, and communication intensive tasks where the cluster center K3 is located.

3. The method for scheduling a hybrid computing center at a lunar base according to claim 1, characterized in that: The computing resources of the hybrid computing center of the lunar base are pooled to obtain M computing resource pools, including: S31, obtaining computing task requirement information; the computing task requirement information is a weighted undirected graph G V = {N V ,L V ,C v }, where N V represents resource utilization, L V represents the task execution time, C v represents the energy consumption of the task; S32, constructing a computing resource network; the computing resource network is a weighted undirected graph Where N S represents the set of computing resource nodes, L S represents a set of computing resource links, Represents a set of computing resource node performance indicators, Represents a set of computing resource link performance indicators; S33, dividing the computing resource network according to the computing task requirement information to obtain M computing resource pools.

4. The method for scheduling a hybrid computing center at a lunar base according to claim 1, characterized in that: The computing task category clusters and the M computing resource pools are matched to obtain the scheduling result of the hybrid computing center of the lunar base, including: S41, processing the computing tasks in the computing task category cluster to obtain a priority index of the computing tasks; S42, monitoring the M computing resource pools to obtain status information of each computing resource; the status information of the computing resources includes CPU information, GPU information, dedicated accelerator available resource information, load information and energy consumption information; S43, allocating computing resources to each computing task according to the priority index of the computing task, and executing the computing task; S44, during the execution of the computing task, when the status information of the computing resource exceeds a preset status information threshold, triggering a feedback mechanism to calculate a penalty coefficient; S45, reallocating computing resources to each computing task according to the penalty coefficient, and obtaining the scheduling result of the hybrid computing power center of the lunar base.

5. The method for scheduling the hybrid computing center of the lunar base according to claim 4, characterized in that: The calculation formula of the priority index of the computing task is: r k =α1A 1k +α2A 2k +α3A 3k Among them, r k is the priority index of the kth computing task, A 1k is the urgency of the kth computing task, A 2k is the resource requirement of the kth computing task, A 3k is the task execution time of the kth computing task, α1 is A 1k The weight of A 2k The weight of A 3k The weight of .

6. The method for scheduling the hybrid computing center of the lunar base according to claim 4, characterized in that: The step of allocating computing resources to each computing task according to the priority index of the computing task and executing the computing task includes: S431, adding l computing tasks to the queue according to the priority index of the computing task; S432, when the number of computing tasks l is less than or equal to a preset threshold L, re-establish the task queue and add other tasks to the queue; S433, when the number of computing tasks to be scheduled on a node of a computing resource network is greater than or equal to a preset threshold value L, it is necessary to select a node of another computing resource network to assist in processing the computing tasks; S434, when the number of computing tasks of a node of a computing resource network is less than a preset threshold value L, computing resources are allocated to each computing task according to the priority index of the computing tasks in the queue to execute the computing tasks.

7. The method for scheduling a hybrid computing center at a lunar base according to claim 4, characterized in that: The expression of the penalty coefficient is: Among them, y(t) is the penalty coefficient at time t, C t is the CPU usage at time t, G t is the GPU utilization rate at time t, M t is the memory usage rate at time t, S t is the memory usage rate at time t, ε is the penalty coefficient gain, T is the preset state information threshold, max(C t ,G t ,M t ,S t ) is the maximum resource utilization rate at time t.

8. A hybrid computing center scheduling device for a lunar base, characterized in that: The device comprises: Computation information acquisition module, used to obtain the computing tasks of the lunar base; A computing task clustering module, used to cluster the computing tasks to obtain computing task category clusters; the computing task category clusters include computing-intensive tasks, data-intensive tasks and communication-intensive tasks; The computing resource pooling module is used to pool the computing resources of the hybrid computing center of the lunar base to obtain M computing resource pools; The scheduling module is used to match the computing task category cluster and the M computing resource pools to obtain the scheduling result of the hybrid computing power center of the lunar base.

9. A hybrid computing center scheduling device for a lunar base, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the lunar base hybrid computing center scheduling method as described in any one of claims 1-7.

10. A computer storable medium, characterized in that: The computer storable medium stores computer instructions, which, when called, are used to execute the lunar base hybrid computing center scheduling method as described in any one of claims 1-7.

Images