Multi-index resource scheduling method and system applied to computing power network

Abstract

The invention relates to the technical field of computing power network resource scheduling, in particular to a multi-index resource scheduling method and system applied to a computing power network, and the method comprises the steps: classifying computing power scenes according to the measurement, analysis and modeling of various resources in the computing power network; corresponding feature factors are extracted according to the calculation models corresponding to the classifications; carrying out standardization processing and weight distribution on all feature factors of the computing power nodes by utilizing the idea of an entropy weight method, and carrying out comprehensive scoring on the capabilities of the nodes in the computing power network region through a computing model corresponding to each category; and according to the comprehensive score of each computing power node, carrying out resource scheduling on the computing task by improving an ant colony algorithm. According to the method, the calculation task can be matched with the corresponding service node, and the stability of task completion can be ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power network resource scheduling, and in particular to a multi-index resource scheduling method and system applied to a computing power network. Background Art

[0002] Research on computing network resource scheduling technology has important theoretical and practical significance. First, in theory, optimized resource scheduling algorithms and strategies will provide an effective means for the rational allocation of computing resources, promoting the further development of distributed systems. Second, efficient resource scheduling technology will enable computing tasks to be executed more quickly and accurately, thereby promoting the development and promotion of various compute-intensive applications, such as large-scale data processing and artificial intelligence model training. Furthermore, reasonable resource scheduling can effectively reduce computing costs and improve resource utilization, saving enterprises money and enhancing their competitiveness.

[0003] In addition, the research on computing network resource scheduling technology is also of great significance for actual production and application. Reference 1:

[21] Zhang Hongke, Quan Wei, Liu Kang. Research and Exploration of Computing Network [J]. ZTE Communications Technology, 2023, 29(1): 1-5, which gives an overview and summary of the development and key core technologies of computing network, indicating that the research and construction of computing network should achieve the in-depth integration goal of "computing power + network" and establish an intelligent, efficient, on-demand computing service platform to meet users' highly differentiated computing service needs.

[0004] However, current challenges include heterogeneous and heterogeneous networks and ubiquitous computing resources, uneven distribution, difficulty in cross-network resource scheduling, and insufficient intelligence. Computing network research and development should focus on breakthroughs in key core technologies: multidimensional identification, intelligent mapping, on-demand networking, collaborative transmission, intelligent computing, and system security. The heterogeneity of networks and devices complicates resource scheduling and limits integration. Therefore, research requires breakthroughs in key multidimensional identification technologies and the establishment of an integrated computing network identification system to achieve unified naming for the computing, storage, transmission capabilities, and other functional attributes of ubiquitous computing resources. Summary of the Invention

[0005] (1) Technical issues to be resolved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a multi-index resource scheduling method and system applied to computing power networks, which solves the technical problems of heterogeneity and heterogeneity of existing networks and devices, resulting in difficulties in computing power network resource scheduling and limited integration.

[0007] (2) Technical solution

[0008] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, an embodiment of the present invention provides a multi-index resource scheduling method applied to a computing power network, comprising the following steps:

[0010] Based on the measurement, analysis, and modeling of various resources in the computing network, computing scenarios are classified into computing-intensive, network-intensive, storage-intensive, image processing, deterministic real-time computing, or secure computing.

[0011] And extract the corresponding characteristic factors according to the calculation model corresponding to each classification;

[0012] The idea of ​​entropy weight method is used to standardize and weight all characteristic factors of computing power nodes. Through the calculation model corresponding to each category, the ability of nodes in the computing power network area is comprehensively scored.

[0013] According to the comprehensive score of each computing power node, the computing tasks are scheduled through the improved ant colony algorithm.

[0014] The multi-index resource scheduling method for computing power networks proposed in the embodiments of the present invention solves the technical problems of heterogeneity and heterogeneity of existing networks and devices, which lead to difficulties in computing power network resource scheduling and limited integration. A model is constructed for the complex scenarios of computing power networks. By integrating commonly used computing task models, the characteristic factors of each model are extracted. After data standardization and weight distribution using the idea of ​​entropy weight method, the capabilities of nodes in the computing power network area are scored according to the model. Based on the scoring, an improved ant colony algorithm is introduced to perform specific scheduling of computing tasks. This is more in line with current actual industrial production scenarios.

[0015] Optionally, the improved ant colony algorithm includes the following steps:

[0016] The node's completion credibility is evaluated based on the ratio of the node's actual task completion volume to the historical task acceptance volume;

[0017] The multi-index decision evaluation value based on the entropy weight method, the transmission capacity and the node completion credibility are used as the initial pheromone of each node; a, b, and c are used to represent the proportion of the node evaluation value pheromone, transmission capacity pheromone and completion credibility pheromone in the entire pheromone, respectively, and a+b+c=1;

[0018] Split any computing task into several task slices, and schedule the task slices in sequence to different computing nodes that meet the processing capabilities in the resource evaluation table of the computing network area; define the path and path length in the scenario: for any ant, the path is the adjacent computing nodes passed by the ant in a traversal; the path length is the time consumed by all nodes; iterate until the optimal path with the lowest overhead when scheduling task slices is found.

[0019] Optionally, the improved ant colony algorithm further includes the following steps:

[0020] ① Initialization parameters: Initialization parameters include the number of ants, number of iterations, and pheromone concentration; initialization of the taboo table and taboo table related parameters;

[0021] ② Initialize ants and pheromones: Randomly place a certain number of ants at any computing power node position in the space set B; initialize the pheromone concentration on each path to be equal;

[0022] ③Tabu table ant colony search algorithm: Each ant selects the next computing node based on pheromone concentration and heuristic rules, updates the pheromone concentration on the path, calculates the path length and fitness of each ant, and converts the path into a taboo solution in the taboo table based on the search results of the ant colony algorithm. According to the taboo strategy, the taboo table is updated, the solutions visited are recorded, and the taboo period is set.

[0023] ④ Update pheromone: Update the pheromone concentration on the path through pre-set pheromone volatilization and pheromone release rules, and adjust the pheromone update strategy to balance the impact of global search and local search;

[0024] ⑤ Determine the stopping condition: If the set number of iterations or other pre-set stopping conditions are reached, the process ends and the optimal solution is output; otherwise, return to step ② and continue searching.

[0025] Optionally, the taboo table related parameters include taboo length and taboo objects.

[0026] Optionally, the characteristic factors corresponding to each category of computing power scenarios are as follows:

[0027] Computationally intensive: CPU performance, memory capacity, and throughput;

[0028] Network intensive: memory size, transmission time;

[0029] Storage-intensive: bandwidth, latency, throughput, packet loss rate, load balancing efficiency, and cache efficiency;

[0030] Graphics processing: frame rate, rendering time, pixel fill rate, resolution, and shader performance;

[0031] Deterministic real-time computing: latency, bandwidth, throughput, data processing speed, window size, and fault recovery time;

[0032] Secure Computing: Hardware security modules, data encryption, threat detection and protection, resiliency, and disaster recovery.

[0033] Optionally, a comprehensive score is given to the capabilities of the nodes in the computing power network area, including the following steps: using the idea of ​​the entropy weight method to perform data standardization on all characteristic factors of the computing power nodes in turn, allocating the ratios of each indicator under each scheme, calculating the information entropy of each characteristic factor of the computing power node and determining the weight of each characteristic factor; and calculating the comprehensive score of the computing power node.

[0034] In comprehensively scoring the capabilities of nodes within the computing power network area, the computing power service platform informs the computing power gateway node of its own status and availability by regularly sending heartbeat messages; the heartbeat mechanism includes node registration, heartbeat message format, heartbeat message sending, heartbeat message receiving, node status update and exception handling, among which node status update adopts active update and passive update.

[0035] Optionally, the calculation formula for the comprehensive score of the computing power node is as follows:

[0036]

[0037] Among them, y ij is the proportion of the i-th computing power node under the j-th characteristic factor; W j is the weight of each characteristic factor, and m is the number of characteristic factors under this category.

[0038] Optionally, data standardization processing includes: using the characteristic factor as an information entropy indicator, dividing the indicator into a positive indicator and a negative indicator according to the meaning of the information entropy indicator, and standardizing them respectively, and ensuring that the normalized data are all positive numbers;

[0039] On the basis of data standardization, calculate the ratio distribution of each indicator under each scheme and assign the proportion y ij As the probability used in information entropy calculation, the forward formula is calculated as follows:

[0040]

[0041] Among them, each computing power node has j characteristic factors, and the value of a characteristic factor of the i-th computing power node is x ij , the value after data standardization is x′ ij .

[0042] Optionally, calculate the information entropy of each characteristic factor of the computing power node:

[0043]

[0044] Among them, e j is the entropy of the jth characteristic factor, and e j ≥0; if y ij =0, define e j=0, m is the number of characteristic factors considered, K is a constant used to ensure e j The result is less than or equal to 1.

[0045] Determine the weight of each characteristic factor, the calculation formula is as follows:

[0046]

[0047] Entropy j Combined with m characteristic factors, it is the weight W of each characteristic factor j .

[0048] In the second aspect, an embodiment of the present invention provides a multi-index resource scheduling system applied to a computing power network, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, and the processor implements the steps of any of the above methods when executing the computer program.

[0049] (3) Beneficial effects

[0050] The beneficial effects of the present invention are as follows: the multi-index resource scheduling method and system for computing power networks of the present invention builds models for complex computing power network scenarios, integrates commonly used computing task models, extracts characteristic factors of each model, and uses the entropy weight method to perform data standardization and weight assignment, then scores the capabilities of nodes within the computing power network area according to the model; based on the scores, an improved ant colony algorithm is introduced to specifically schedule computing tasks. This makes it more suitable for current actual industrial production scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the classification of the computing power network scenario model of the preferred embodiment of the present invention;

[0052] Figure 2 1 is a flow chart of a multi-index resource scheduling method applied to a computing power network according to a preferred embodiment of the present invention;

[0053] Figure 3 It is the heartbeat mechanism of the preferred embodiment of the present invention;

[0054] Figure 4 This is a resource scheduling minimum overhead path diagram of a preferred embodiment of the present invention;

[0055] Figure 5 is the average overhead of the improved ant colony algorithm of the preferred embodiment of the present invention;

[0056] Figure 6 This is the lowest overhead change per iteration for the preferred embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0058] The present invention integrates commonly used computing task models, extracts the characteristic factors of each model, and uses the entropy weight method to perform data standardization and weight assignment. The capabilities of nodes within the computing network area are then scored according to the model. Based on the scores, an improved ant colony algorithm is introduced to schedule computing tasks. A completion credibility mechanism is also added to ensure the completion rate of computing tasks. This solves the technical problem that the heterogeneous nature of existing networks and devices makes computing network resource scheduling difficult and integration is limited.

[0059] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0060] This paper divides resource scheduling in computing networks into evaluation and scheduling, and proposes a multi-metric resource scheduling method for computing networks. Resource collection and scheduling are performed by computing gateway nodes, while resource evaluation and some decision-making are performed by the computing service platform.

[0061] Example

[0062] Figure 1 This is a classification diagram of the computing power network scenario model in an embodiment of the present invention. Figure 2 The flowchart of the multi-index resource scheduling method applied to a computing power network in an embodiment of the present invention is as follows. The multi-index resource scheduling method applied to a computing power network in this embodiment includes the following steps:

[0063] S1: Based on the measurement, analysis, and modeling of various resources in the computing network, computing scenarios are classified into compute-intensive, network-intensive, storage-intensive, image processing, deterministic real-time computing, or secure computing.

[0064] S2: Extract the corresponding characteristic factors according to the calculation model corresponding to each classification;

[0065] S3: Use the entropy weight method to standardize and weight all characteristic factors of computing power nodes. Then, use the corresponding calculation model for each category to comprehensively score the capabilities of nodes in the computing power network area.

[0066] S4: Based on the comprehensive score of each computing power node, the computing task resources are scheduled through the improved ant colony algorithm.

[0067] The multi-index resource scheduling method for computing power networks proposed in the embodiments of the present invention solves the technical problems of heterogeneity and heterogeneity of existing networks and devices, which lead to difficulties in computing power network resource scheduling and limited integration. A model is constructed for the complex scenarios of computing power networks. By integrating commonly used computing task models, the characteristic factors of each model are extracted. After data standardization and weight distribution using the idea of ​​entropy weight method, the capabilities of nodes in the computing power network area are scored according to the model. Based on the scoring, an improved ant colony algorithm is introduced to perform specific scheduling of computing tasks. This is more in line with current actual industrial production scenarios.

[0068] The computing power network indicator model of this embodiment is constructed.

[0069] Traditional research on computing power measurement revolves around hardware communications and infrastructure service capabilities. These approaches are computationally centered, while networks, as the technology connecting these hardware facilities, typically focus solely on metrics such as bandwidth, latency, and packet loss. This general and single-minded quantitative approach has its merits in traditional communications development. However, with the continuous evolution of networks and the introduction of emerging technologies, network performance assessment has become increasingly complex. Given the diverse service demands, quantitative methods may not provide sufficient information to explain network performance. A more comprehensive measurement system is needed to more comprehensively assess the overall performance of modern computing networks. Similar to traditional networks, computing power networks also utilize multiple resources, including computing resources, storage resources, and network bandwidth. However, unlike traditional networks, which separate computing and networking, computing power networks not only highly integrate computing, networking, and storage, but their architecture is also designed to enable flexible resource scheduling and on-demand allocation based on user service needs across the cloud, edge, and end. This highly integrated model and diverse demands have further complicated the interrelationships between resources, leading to the development of resource measurement research into a multidimensional analytical field.

[0070] This embodiment summarizes and classifies complex services according to user computing tasks, and compiles a set of computing network computing model classifications that adapt to the needs of diverse business scenarios to better manage and utilize computing resources in the computing network, such as Figure 1 shown.

[0071] The computing power node data collection mechanism of this embodiment.

[0072] The information collection of computing power nodes is a crucial part of realizing resource evaluation. This embodiment uses a heartbeat mechanism to obtain the characteristic information of the nodes and maintain the real-time nature of the information. The heartbeat mechanism is a technology commonly used in network communications for monitoring and maintaining the online status of network equipment. In the evaluation of computing power network resources, the computing power service platform can inform the computing power gateway node of its own status and availability by regularly sending heartbeat messages. The content of the heartbeat mechanism of this embodiment includes node registration, heartbeat message format, heartbeat message sending, heartbeat message receiving, node status update, and exception handling, among which the node status update adopts two methods: active update and passive update.

[0073] First, node registration is required. When a node joins the computing power network, it must first register. The registration process primarily involves sending a registration request to the computing power service platform and waiting for a confirmation response. After successful registration, the node can be discovered by other nodes and added to the network. The node then periodically sends heartbeat messages containing key information, such as its ID, IP address, computing power, and storage capacity. This information helps the computing power service platform understand the node's basic status and resource availability, enabling it to assess the node's model service capabilities. To ensure that the node remains active and available, the node sends heartbeat messages at predetermined intervals. Typically, the heartbeat message interval is fixed, but can be adjusted based on network needs. The message sent can be a simple message simply confirming the node's existence. The more frequently the heartbeat messages are sent, the more timely the node's status in the network will be updated. Upon receiving the node heartbeat message, the computing power service platform processes it. If the message indicates that the node is normal, a confirmation message will be sent in response. If the message indicates that the node is abnormal, appropriate measures can be taken.

[0074] For updating the node status, both active and passive methods are used. The specific process is: after the computing power service platform obtains the information of all computing power nodes for the first time, it will send a specific update request message to the computing power node at regular intervals. When the computing power node receives the message, it will return the latest node information to the computing power service platform. This update is a passive update; for active updates, after each time the computing power node sends information to the computing power service platform, it will also record the information sent and the current status. Every once in a while, it will compare the current information and status with the last recorded information status. If the information difference exceeds a certain range (such as the storage space occupies 10% more than the last time), the latest information will be actively sent to the computing power service platform to facilitate timely updates of resource conditions in the area.

[0075] Finally, for exception handling, if the computing power service platform does not receive a confirmation message from a node for a long time, it may think that the node is offline or faulty. In this case, the computing power service platform can take some measures, such as reconnecting the node, re-registering to the network, etc., to restore the connection with the network. Figure 3 shown.

[0076] This embodiment uses multi-index resource evaluation.

[0077] In the current computing power network scenario, the trend of diversified business scenario requirements is obvious. For some businesses, link transmission rate and delay are not important factors. The computing power of the node itself and its own data processing capabilities are the key points. Although the computing network integration model abstracts the computing, storage, and network of all nodes into a whole, when scheduling resources, the corresponding specific nodes are still found, that is, to a certain extent, they rely on the processing capabilities of the computing power nodes themselves. This embodiment incorporates refined measurement attributes into the calculation of resource calls. The entropy weight method is a commonly used multi-indicator decision-making method, which is mainly used to distribute weights among multiple indicators. Its function is to solve the weight distribution problem in multi-indicator decision-making by calculating the entropy value and weight of each indicator, so as to obtain relatively accurate decision results. This method does not rely on the weights assigned by humans subjectively, but uses certain mathematical methods to calculate the weight ratios of different indicators.

[0078] In this embodiment, the following steps are preferably adopted:

[0079] In S1, the currently commonly used computing scenarios are classified and six computing models are summarized: computing-intensive, network-intensive, storage-intensive, image processing, deterministic real-time computing, and secure computing.

[0080] In S2, each computing model has characteristic factors (such as bias towards bandwidth, memory, hardware conditions, etc.), and these characteristic factors are extracted to quantify the importance of indicators. Figure 1 In this embodiment, the characteristic factors corresponding to each category of computing power scenarios are as follows:

[0081] Computationally intensive: CPU performance, memory capacity, and throughput;

[0082] Network intensive: memory size, transmission time;

[0083] Storage-intensive: bandwidth, latency, throughput, packet loss rate, load balancing efficiency, and cache efficiency;

[0084] Graphics processing: frame rate, rendering time, pixel fill rate, resolution, and shader performance;

[0085] Deterministic real-time computing: latency, bandwidth, throughput, data processing speed, window size, and fault recovery time;

[0086] Secure Computing: Hardware security modules, data encryption, threat detection and protection, resiliency, and disaster recovery.

[0087] In S3, the idea of ​​entropy weight method is used to perform data standardization on all characteristic factors of computing power nodes, assign ratios of various indicators under various schemes, calculate the information entropy of each characteristic factor of computing power nodes, determine the weight of each characteristic factor, and calculate the comprehensive score of computing power nodes. The first step is standardization. According to the meaning of information entropy indicators (characteristic factors), indicators can be divided into positive indicators (the larger the value, the better) and negative indicators (the smaller the value, the better). They are standardized by certain methods respectively, and it is necessary to ensure that the standardized data are all positive numbers. Assuming that the number of computing power nodes in the computing power network scenario is n, and each computing power node has j characteristic factors, then the value of a characteristic factor of the i-th computing power node is x ij , the value after data standardization is x′ ij . Then use the min-max normalization method:

[0088] The forward formula is:

[0089]

[0090] The reverse formula is:

[0091]

[0092] Based on data standardization, the ratio of the scheme is distributed. Taking the result of the forward formula as an example, the proportion y of the i-th computing power node under the j-th characteristic factor is calculated. ij , and regard it as the probability used in the information entropy calculation, the formula is:

[0093]

[0094] Continue to calculate the information entropy of each characteristic factor of the computing power node, and define the entropy of the j-th characteristic factor as e j :

[0095]

[0096] where e j ≥0. If y ij =0, define e j =0, m is the number of characteristic factors considered; then determine the weight of each characteristic factor, and e j The entropy value result is combined with m characteristic factors to calculate the weight W of each characteristic factor j :

[0097]

[0098] Finally, calculate the comprehensive score of the computing power node:

[0099]

[0100] In actual implementation, the characteristic factors extracted by the multi-indicator resource assessment method are not limited to one or several items, and the calculation model should also be defined and constructed according to specific scenarios. The multi-indicator assessment of the entropy weight method mainly distributes weights objectively through data differences, thereby reducing misjudgments caused by human subjective judgment.

[0101] During implementation, the improved ant colony algorithm of this embodiment preferably includes the following steps:

[0102] The node's completion credibility is evaluated based on the ratio of the node's actual task completion volume to the historical task acceptance volume;

[0103] The multi-index decision evaluation value based on the entropy weight method, the transmission capacity and the node completion credibility are used as the initial pheromone of each node; a, b, and c are used to represent the proportion of the node evaluation value pheromone, transmission capacity pheromone and completion credibility pheromone in the entire pheromone, respectively, and a+b+c=1;

[0104] Split any computing task into several task slices, and schedule the task slices in sequence to different computing nodes that meet the processing capabilities in the resource evaluation table of the computing network area; define the path and path length in the scenario: for any ant, the path is the adjacent computing nodes passed by the ant in a traversal; the path length is the time consumed by all nodes; iterate until the optimal path with the lowest overhead when scheduling task slices is found.

[0105] Computing task scheduling in this embodiment.

[0106] Traditional resource scheduling methods have limitations when it comes to addressing the completion rate of computing tasks, particularly when considering the trustworthiness of nodes in completing tasks. Traditional resource scheduling methods typically allocate and schedule resources based on metrics such as node load, network bandwidth, and task priority to achieve efficient task execution. However, these methods often overlook the stability and reliability of the nodes themselves, which can negatively impact task completion rates. Traditional resource scheduling methods typically prioritize node load to ensure that tasks are assigned to idle or lightly loaded nodes. This can improve task execution efficiency to a certain extent, but it doesn't fully consider node stability. If a node experiences failure or instability during task execution, it can lead to task failure or delay, impacting task completion rates. Furthermore, traditional resource scheduling methods often make scheduling decisions based on static node capability information, ignoring the dynamic changes and uncertainties of node operation. Node performance, reliability, and stability can be affected by a variety of factors, such as hardware failures, network congestion, and energy availability, all of which can significantly impact task completion rates. Traditional methods fail to effectively consider these dynamic factors, resulting in a lack of timely response and adjustment of resource scheduling to node status changes, thereby reducing the credibility of task completion.

[0107] In response to the above problems, this embodiment provides a computing network resource scheduling method based on an improved ant colony algorithm. This method introduces an improved ant colony algorithm to train the optimal path for computing nodes on the basis of multi-index resource evaluation results. First, the global optimal search capability is achieved by optimizing the algorithm's own defects, and a scheduling strategy with high matching between tasks and nodes is realized; then, by introducing a completion credibility mechanism, the completion degree of the computing tasks after scheduling is improved.

[0108] In S4, based on the multi-index resource evaluation method of entropy weight, an improved ant colony algorithm is introduced to perform specific transmission of computing tasks. First, the taboo table method is used to solve the problem that the ant colony algorithm is prone to falling into local optimality.

[0109] Determine the information initialization parameters and the composition of pheromones, and add a completion credibility mechanism to explore the optimal path. Specifically, the completion credibility mechanism can be judged by the ratio of the node's actual task completion amount to the historical task reception amount. The formula is as follows:

[0110] T d =T s / T a

[0111] T k (0) = aT b (0)+bT b (0)+cT d(0),a+b+c=1

[0112] Among them, T d is the completion credibility of the computing network node, T a is the total number of transmission tasks that the node has received historically, T s The number of tasks completed for node transmission. Obviously, when T s The larger the value, the T d The higher it is, the lower the failure rate of its data transmission task. s =T a When , it means that the node has successfully transmitted all the historical tasks it has received, and its credibility is 1, indicating that the node's transmission stability and reliability are very high. This improved method is brought into the initialization of pheromone, and the improved ant colony algorithm selects the node based on the multi-index decision evaluation value T of the entropy weight method. p (T p =S i ), transmission capacity T b And the node completion credibility T d As the initial pheromone of each node, a, b, and c represent the proportion of the node's evaluation value pheromone, transmission capacity pheromone, and completion credibility pheromone in the entire pheromone. It is worth mentioning that the construction of this completion credibility mechanism is based on the premise that the historical task processing volume of the traversed nodes is sufficient. This means that in the early stage of the construction of the computing power network environment, when all nodes in the area rarely participate in or even do not participate in the processing of computing tasks, T a The reference degree is low, which is why this method adds the transmission capacity T b As one of the evaluation criteria, in general industrial scenarios, communication transmission focuses more on real-time performance. Therefore, parameters such as bandwidth, delay, and packet loss rate can be standardized according to the min-max method, and then the corresponding proportions can be allocated as needed. Finally, the cumulative value of T can be obtained. b In the early stage of scene construction, you can increase T b Weight value. To a certain extent, it can guarantee the stability of resource scheduling. At the same time, because the resource scheduling of the improved ant colony algorithm is based on the multi-index evaluation value S of the entropy weight method, i The nodes are selected from the top several nodes, and these nodes that have been screened by the initial model also ensure that they match the computing task requirements to a greater extent.

[0113] Integrating the above process, when the number of iterations is completed, the best path can be selected. Specifically, suppose a computing power gateway node receives a computing task A and splits it into several task slices. The task slice set is represented by A={A1,A2,A3…A n+, compare the resource evaluation table of the area to find a group of computing nodes that meet the processing capabilities, and the set is represented as B = {B1, B2, B3…B n To prevent nodes from being overloaded, task slices must be scheduled sequentially to different computing nodes. The computing gateway node sends the task slices all at once, and all computing nodes receive them evenly. This raises the question of how to minimize the overhead of task slice scheduling. The ant colony algorithm with a tabu table is introduced to address this problem. The path and path length in this scenario are defined as follows: For any ant, the path is the number of adjacent computing nodes that the ant passes through during a single traversal; the path length is the time spent passing through all these nodes.

[0114] The improved ant colony algorithm of this embodiment preferably includes the following steps:

[0115] ① Initialization parameters: Initialization parameters include the number of ants, number of iterations, and pheromone concentration; initialization of the taboo table and taboo table related parameters;

[0116] ② Initialize ants and pheromones: Randomly place a certain number of ants at any computing power node position in the space set B; initialize the pheromone concentration on each path to be equal;

[0117] ③Tabu table ant colony search algorithm: Each ant selects the next computing node based on pheromone concentration and heuristic rules, updates the pheromone concentration on the path, calculates the path length and fitness of each ant, and converts the path into a taboo solution in the taboo table based on the search results of the ant colony algorithm. According to the taboo strategy, the taboo table is updated, the solutions visited are recorded, and the taboo period is set.

[0118] ④ Update pheromone: Update the pheromone concentration on the path through pre-set pheromone volatilization and pheromone release rules, and adjust the pheromone update strategy to balance the impact of global search and local search;

[0119] ⑤ Determine the stopping condition: If the set number of iterations or other pre-set stopping conditions are reached, the process ends and the optimal solution is output; otherwise, return to step ② and continue searching.

[0120] The improved ant colony algorithm performs scheduling based on multi-index resource evaluation. The subsequent algorithm accuracy mainly depends on the construction of pheromones. Therefore, when constructing pheromones, in order to ensure that the computing task can be completed within the specified time, the network transmission delay factor should be fully considered.

[0121] Experiment and result analysis of this embodiment.

[0122] Experimental Environment: The experimental nodes will use the current mainstream network and terminal equipment in the market, and the parameters of each node will be configured according to the official standards. The experiment uses Pycharm software to build the simulation process. Ten positive indicators are extracted from the node: bandwidth, throughput, latency, packet loss rate, current storage space, number of processor cores, CPU computing performance, GPU processing performance, memory size, and security protection capability. Among them, bandwidth and throughput use random parameters with a random range of 100Mbps / s to 10,000Mbps / s, packet loss rate is controlled in the range of 1% to 20%, and security protection capability is based on the assumed score set by the operating system, ranging from 1 to 100. Each parameter is dynamically valued, that is, the actual device operating status and resource usage at the time of value selection. Specific parameters are shown in Table 1, and node allocation is shown in Table 2.

[0123] Table 1 Node types and parameters

[0124]

[0125] Table 2: Distribution of computing power nodes

[0126]

[0127] Multi-indicator resource assessment.

[0128] To ensure the stability of the sample, data was obtained three times every 10 seconds for the same node, and the average of the three samples was taken for verification calculations. The max-min method was used to normalize all initial data. To avoid the impact of excessively large or small data on subsequent calculations, all values ​​were uniformly controlled between 1 and 100. The characteristic factors that have the greatest impact on each calculation model were extracted from the ten indicators and used to participate in the entropy weight calculation. The final results are shown in Table 2. The configuration of all calculation model node parameters shows that the weighted sum of different nodes in each calculation model and the degree of influence of the node's existing resource situation on the calculation model are closely related. For a calculation model with more characteristic factors, the size of the individual characteristic factor value has less impact on the total weight. This multi-indicator resource call calculation method also meets actual business needs.

[0129] Table 3 Calculation model results

[0130]

[0131] Resource scheduling paths and overhead.

[0132] The number of tasks was set to 100. The top 25 nodes within the region whose credibility scores matched the evaluation value range were selected to participate in the allocation and processing of computing tasks. The coordinates and attributes of each node are shown in Table 4. The communication overhead between nodes is positively correlated with the distance between node coordinates. The ant colony algorithm parameters were also set: the total number of ants was 15, the maximum number of algorithm iterations was 200, the pheromone release concentration was 1, the pheromone evaporation rate was 0.1, the algorithm was run 20 times, and the average result was calculated. The completion credibility simulation experiment parameters were set. The model requirements were based on the resource scheduling of a computationally intensive model. The initial number of nodes was set to 100. A multi-criteria decision-making evaluation method based on the entropy weight method was used to predict computationally intensive nodes. Nodes with an evaluation value greater than 80 were selected. The evaluation value in the constructed pheromone was weighted at 50%, the transmission capacity at 20%, and the completion credibility at 30%. The number of ant colony algorithm iterations was 100.

[0133] Table 4 Resource scheduling node settings

[0134]

[0135] Figure 4 shows the final convergence path, Figure 5 Shows the average convergence cost of the path after 100 iterations of the algorithm Figure 6 Shows the shortest cost statistics after each iteration. Figure 4 It can be seen intuitively that all paths form a shortest loop route, which makes it possible to schedule computing tasks from any point to all nodes with the optimal cost. Figure 5 As shown in the figure, the ant colony algorithm with the taboo table did not converge prematurely and fall into the local optimum, because its average cost trend was a wave-like downward trend, and there was a small peak in the later period, indicating that the node traversal has always been a global perspective. Figure 6 The shortest cost change shows that as the number of iterations increases, the result gradually converges to the optimal value. In summary, these results show that the application of the ant colony algorithm with the tabu table in resource scheduling of computing power networks is conducive to determining the global optimal solution.

[0136] Combine Figure 5 and Figure 6 The experimental results show that this method, when combined with the multi-index resource evaluation method based on the entropy weight method, can not only match the computing tasks with the corresponding service nodes, but also ensure the stability of task completion.

[0137] In the second aspect, an embodiment of the present invention provides a multi-index resource scheduling system applied to a computing power network, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, and the processor implements the steps of any of the above methods when executing the computer program.

[0138] In summary, the present invention aims at the resource scheduling problem of computing power network, splits resource scheduling into two aspects: evaluation and scheduling, and combines the idea of ​​multi-index decision-making of entropy weight method to propose a computing power network resource evaluation method based on entropy weight method. This method abstracts the important characteristic factors involved in each model application scenario, and then integrates these characteristic factors using the idea of ​​entropy weight method, and uses the multi-index calculation method to calculate the weighted comprehensive value of a characteristic factor to predict the node's matching ability for the computing model. Then, the improved ant colony algorithm is introduced for specific resource scheduling, and a completion credibility mechanism is introduced to ensure the completion rate of the task. An experimental environment is set up to conduct experiments from two aspects: resource evaluation and scheduling of computing tasks. The experimental results show that the scheme has high feasibility.

[0139] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0140] It should be understood that the specific order or hierarchy of steps in the disclosed processes is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0141] In the foregoing detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention comprises less than all the features of any individual disclosed embodiment. The appended claims are therefore hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0142] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein may be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described around their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person may implement the described functions in an adaptable manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of this disclosure.

[0143] The steps of the methods or algorithms described in conjunction with the embodiments herein may be directly embodied as hardware, software modules executed by a processor, or a combination thereof. The software module may be located in a RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and storage medium may also be present in a user terminal as discrete components.

[0144] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or external to the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well known in the art.

[0145] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purposes of describing the above embodiments, but one of ordinary skill in the art will recognize that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to encompass all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent the term "comprising" is used in the specification or claims, the term is intended to be encompassed in a manner similar to the term "including," as explained in terms of "including," used as a transitional word in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

[0146] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A multi-index resource scheduling method applied to a computing power network, characterized in that: The following steps are involved: Based on the measurement, analysis, and modeling of various resources in the computing network, computing scenarios are classified into computing-intensive, network-intensive, storage-intensive, image processing, deterministic real-time computing, or secure computing. And extract the corresponding characteristic factors according to the calculation model corresponding to each classification; The idea of ​​entropy weight method is used to standardize and weight all characteristic factors of computing power nodes. Through the calculation model corresponding to each category, the ability of nodes in the computing power network area is comprehensively scored. According to the comprehensive score of each computing power node, the computing tasks are scheduled through the improved ant colony algorithm.

2. The multi-index resource scheduling method applied to a computing power network according to claim 1, characterized in that: The improved ant colony algorithm comprises the following steps: The node's completion credibility is evaluated based on the ratio of the node's actual task completion volume to the historical task acceptance volume; The multi-index decision evaluation value based on the entropy weight method, the transmission capacity and the node completion credibility are used as the initial pheromone of each node; a, b, and c are used to represent the proportion of the node evaluation value pheromone, transmission capacity pheromone and completion credibility pheromone in the entire pheromone, respectively, and a+b+c=1; Split any computing task into several task slices, and schedule the task slices in sequence to different computing nodes that meet the processing capabilities in the resource evaluation table of the computing network area; define the path and path length in the scenario: for any ant, the path is the adjacent computing nodes passed by the ant in a traversal; the path length is the time consumed by all nodes passed; iterate until the optimal path with the lowest overhead when scheduling task slices is found.

3. The multi-index resource scheduling method applied to a computing power network according to claim 1 or 2, characterized in that: The improved ant colony algorithm further includes the following steps: ① Initialization parameters: Initialization parameters include the number of ants, number of iterations, and pheromone concentration; initialization of the taboo table and taboo table related parameters; ② Initialize ants and pheromones: Randomly place a certain number of ants at any computing power node position in the space set B; initialize the pheromone concentration on each path to be equal; ③Tabu table ant colony search algorithm: Each ant selects the next computing node based on pheromone concentration and heuristic rules, updates the pheromone concentration on the path, calculates the path length and fitness of each ant, and converts the path into a taboo solution in the taboo table based on the search results of the ant colony algorithm. According to the taboo strategy, the taboo table is updated, the solutions visited are recorded, and the taboo period is set. ④ Update pheromone: Update the pheromone concentration on the path through pre-set pheromone volatilization and pheromone release rules, and adjust the pheromone update strategy to balance the impact of global search and local search; ⑤ Determine the stopping condition: If the set number of iterations or other pre-set stopping conditions are reached, the process ends and the optimal solution is output; otherwise, return to step ② and continue searching.

4. The multi-index resource scheduling method applied to a computing power network according to claim 3, characterized in that: The taboo table related parameters include taboo length and taboo objects.

5. The multi-index resource scheduling method applied to a computing power network according to claim 1, characterized in that: The characteristic factors corresponding to each category of computing power scenarios are as follows: Computationally intensive: CPU performance, memory capacity, and throughput; Network intensive: memory size, transmission time; Storage-intensive: bandwidth, latency, throughput, packet loss rate, load balancing efficiency, and cache efficiency; Graphics processing: frame rate, rendering time, pixel fill rate, resolution, and shader performance; Deterministic real-time computing: latency, bandwidth, throughput, data processing speed, window size, and fault recovery time; Secure Computing: Hardware security modules, data encryption, threat detection and protection, resiliency, and disaster recovery.

6. The multi-index resource scheduling method applied to a computing power network according to claim 1 or 5, characterized in that: The comprehensive scoring of the capabilities of nodes in the computing power network area includes the following steps: using the idea of ​​entropy weight method to perform data standardization processing on all characteristic factors of the computing power node in sequence, allocating the ratio of each indicator under each scheme, calculating the information entropy of each characteristic factor of the computing power node and determining the weight of each characteristic factor; and calculating the comprehensive score of the computing power node; In comprehensively scoring the capabilities of nodes within the computing power network area, the computing power service platform informs the computing power gateway node of its own status and availability by regularly sending heartbeat messages; the heartbeat mechanism includes node registration, heartbeat message format, heartbeat message sending, heartbeat message receiving, node status update and exception handling, among which node status update adopts active update and passive update.

7. The multi-index resource scheduling method applied to a computing power network according to claim 6, characterized in that: The calculation formula for the comprehensive score of computing power nodes is as follows: Among them, y ij is the proportion of the i-th computing power node under the j-th characteristic factor; W j is the weight of each characteristic factor, and m is the number of characteristic factors under this category.

8. The multi-index resource scheduling method applied to a computing power network according to claim 7, characterized in that: The data standardization process includes: taking the characteristic factor as the information entropy index, dividing the index into positive index and negative index according to the meaning of the information entropy index, and standardizing them respectively, and ensuring that the normalized data are all positive numbers; On the basis of data standardization, calculate the ratio distribution of each indicator under each scheme and assign the proportion y ij As the probability used in information entropy calculation, the forward formula is calculated as follows: Among them, each computing power node has j characteristic factors, and the value of a characteristic factor of the i-th computing power node is x ij , the value after data standardization is x′ ij .

9. The multi-index resource scheduling method applied to a computing network according to claim 8, characterized in that: Calculate the information entropy of each characteristic factor of the computing power node: Among them, e j is the entropy of the jth characteristic factor, and e j ≥0; if y ij =0, define e j =0, m is the number of characteristic factors considered; K is a constant used to ensure e j The result is less than or equal to 1. The weight of each characteristic factor is determined by the following calculation formula: Entropy j Combined with m characteristic factors, it is the weight W of each characteristic factor j .

10. A multi-index resource scheduling system applied to a computing power network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images