Intelligent network computing power resource management system

Abstract

The invention relates to the technical field of computing power management, in particular to an intelligent network computing power resource management system which comprises a node response load combing module, a task computing power grading module, a computing power scheduling adaptation module, a link load blocking detection module and a computing power flow direction reprogramming module. In the invention, by tracking the continuous change and inversion frequency of the node response direction, combining the computing power use state and the cache occupation ratio, dynamically evaluating the node stability, extracting the task logic depth and the cache interaction deviation to divide the computing power level, the hierarchical matching of the task demand and the node capability is realized; nodes with high cache load and unstable response are screened out, the resource scheduling accuracy and the integrating degree of task allocation are improved, the link abnormal growth trend and the queuing backlog proportion are monitored, the transmission blocking risk is recognized in time, and the flow direction is reconstructed by preferably selecting replaceable nodes based on the idle computing power and the bearing ratio; and the agility of resource calling and the continuity of task execution in a complex network are improved.

Description

Technical Field

[0001] The present invention relates to the field of computing power management technology, and in particular to an intelligent network computing power resource management system. Background Art

[0002] The field of computing power management technology lies at the intersection of information technology and computer science, primarily involving technologies for the unified scheduling, allocation, and optimized management of distributed computing resources. These technologies encompass core issues such as computing power scheduling strategies, resource awareness mechanisms, computing node selection, load balancing methods, resource collaboration in heterogeneous environments, and the converged scheduling of edge and cloud computing. Developments in this area are dedicated to improving the efficiency and responsiveness of computing power resources, supporting the efficient operation of complex computing tasks in multi-source, heterogeneous environments. They serve as a foundational technology supporting artificial intelligence, big data analytics, IoT applications, and intelligent network systems.

[0003] Among them, the traditional intelligent network computing resource management system refers to a system that coordinates the configuration and unified control of cloud and edge computing resources in intelligent network scenarios. The technical matter targeted by this patent subject is how to achieve effective unified management and task scheduling of multi-source heterogeneous computing resources in an intelligent network environment. Traditional intelligent network computing resource management systems typically use a static computing power description method based on a resource representation model. By pre-defining node characteristics and computing power indicators, a resource information database is established. Then, a static scheduling algorithm based on the shortest task completion time or minimum energy consumption is combined to perform task matching and resource allocation.

[0004] The existing system is built on a static model. The node status description lacks the ability to capture real-time fluctuations and cache interaction dynamics. The task matching process cannot reflect the correlation between task hierarchy and resource carrying. Faced with an environment where node load changes frequently or cache response pressure drops continuously, the scheduling logic is prone to matching errors. The scheduling efficiency drops significantly in scenarios where cache occupancy is high. The lack of link monitoring leads to delayed identification of transmission bottlenecks. Task dispatch does not form a feedback loop with the link status. When the data channel is blocked, the original path output is maintained, resulting in the coexistence of task transmission backlog and resource idleness, which limits the system's adaptability to task allocation in complex network structures and flow control efficiency. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose an intelligent network computing power resource management system.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions: A smart network computing resource management system includes: The node response load analysis module uses node response fluctuation records from the Zhiwang computing resource management to analyze the direction of continuous changes, the number of adjacent response time reversals, and the cumulative number of reversals. It then combines node computing power usage, cache read and write backlogs, and input and output processing ratios to generate a node computing power stability table. The task computing power level classification module extracts the task logic call depth, total number of instructions and cache interaction times based on the node computing power stable state table, calculates the difference and classifies the levels, matches the task intensity, and generates a task computing power usage level list; The computing power scheduling adaptation module refers to the task computing power usage hierarchy list, analyzes the computing power usage and cache usage of online nodes, selects nodes whose computing power margin matches the task intensity, eliminates nodes whose cache is close to saturation, and generates a set of nodes with available computing power for the task; The link load blocking detection module monitors transmission anomalies and queue volume based on the set of available computing nodes for the task, analyzes the continuous increase in the backlog ratio, and generates a link transmission blocking list; The computing power flow re-editing module refers to the link transmission blocking list and the unassigned nodes, selects the ratio nodes to replace the connection, adjusts the task flow, and generates the computing power resource management result.

[0007] As a further solution of the present invention, the node computing power stable state table includes the number of direction reversals, the response time fluctuation frequency, the cache access processing ratio, and the node computing power usage ratio; the task computing power usage hierarchy list includes the logic call depth interval, the instruction call quantity level, the cache interaction frequency range, and the call interaction difference segment; the task available computing power node set includes computing power margin nodes, cache load controllable nodes, and online adjustable points; the link transmission congestion list includes abnormal transmission frequency, connection node queuing intensity, and task backlog growth ratio; the computing power resource management results include blocked node numbers, replacement node numbers, and adjusted task flow paths; The nodes whose computing power margin matches the task intensity refer to online nodes whose remaining computing power is sufficient to meet the intensity requirements of the specified task and whose cache occupancy is not close to saturation; The ratio node replaces the connected node: refers to selecting a node with high task transmission efficiency from unassigned nodes and replacing the original connected node in the case of link congestion.

[0008] As a further solution of the present invention, the node response load combing module includes: The response fluctuation extraction submodule obtains node response fluctuation records, extracts continuous response directions, determines whether the directions of adjacent response times are consistent, records continuous direction sequences, and generates node direction change trend sequences; The reversal identification submodule calls the node direction change trend sequence, compares whether adjacent directions are opposite, counts the number of direction reversals, and generates a direction reversal intensity result based on the node computing power usage; The steady state generation submodule obtains the cache read-write backlog and the input-output processing capacity based on the direction reversal intensity result, calculates the cache access and processing ratio, calls the reversal intensity comparison frequency and ratio relationship, and generates a node computing power steady state table; The node computing power usage refers to the total amount of computing power resources consumed by the node in actually executing tasks within a time period.

[0009] As a further scheme of the present application, the task computing power grade division module comprises: The task parameter extraction submodule obtains the node computing power steady state table, extracts the logical call depth, instruction call total number and cache interaction times corresponding to each task, integrates and records the three data according to the task dimension, establishes a task execution structure item set, and generates a task execution index set; The intensity difference calculation submodule calls the task execution index set, calculates the difference value of the logical call depth and the cache interaction times in each record, matches the logical and cache difference value intervals according to the difference value, and generates a task intensity grade reference table; The grade list generation submodule filters and classifies the execution intensity of the differentiated grade tasks according to the grade labels in the task intensity grade reference table in combination with the instruction call total number, constructs the computing power distribution record under the grade tasks, and generates a task computing power usage level list; The task execution structure item set refers to the integrated set of the three core parameters of the logical call depth, instruction call total number and cache interaction times corresponding to each task; The difference value matching logic refers to a matching rule for mapping the difference value of the logical call depth and the cache interaction times to a preset task intensity grade; The cache difference value interval refers to the intensity grade division section corresponding to the logical depth and cache interaction difference value according to the task type; The execution intensity refers to the consumption degree of the node computing power resources during the running of the task.

[0010] As a further scheme of the present application, the computing power scheduling adaptation module comprises: The node state extraction submodule obtains the task computing power usage level list, extracts the resource state of the online node, reads the computing power occupation value of the node, calculates the computing power occupation distribution index, integrates the associated index and the node identifier in structure, and generates an online resource occupation index set; The resource matching judgment submodule calls the online resource occupation index set, compares the current computing power occupation value of the node and the task intensity value according to the computing power intensity grade record in the task list, and generates a task adaptation node list; The computing power node screening submodule calls the task adaptation node list to determine whether the cache occupancy ratio of the node is saturated, removes nodes with cache occupancy in the saturated range, retains the remaining nodes as schedulable targets, and generates a set of computing power nodes available for the task; The computing power utilization value refers to the amount and proportion of computing power resources occupied by the node being executed; The node whose cache occupancy is in the saturation range refers to a node whose cache resource usage has reached and exceeded a preset upper limit threshold, resulting in a performance bottleneck.

[0011] As a further solution of the present invention, the calculation formula of the computing power occupancy distribution index is specifically: ; in, Represents the computing power distribution index of the k-th node, Represents the computing power usage of the k-th node on the m-th task, Represents the average computing power usage of all tasks of the k-th node, Represents the total cache usage of the kth node, Represents the total number of tasks currently processed by the k-th node, Represents a small constant that prevents division by zero.

[0012] As a further solution of the present invention, the link load blocking detection module includes: The abnormal record monitoring submodule calls the set of available computing power nodes for the task, monitors the number of transmission anomalies of the connected nodes, records the growth of abnormal events in chronological order, eliminates records without continuous changes, and generates an abnormal event growth sequence; The queue ratio calculation submodule calls the abnormal event growth sequence, extracts the task queue volume corresponding to the connection node and the node's task capacity, calculates the queue ratio index value, determines whether the node ratio is continuously increasing, filters the upward trend records, and generates the queue ratio change trend; The blocking list generation submodule identifies the node identifiers corresponding to the continuously rising records according to the queue ratio change trend, summarizes the abnormal sequences and queue status, eliminates the node items that do not meet the conditions, and generates a link transmission blocking list; The record items without continuous changes refer to the node record data where the number of abnormal events does not continue to increase in adjacent time periods; The upward trend record refers to a record item in which the queue ratio index value shows an increasing trend at consecutive time points; The node items that do not meet the conditions refer to node data that do not show a continuous upward trend in the abnormal growth and queue ratio analysis and do not meet the congestion judgment criteria.

[0013] As a further solution of the present invention, the calculation formula of the queue ratio index value is specifically: ; in, Representative connected nodes at time The queue ratio index value at the time, Representative connected nodes at time The number of tasks queued at the moment, Representative time Moment The weight of the task to be queued, Representative time Moment The estimated processing time of the tasks to be queued, Representative connected nodes at time The maximum task capacity that can be carried at any time, Indicates time The total number of queued tasks associated with the moment.

[0014] As a further solution of the present invention, the computing power flow reprogramming module includes: The blocking node positioning submodule extracts the node number corresponding to each blocking record according to the link transmission blocking list, organizes the node number and the blocking information into corresponding mapping items, removes duplicate numbers and summarizes them to generate a blocking node number set; The replacement node screening submodule calls the blocked node number set, reads the idle computing power and task load of the currently unallocated nodes in the intelligent network computing resource management, calculates the computing power difference index of the unallocated nodes, compares the ratios of all unallocated nodes, screens the nodes with the dominant ratios, marks them with numbers, and generates a set of candidate replacement nodes; The task flow adjustment submodule calls the replacement candidate node set, redirects the corresponding tasks to the replacement nodes according to the connection path structure of the original blocking node, restructures the task scheduling relationship, updates the node connection and task allocation path, and generates computing power resource management results; The repeated number refers to the same node number that appears multiple times in the link transmission blocking list and corresponds to multiple blocking records; The numbering and marking of the node with the dominant ratio refers to the individual with the highest ratio of computing power difference index among the unallocated nodes, which is used as the replacement node and its node number is identified.

[0015] As a further solution of the present invention, the calculation formula of the computing power difference index of the unassigned nodes is specifically: ; in, Representative number is The computing power difference indicator of the unassigned nodes, Representative number is The unassigned nodes in Idle computing power value under dimension, Representative number is The number of tasks carried by unassigned nodes, Represents the number of idle computing power dimensions involved in the calculation, Representative number is The scheduling delay weight value corresponding to the unassigned node, A positive offset factor.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by tracking the continuous changes and reversal frequency of node response directions, combining the computing power usage status and cache occupancy ratio, dynamically evaluating node stability, extracting task logic depth and cache interaction deviation to divide computing power levels, achieving hierarchical matching of task requirements and node capabilities, screening out nodes with high cache load but unstable response, improving the accuracy of resource scheduling and the fit with task allocation, monitoring abnormal link growth trends and queue backlog ratios, timely identifying transmission congestion risks, and optimizing alternative nodes based on idle computing power and load ratio to reconstruct flow directions, thereby improving the agility of resource calling and the continuity of task execution in complex networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a system flow chart of the present invention; Figure 2 is a system block diagram of the present invention; Figure 3 This is a flow chart of the node response load combing module of the present invention; Figure 4 This is a flow chart of the task computing power level division module of the present invention; Figure 5 This is a flow chart of the computing power scheduling adaptation module of the present invention; Figure 6 This is a flow chart of the link load blocking detection module of the present invention; Figure 7 This is a flow chart of the computing power flow reprogramming module of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0020] Referring to Figure 1 and Figure 2 A wisdom network computing resource management system comprises: The node response load analysis module obtains the node response fluctuation record in the wisdom network computing resource management, extracts the continuous change direction, judges the inversion phenomenon of adjacent response time, accumulates the number of direction inversion, reads the node computing power usage, extracts the cache read-write backlog and input-output processing amount, compares the cache access and input-output processing ratio, combines the fluctuation frequency and processing ratio, and generates a node computing power stable state table; The task computing power level division module extracts the task logic call depth, instruction call total number and cache interaction times from the node computing power stable state table, calculates the difference value of logic call and cache interaction, divides the level according to the difference value interval, matches the execution intensity and level, and generates a task computing power usage level list; The computing power scheduling adaptation module references the task computing power usage level list, extracts the online node resource state, reads the computing power occupation value and cache occupation ratio, judges the matching of computing power margin and task intensity, eliminates the nodes with nearly saturated cache occupation, and forms a task available computing power node set; The link load blockage detection module calls the task available computing power node set, monitors the number of transmission exceptions, records the continuous abnormal growth, extracts the connection node task queuing amount, calculates the proportion of backlog tasks to the loadable amount, judges the continuous rising proportion, and generates a link transmission blockage list; The computing power flow direction reprogramming module positions the blocked node number according to the link transmission blockage list, calls the unallocated nodes in the wisdom network computing resource management, reads the idle computing power and the number of bearers, filters the nodes with large ratio for replacement connection, adjusts the task flow direction, and generates a computing power resource management result.

[0021] The node computing power stable state table includes the number of direction reversals, response time fluctuation frequency, cache access processing ratio, and node computing power usage ratio. The task computing power usage hierarchy list includes the logic call depth range, instruction call quantity level, cache interaction frequency range, and call interaction difference segment. The task available computing power node set includes computing power margin nodes, cache load controllable nodes, and online adjustable points. The link transmission congestion list includes abnormal transmission frequency, connection node queuing intensity, and task backlog growth ratio. The computing power resource management results include blocked node number, replacement node number, and adjusted task flow path.

[0022] See also Figure 3 and Figure 2 , the node response load combing module includes: The response fluctuation extraction submodule obtains node response fluctuation records, extracts continuous response directions, determines whether the directions of adjacent response times are consistent, records continuous direction sequences, and generates node direction change trend sequences; First, set the sampling period to every 0.5 seconds, and perform periodic response sampling on the nodes. Continuous sampling for 10 seconds can obtain 20 groups of response time data. For example, the response sequence is 12ms, 14ms, 14ms, 13ms, 11ms, 10ms, 10ms, etc. After the system reads each group of response time, it compares the response difference with the previous group to determine whether there is a change. When the response time changes from 12ms to 14ms, the difference is positive, and it is determined to be a positive fluctuation. If it drops from 14ms to 13ms, it is a reverse fluctuation. When the two groups of response times are the same, the direction is not recorded. The system compares each group of sampling pairs in turn, and marks each point with a change as "positive" or "reverse" according to the direction of change, thereby constructing a response direction sequence, such as " Forward, none, reverse, reverse, none", where "none" means the response time of two consecutive points is consistent. The system skips the record and only continuously splices the "forward" and "reverse" directions into a direction subsequence for trend segment division. In trend segment identification, the first valid direction is recorded from the beginning of the sequence and the initial segment index is set. Each direction is then compared to see if it is consistent with the previous direction. If they are consistent, they are included in the current segment. If they are inconsistent, a new segment is started and the start and end index values ​​and direction label of the previous segment are recorded. For example, three segments are formed: "forward 2 times", "reverse 3 times", and "forward 2 times". Each segment is marked by an index position and combined into a direction change trend sequence. Through this sequence, the continuous direction changes of the node in different response stages can be identified and processed.

[0023] The reversal identification submodule calls the node direction change trend sequence, compares adjacent directions to see if they are opposite, counts the number of direction reversals, and combines this with the node computing power usage to generate the direction reversal intensity result. The reversal identification submodule calls the generated direction change trend sequence, and compares the direction labels of the current segment and the next segment from the starting position of the sequence. For example, when the trend sequence is "forward 3 times", "reverse 2 times", "reverse 2 times", "forward 1 time", and "reverse 1 time", the system compares the directions of the first and second segments, the second and third segments, and other adjacent segments respectively. If the directions are opposite, it is recorded as a reversal, and if they are the same, it is skipped. In the above sequence, the directions of the first and second segments are opposite, and a reversal is recorded. The second and third segments are the same and skipped. The directions of the third and fourth segments are different, and a reversal is recorded again. In this way, the number of reversals is counted, and a total of three direction reversals are identified; then the computing power usage of the node in the current cycle is called, and this value is processed by the node The ratio of the number of tasks to the set maximum processing capacity is used to calculate the value. For example, if the maximum processing capacity of a node cycle is 100 task units and 65 task units are currently being processed, the usage is 0.65. The number of reversals is multiplied by the computing power usage to obtain the direction reversal intensity, i.e. 3×0.65=1.95. This intensity value is used for subsequent stability judgment. To ensure the consistency of the judgment standard, the system presets four reversal intensity intervals, with the maximum number of reversals set to 10 times. Combined with the maximum computing power usage of 1 and the highest intensity of 10, the system is divided into four intervals proportionally: low intensity is (0–2.5), medium intensity is (2.5–5), high intensity is (5–7.5), and extremely high intensity is (7.5–10). The current value of 1.95 is in the low intensity interval.

[0024] The stable state generation submodule obtains the cache read and write backlog and input and output processing volume based on the direction reversal strength results, calculates the cache access and processing ratio, calls the reversal strength comparison frequency and ratio relationship, and generates the node computing power stable state table; The stable state generation submodule further reads the cache read and write backlog and input and output processing data of the node in the current cycle based on the direction reversal strength value to obtain the cache access and processing ratio. For example, if the backlog is 120MB and the processing volume is 300MB, the ratio is 120 / 300=0.4. The ratio value reflects the current cache data processing load. The smaller the ratio, the more timely the processing, and vice versa, it means that the cache load is heavier. The system compares this ratio with the direction reversal strength value and uses a two-dimensional state determination table to complete the stable state identification. The cache ratio is divided into four intervals: low ratio (0–0.3), medium ratio (0. The ratio is 3–0.6), the high ratio is (0.6–0.8), and the extremely high ratio is (0.8–1.0). According to the actual settings, when the ratio is 0.4, it is a medium ratio, and the aforementioned reversal strength value is 1.95, it is a low strength. In the state matching table, the low strength and medium ratio correspond to a stability level of "medium to high". The system uses this level as the current node computing power stability state, and records this state together with the relevant parameter values ​​in the calculation process into the stable state table to form a single record entry, including a reversal strength value of 1.95, a cache ratio value of 0.4, and a stability state level of "medium to high", which is used for the unified management and retrieval analysis of the subsequent node operation status.

[0025] See also Figure 4 and Figure 2 , the task computing power level division module includes: The task parameter extraction submodule obtains the node computing power stability status table, extracts the logic call depth, total number of instruction calls and cache interaction times corresponding to each task, integrates and records the three data according to the task dimension, establishes a task execution structure item set, and generates a task execution indicator set; First, the computing power status field related to each specific task is extracted according to each record in the table, and the task is associated with its corresponding record in the stable state table through the task unique identification code. After obtaining the corresponding record of each task, the three parameters of logic call depth, total number of instruction calls and number of cache interactions are read in turn. The logic call depth is identified by recording the recursion and nesting layers of the task in the multi-layer function or module. For example, if a task calls three nested function modules in succession and the recursion depth of each layer is 1, then its logic call depth is 3. The total number of instruction calls is recorded by the assembly or intermediate instruction counter scheduled by the processor during the task execution cycle. For example, if a task is executed A total of 15,000 intermediate instructions are executed in the process, and the total number of instruction calls is recorded as 15,000. The number of cache interactions refers to the sum of all read and write operations generated by the node cache system for the task. For example, if the cache is read 90 times and written 60 times, the number of cache interactions for the task is 150. The system integrates the above three data into structured records according to the task dimension. Each record is bound to the task ID and the three data fields are classified into a unified data structure to form a set of task execution structure items. After all task records are completed, all structure items are batch summarized to form a unified task execution indicator set. The indicator set is used for subsequent classification, comparison and hierarchical division of the execution intensity of different tasks.

[0026] The intensity difference calculation submodule calls the task execution indicator set, calculates the difference between the logic call depth and the number of cache interactions in each record, and generates a task intensity level comparison table based on the difference matching logic and cache difference interval; First, the logical call depth and cache interaction count fields are read in each record, and the numerical difference calculation is performed. The difference value item is obtained by subtracting the cache interaction count from the logical call depth. For example, if the logical call depth of a task is 8 layers and the cache interaction count is 6 times, the calculation result is 2. If the logical call depth is 5 layers and the cache interaction count is 9 times, the result is -4. The system matches the preset logic and cache difference intervals based on this difference, and sets the difference intervals as follows: interval A is the difference greater than or equal to 5, interval B is the difference between 2 and 5, interval C is the difference between -2 and 2, and interval D is the difference less than -2. Each interval corresponds to a task intensity level, namely "high intensity", "medium intensity", "normal intensity" and "low intensity". The difference value item is compared with the above intervals. If the difference is 6, it matches interval A, corresponding to "high intensity". If the difference is -3, it matches interval D, corresponding to "low intensity". After completing the matching of each record, the system marks the level label corresponding to the task after the record item. After all task records complete the difference calculation and level assignment, the system generates a task intensity level comparison table. Each record in the table contains the task ID, difference, and matching level fields, which are used for the hierarchical classification of execution intensity in the next stage.

[0027] The level list generation submodule filters and classifies the execution intensity of differentiated level tasks based on the level labels in the task intensity level comparison table and the total number of instruction calls, constructs a computing power distribution record under the level tasks, and generates a hierarchical list of task computing power usage; According to the level labels in the task intensity level comparison table, the level identification fields are traversed one by one, the task records under each level are extracted, and the total number of instruction calls is summarized and classified. Under each level category, the system filters the record items with the same level label field in all task records, and reads the corresponding total number of instruction calls parameters to perform hierarchical statistics and task distribution construction. For example, there are 10 tasks under the "high intensity" level, and the total number of instruction calls are 13,000, 14,500, 16,000, etc. The system calculates the average number of instruction calls and the distribution range based on these values, and then groups them into computing power usage distribution segments. For example, the total number of calls below 14,000 is classified as "low segment", and 14,000 is classified as "low segment". Up to 16,000 are classified as the "normal segment", and above 16,000 are classified as the "high segment". The threshold setting is based on the mean standard of the 95% confidence interval in empirical data or training set statistics. For example, if the average total number of calls in multiple task samples is 15,000 and the standard deviation is set to 2,000, the normal segment can be set to the range of 13,000 to 17,000. Below 13,000 or above 17,000 is the abnormal segment. Finally, the system organizes the call distribution of tasks of each level in the form of structured records into a hierarchical list of task computing power usage. Each record contains four fields: level label, task ID, total number of instruction calls and distribution segment identifier, which are used for reference in subsequent node resource scheduling and task allocation strategies.

[0028] See also Figure 5 and Figure 2 , the computing power scheduling adaptation module includes: The node status extraction submodule obtains the task computing power usage hierarchy list, extracts the resource status of the online node, reads the computing power usage value of the node, calculates the computing power usage distribution index, and structurally integrates the related indicators with the node identifier to generate an online resource usage index set; The calculation formula for computing power distribution index is as follows: ; in, Represents the computing power distribution index of the k-th node, Represents the computing power usage of the k-th node on the m-th task, Represents the average computing power usage of all tasks of the k-th node, Represents the total cache usage of the kth node, Represents the total number of tasks currently processed by the k-th node, Represents a small constant that prevents division by zero; The computing power occupation distribution index is used to measure the computing power use balance and cache usage of a node in multiple tasks, and is a key index for evaluating the load state of the node. The calculation formula thereof comprehensively considers the computing power use value, average use value, cache occupation amount and task number of the node in different tasks; The formula logic is as follows: First, the difference between the computing power use value of the node on all tasks and the average value thereof is squared and summed (to measure the use imbalance degree); Then, the cache occupation amount of the node divided by the task number is added as the second part (to measure the resource density); Finally, a small constant is set to avoid division by zero; The larger the result is, the more uneven the computing power distribution of the node or the higher the cache pressure is. The scheduling system can identify the node with unreasonable load according to the result, and avoid the participation of the node in high-intensity task allocation; The computing power occupation distribution index of the kth node is calculated as follows: The following parameters are required: : The computing power use value of the kth node on the mth task. The CPU use rate data is collected by a node monitoring system (such as Prometheus), and the unit is percentage (%). For example, the CPU use rates of node k on tasks 1 to 4 are 65%, 70%, 60% and 75% respectively.

[0029] : The average computing power use value of the kth node on all tasks. The calculation formula is as follows: ; Substitute the above data: ; : The total cache occupation amount of the kth node. The cache usage is obtained by a cache monitoring tool (such as the RedisINFO command), and the unit is GB. It is assumed that the cache occupation of node k is 2.5 GB.

[0030] n: The total number of tasks currently processed by the kth node. It is obtained by querying a task scheduling system (such as Kubernetes), and it is known that node k currently processes 4 tasks, i.e. n = 4.

[0031] δ: A small constant to prevent division by zero, which is set to 0.01.

[0032] Substitute the above numerical values into the formula to calculate : The first part is calculated as follows: ; The second part is calculated as follows: ; therefore, The calculation result is: ; The result shows that the computing power usage of the kth node under the current task allocation is relatively balanced (the first part is 0), but the cache occupancy is relatively high (the second part is 1.2438). It is necessary to pay attention to the allocation and usage of cache resources to avoid potential performance bottlenecks.

[0033] The resource matching judgment submodule calls the online resource occupancy indicator set, compares the current computing power occupancy value of the node with the task intensity value according to the computing power intensity level record in the task list, and generates a task adaptation node list; The resource matching judgment submodule calls the online resource occupancy index set, and while traversing all online node records, reads the task intensity level record in each task list, and compares the adaptability between the task level and the distribution level of the node's current computing power occupancy value. The matching rule is: if the task is "high intensity", the node must be in the "low occupancy" or "medium occupancy" interval; if the task is "medium intensity", it can adapt to "medium occupancy" or "high occupancy" nodes; if the task is "low intensity", it can accept nodes of any occupancy level. In this process, the system does not directly use fuzzy logic, but adopts value The interval demarcation is matched with the task-node level comparison table. Assuming that the task intensity level is "high intensity", the matching node computing power occupancy value must be less than 0.7. If the occupancy value of a node is 0.65, the adaptation is determined to be successful. Conversely, if the node occupancy value is 0.85, the condition is not met, and the node is filtered out. After the system completes the adaptation judgment between all tasks and nodes, it generates a corresponding task adaptation node list for each task. Each record in the list contains the task ID, task intensity level, node ID and node computing power occupancy value fields, which are used for subsequent cache status judgment and node screening.

[0034] The computing power node screening submodule calls the task adaptation node list to determine whether the cache occupancy ratio of the node is saturated, removes nodes with cache occupancy in the saturated range, retains the remaining nodes as schedulable targets, and generates the set of computing power nodes available for the task; The computing power node screening submodule calls the task adaptation node list and further extracts the cache usage of the adaptation nodes under each task record. It reads the cache occupancy ratio of each node in the current cycle. This value is calculated by dividing the node's cache used capacity by the maximum cache capacity. For example, if a node has a total cache capacity of 500MB and currently uses 450MB, the occupancy ratio is 0.9. The system sets cache occupancy intervals as follows: normal interval (0–0.8), warning interval (0.8–0.9), and saturation interval (0.9–1). These intervals are determined by the critical thresholds for cache response timeouts in the system's operation history. If the node's cache occupancy ratio is greater than or equal to 0.9, the cache is considered saturated. The system traverses each node in the task adaptation node list in sequence. If a node is in the saturation interval, it is removed from the candidate target node for the current task. The remaining nodes are retained as the valid target node set, ultimately forming the set of available computing power nodes for each task. Each record in this set contains fields such as task ID, node ID, computing power occupancy value, and cache occupancy ratio, which are used by the scheduler to implement task allocation based on resource conditions.

[0035] See also Figure 6 and Figure 2 , the link load blocking detection module includes: The abnormal record monitoring submodule calls the available computing power node set of the task, monitors the number of transmission anomalies of the connected nodes, records the growth of abnormal events in chronological order, eliminates records without continuous changes, and generates an abnormal event growth sequence; The abnormal record monitoring submodule calls the task available computing power node set, and the system collects abnormal data for each connected node in the task schedulable node list. The collection content is the number of transmission abnormalities that occurred in each node in each time period. The abnormality type is limited to abnormal state events caused by communication packet loss, response delay or protocol interruption. The system collects abnormal count values ​​for each node at a fixed time interval within the sampling period. For example, the number of abnormalities of node A in five consecutive periods is 1, 3, 4, 4, and 6 times respectively. The system summarizes the number of abnormalities of the node in chronological order as an abnormal event time series, and then judges the difference between adjacent data in the series. If two consecutive sampling If the difference in the number of abnormal times of a point is 0, it is considered as a record item with no change. For example, in the above example, from 3 to 4 is +1, from 4 to 4 is 0, and from 4 to 6 is +2. The system removes the time points corresponding to the middle pair of values ​​​​(4,4) from the sequence and only retains the items with positive changes, that is, the record items must meet the continuous growth condition. The record items constitute the abnormal event growth sequence. If the number of node abnormalities shows a downward or stagnant trend, the record of the corresponding time point is not retained. In this way, the system traverses all connected nodes in the available nodes of the task one by one, extracts all record items with abnormal growth trends, and summarizes them by node grouping to form an abnormal event growth sequence for subsequent queue status evaluation.

[0036] The queuing ratio calculation submodule calls the abnormal event growth sequence, extracts the task queuing amount of the connection node and the node bearable task capacity, calculates the queuing ratio index value, judges whether the node ratio is continuously rising, filters the rising trend record, and generates the queuing ratio change trend; The queuing ratio index value calculation formula is specifically: ; Wherein, represents the queuing ratio index value of the first connection node at time , represents the task queuing amount of the first connection node at time , represents the weight of the first to-be-queued task at time , represents the task expected processing time of the first to-be-queued task at time , represents the maximum bearable task capacity of the first connection node at time , represents the number of all to-be-queued task items associated at time ; The queuing ratio index value reflects the proportion of the current task queuing degree of the node relative to its maximum task bearing capacity, and is an important signal of whether the link is congested. Its calculation considers the number of tasks, weight, expected processing time and node bearable task upper limit and other factors; The calculation logic is as follows: For all to-be-processed tasks, a weighted index is first calculated according to the weight and expected processing time; Then, the queuing task amount of the current node is compared with its processing upper limit to obtain the queuing degree; If the index rises at a plurality of time points in succession, it indicates that the node is more and more congested; The index can assist in identifying the network bottleneck node, so that the scheduler can timely adjust the flow direction to avoid congestion; At time , for the first connection node, the following parameters are collected and calculated: : The current task queuing amount of the node obtained through real-time monitoring system, in the unit of task number.

[0037] : The maximum task capacity that a node can carry, determined by the node's hardware configuration and historical load data, and the unit is the number of tasks.

[0038] : No. The estimated processing time of the queued task, in seconds, based on historical processing data and task complexity assessment.

[0039] : No. The weight of the task to be queued is set according to the priority and resource consumption of the task. It has no unit.

[0040] : The total number of tasks currently waiting to be queued, calculated in real time.

[0041] Set the specific values ​​as follows: ; ; ; 、 、 ; 、 、 ; Calculation steps: Calculate the weighted square root processing time for each task: ; ; ; Sum to get the total weighted square root processing time: ; ; ; Sum to get the total weighted square root processing time: ; Substitute the formula to calculate the queue ratio index value: ; This result shows that at time Moment, The queue ratio index for the connected nodes is 0.424, indicating the relative difference between the current task queue and the node's capacity. This index can be used to determine node load and assist in resource scheduling and load balancing strategies.

[0042] The blocking list generation submodule identifies the node identifiers corresponding to the continuously rising records based on the queue ratio change trend, summarizes the abnormal sequences and queue status, removes the node items that do not meet the conditions, and generates a link transmission blocking list; The blocking list generation submodule reads the node identifier from each record based on the queue share trend records. It then cross-compares the node's cycle number in the abnormal event growth sequence with the cycle number in the queue share sequence. If there is an overlap between the two and the queue share continues to rise during the overlapping period, the system identifies the node as a suspected blocking target. It then reads the cumulative number of abnormal growth events for the node within the overlapping period, setting a threshold of 5 and a final queue share value greater than or equal to 0.85 as the blocking confirmation condition. If a node has 6 abnormal growth events within 4 periods and a final queue share value of 0.88, both conditions are met and the system identifies the node as an actual transmission blocking node. Conversely, if the cumulative number of abnormal events is less than the threshold or the share value does not exceed the set threshold, the record is discarded and does not enter the final result. After the system completes the judgment of all nodes, it summarizes the identifier, cumulative number of abnormal events, final queue share value, and trend period segment number of the confirmed blocked node and integrates them into a structured record. This list ultimately generates a link transmission blocking list, which serves as an input reference for the system to determine link performance bottlenecks. Each record in the list provides a complete and traceable link status.

[0043] See also Figure 7 and Figure 2 , the computing power flow reprogramming module includes: The blocking node positioning submodule extracts the node number corresponding to each blocking record based on the link transmission blocking list, organizes the node number and blocking information into corresponding mapping items, removes duplicate numbers and summarizes them to generate a blocking node number set; The blocked node location submodule first reads the node identification field marked in each record based on the link transmission blocking list. It then uses the node number as the primary key to extract and generate a preliminary number list. In this list, there may be multiple records corresponding to the same node number. The system performs a deduplication operation on the number list and merges all duplicate node numbers using a hash structure to retain only unique identification items. For example, the preliminary records are (N1, N2, N3, N1, N4, N2). After deduplication, the system forms (N1, N2, N3, N4). Each node number is then mapped and assembled with the corresponding blocking information in the original list. For example, the cumulative number of anomalies corresponding to node N1 is 6 and the queue ratio is 0.88, while the cumulative number of anomalies corresponding to node N2 is 5 and the queue ratio is 0.9. Each number is mapped to its corresponding blocking information one by one, and a mapping structure is constructed. The system summarizes all mapping items and forms a structured data set. Each record contains fields such as the node number, anomaly count value, and the final queue ratio value. Finally, the blocked node number set is generated, which is used to lock the target node range in the subsequent node replacement process.

[0044] The replacement node screening submodule calls the blocked node number set, reads the idle computing power and task load of the currently unassigned nodes in the Zhiwang computing power resource management, calculates the computing power difference index of the unassigned nodes, compares the ratios of all unassigned nodes, selects the nodes with the dominant ratio, marks them with numbers, and generates a set of candidate replacement nodes; The calculation formula for the computing power difference index of unassigned nodes is as follows: ; in, Representative number is The computing power difference indicator of the unassigned nodes, Representative number is The unassigned nodes in Idle computing power value under dimension, Representative number is The number of tasks carried by unassigned nodes, Represents the number of idle computing power dimensions involved in the calculation, Representative number is The scheduling delay weight value corresponding to the unassigned node, is a positive offset factor; The computing power difference indicator is used to measure the resource advantage of unallocated nodes and is used to select the optimal replacement node when a task is blocked. Its formula combines the idle computing power value of various dimensions of the node (such as CPU, GPU, etc.), the number of current tasks carried, the scheduling delay weight, and other factors to reflect the ratio between the total amount of available computing power of the node and its load capacity. The calculation steps are: The sum of the idle value and delay weight of the node in all resource dimensions is accumulated as the numerator; Use the current number of tasks of the node plus a small constant as the denominator; The higher the index value, the more idle resources the node has and the lighter the load, making it suitable for use as a replacement node. In this way, the system can dynamically select alternative nodes when the network is congested to maintain scheduling continuity; 1. Parameter acquisition and quantification method: (No. The unassigned nodes in Idle computing power value in each dimension): By real-time monitoring of the node's CPU, GPU, memory and other resource usage, the idle computing power values ​​of each dimension can be obtained.

[0045] (No. The number of tasks carried by unassigned nodes): The number of tasks currently being processed by the statistics node is used as the task load quantity.

[0046] (Number of idle computing power dimensions involved in the calculation): Determined by the number of monitored resource dimensions, such as CPU, GPU, memory, etc.

[0047] (No. The scheduling delay weight value corresponding to the unassigned nodes): By measuring the network delay, processing delay and other indicators of the node, the scheduling delay is calculated and the corresponding weight value is assigned according to the degree of delay.

[0048] (positive offset factor): Set to a small positive number, such as 0.01, to avoid the denominator being zero.

[0049] Specific numerical settings: Hypothetical Node The idle computing power of in three dimensions is: (CPU idle computing power); (GPU idle computing power); (memory idle computing power); node Number of tasks carried: ; The number of idle computing power dimensions involved: ; node The scheduling delay weight value of: ; Positive offset factor: ; Formula calculation process: Calculate the first term: ; Calculate the second term: ; Adding the two together, we get Value: ; The result shows that the computing power difference index of unallocated nodes is 12.883. This value is used to measure the degree of difference between the idle computing power of the node and the number of tasks it carries. The larger the value, the greater the difference. It can be used to screen nodes with a dominant ratio, mark them by number, and generate a set of candidate nodes for replacement.

[0050] The task flow adjustment submodule calls the replacement candidate node set and redirects the corresponding tasks to the replacement nodes based on the connection path structure of the original blocking node. It then restructures the task scheduling relationship, updates the node connection and task allocation path, and generates computing resource management results. First, the system reads each node ID in the blocked node ID set and extracts the upstream and downstream node relationships connected to the node in the topology structure to construct its original connection path structure. After the system identifies a complete path, for example, the path consists of A→N1→B, where N1 is the blocked node. The system replaces the node ID at position N1 in the original path with the corresponding candidate node ID, such as M3, and reconstructs the path to A→M3→B. During this replacement process, the system needs to check whether the candidate node currently has a physical connection with the upstream and downstream nodes of the task. If not, the system issues a connection instruction through the console to instruct the node to establish link communication. After the connection is successful, the system updates the scheduling mapping record in the task scheduling relationship table, changing the original binding node N1 of the task to M3. The task number remains unchanged, only the target node field is updated, the path structure is updated simultaneously, and the new connection relationship is registered in the node access table. All changes are synchronously written to the computing resource management system log. The system finally completes the task scheduling target update and connection structure reorganization, and outputs the structure reorganization result record to form the final computing resource management result dataset. Each record contains fields such as the original blocked node, replacement node, original connection path, updated path, and task number.

[0051] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A smart network computing resource management system, characterized in that: The system comprises: The node response load analysis module uses node response fluctuation records from the Zhiwang computing resource management to analyze the direction of continuous changes, the number of adjacent response time reversals, and the cumulative number of reversals. It then combines node computing power usage, cache read and write backlogs, and input and output processing ratios to generate a node computing power stability table. The task computing power level classification module extracts the task logic call depth, total number of instructions and cache interaction times based on the node computing power stable state table, calculates the difference and classifies the levels, matches the task intensity, and generates a task computing power usage level list; The computing power scheduling adaptation module refers to the task computing power usage hierarchy list, analyzes the computing power usage and cache usage of online nodes, selects nodes whose computing power margin matches the task intensity, eliminates nodes whose cache is close to saturation, and generates a set of nodes with available computing power for the task; The link load blocking detection module monitors transmission anomalies and queue volume based on the set of available computing nodes for the task, analyzes the continuous increase in the backlog ratio, and generates a link transmission blocking list; The computing power flow re-editing module refers to the link transmission blocking list and the unassigned nodes, selects the ratio nodes to replace the connection, adjusts the task flow, and generates the computing power resource management result.

2. The intelligent network computing resource management system according to claim 1, characterized in that: The node computing power stability state table includes the number of direction reversals, response time fluctuation frequency, cache access processing ratio, and node computing power usage ratio. The task computing power usage hierarchy list includes the logic call depth interval, instruction call quantity level, cache interaction frequency range, and call interaction difference segment. The task available computing power node set includes computing power margin nodes, cache load controllable nodes, and online adjustable points. The link transmission congestion list includes abnormal transmission frequency, connection node queuing intensity, and task backlog growth ratio. The computing power resource management results include blocked node numbers, replacement node numbers, and adjusted task flow paths. The nodes whose computing power margin matches the task intensity refer to online nodes whose remaining computing power is sufficient to meet the intensity requirements of the specified task and whose cache occupancy is not close to saturation; The ratio node replaces the connected node: refers to selecting a node with high task transmission efficiency from unassigned nodes and replacing the original connected node in the case of link congestion.

3. The intelligent network computing resource management system according to claim 1, characterized in that: The node response load combing module includes: The response fluctuation extraction submodule obtains node response fluctuation records, extracts continuous response directions, determines whether the directions of adjacent response times are consistent, records continuous direction sequences, and generates node direction change trend sequences; The reversal identification submodule calls the node direction change trend sequence, compares whether adjacent directions are opposite, counts the number of direction reversals, and generates a direction reversal intensity result based on the node computing power usage; The stable state generation submodule obtains the cache read and write backlog and input and output processing volume based on the direction reversal strength result, calculates the cache access and processing ratio, calls the reversal strength comparison frequency and ratio relationship, and generates a node computing power stable state table; The node computing power usage refers to the total amount of computing power resources consumed by the node to actually execute tasks within a time period.

4. The intelligent network computing resource management system according to claim 1, characterized in that: The task computing power level division module includes: The task parameter extraction submodule obtains the node computing power stability state table, extracts the logic call depth, total number of instruction calls and cache interaction times corresponding to each task, integrates and records the three data according to the task dimension, establishes a task execution structure item set, and generates a task execution indicator set; The intensity difference calculation submodule calls the task execution indicator set, calculates the difference between the logic call depth and the cache interaction times in each record, and generates a task intensity level comparison table based on the difference matching logic and cache difference interval; The level list generation submodule screens and classifies the execution intensity of differentiated level tasks based on the level labels in the task intensity level comparison table and the total number of instruction calls, constructs a computing power distribution record under the level tasks, and generates a task computing power usage level list; The task execution structure item set refers to the integrated set of three core parameters corresponding to each task: logic call depth, total number of instruction calls and number of cache interactions; The difference matching logic refers to a matching rule for mapping the difference between the logic call depth and the number of cache interactions to a preset task intensity level; The cache difference interval refers to the segmentation of the strength level corresponding to the logical depth preset according to the task type and the cache interaction difference; The execution intensity refers to the degree to which a task consumes node computing resources during execution.

5. The intelligent network computing resource management system according to claim 1, characterized in that: The computing power scheduling adaptation module includes: The node status extraction submodule obtains the task computing power usage hierarchy list, extracts the resource status of the online node, reads the computing power occupancy value of the node, calculates the computing power occupancy distribution index, structurally integrates the associated index with the node identifier, and generates an online resource occupancy index set; The resource matching judgment submodule calls the online resource occupancy indicator set, compares the current computing power occupancy value of the node with the task intensity value according to the computing power intensity level record in the task list, and generates a task adaptation node list; The computing power node screening submodule calls the task adaptation node list to determine whether the cache occupancy ratio of the node is saturated, removes nodes with cache occupancy in the saturated range, retains the remaining nodes as schedulable targets, and generates a set of computing power nodes available for the task; The computing power utilization value refers to the amount and proportion of computing power resources occupied by the node being executed; The node whose cache occupancy is in the saturation range refers to a node whose cache resource usage has reached and exceeded a preset upper limit threshold, resulting in a performance bottleneck.

6. The intelligent network computing resource management system according to claim 5, characterized in that: The calculation formula for the computing power occupancy distribution index is as follows: ； in, Represents the computing power distribution index of the k-th node, Represents the computing power usage of the k-th node on the m-th task, Represents the average computing power usage of all tasks of the k-th node, Represents the total cache usage of the kth node, Represents the total number of tasks currently processed by the k-th node, Represents a small constant that prevents division by zero.

7. The intelligent network computing resource management system according to claim 1, characterized in that: The link load blocking detection module includes: The abnormal record monitoring submodule calls the set of available computing power nodes for the task, monitors the number of transmission anomalies of the connected nodes, records the growth of abnormal events in chronological order, eliminates records without continuous changes, and generates an abnormal event growth sequence; The queue ratio calculation submodule calls the abnormal event growth sequence, extracts the task queue volume corresponding to the connection node and the node's task capacity, calculates the queue ratio index value, determines whether the node ratio is continuously increasing, filters the upward trend records, and generates the queue ratio change trend; The blocking list generation submodule identifies the node identifiers corresponding to the continuously rising records according to the queue ratio change trend, summarizes the abnormal sequences and queue status, eliminates the node items that do not meet the conditions, and generates a link transmission blocking list; The record items without continuous changes refer to the node record data where the number of abnormal events does not continue to increase in adjacent time periods; The upward trend record refers to a record item in which the queue ratio index value shows an increasing trend at consecutive time points; The node items that do not meet the conditions refer to node data that do not show a continuous upward trend in the abnormal growth and queue ratio analysis and do not meet the congestion judgment criteria.

8. The intelligent network computing resource management system according to claim 7, characterized in that: The calculation formula of the queue ratio index value is specifically as follows: ； in, Represents the connected node at time The queue ratio index value at the time, Representative connected nodes at time The number of tasks queued at the moment, Representative time Moment The weight of the task to be queued, Representative time Moment The estimated processing time of the tasks to be queued, Representative connected nodes at time The maximum task capacity that can be carried at any time, Indicates time The total number of queued tasks associated with the moment.

9. The intelligent network computing resource management system according to claim 1, characterized in that: The computing power flow reprogramming module includes: The blocking node positioning submodule extracts the node number corresponding to each blocking record according to the link transmission blocking list, organizes the node number and the blocking information into corresponding mapping items, removes duplicate numbers and summarizes them to generate a blocking node number set; The replacement node screening submodule calls the blocked node number set, reads the idle computing power and task load of the currently unallocated nodes in the intelligent network computing resource management, calculates the computing power difference index of the unallocated nodes, compares the ratios of all unallocated nodes, screens the nodes with the dominant ratios, marks them with numbers, and generates a set of candidate replacement nodes; The task flow adjustment submodule calls the replacement candidate node set, redirects the corresponding tasks to the replacement nodes according to the connection path structure of the original blocking node, restructures the task scheduling relationship, updates the node connection and task allocation path, and generates computing power resource management results; The repeated number refers to the same node number that appears multiple times in the link transmission blocking list and corresponds to multiple blocking records; The numbering and marking of the node with the dominant ratio refers to the individual with the highest ratio of computing power difference index among the unallocated nodes, which is used as the replacement node and its node number is identified.

10. The intelligent network computing resource management system according to claim 9, characterized in that: The calculation formula for the computing power difference index of the unassigned nodes is specifically as follows: ； in, Representative number is The computing power difference indicator of the unassigned nodes, Representative number is The unassigned nodes in Idle computing power value under dimension, Representative number is The number of tasks carried by unassigned nodes, Represents the number of idle computing power dimensions involved in the calculation, Representative number is The scheduling delay weight value corresponding to the unassigned node, A positive offset factor.

Images