Elastic resource scheduling method and system for campus cloud platform

Abstract

The invention discloses an elastic resource scheduling method and system for a campus cloud platform, and relates to the technical field of resource management, and the method comprises the steps: collecting execution state information and a storage resource state, carrying out the stabilization processing, and carrying out the solidification to generate an execution state evidence; performing execution state co-evolution judgment on the execution state evidence, positioning a stage switching boundary in a program execution process, and dividing execution stage intervals; performing staged screening on the execution state evidence according to the execution stage interval to generate an execution state constraint condition; based on the execution state constraint condition, a continuous deviation judgment algorithm is adopted to carry out attenuation updating processing on the continuous border crossing condition, and an execution scheduling trigger state is generated; and according to the execution scheduling trigger state, executing progressive resource scheduling control according to the execution stage interval, and updating the execution state evidence through the cooling constraint mark. According to the method, the operation stability and the resource scheduling precision of the industrial cloud platform in a complex service scene can be improved on the premise of not depending on complex hardware transformation.

Description

Technical Field

[0001] This invention relates to the field of resource management technology, and in particular to a method and system for elastic resource scheduling on a campus cloud platform. Background Technology

[0002] With the continuous development of cloud computing and virtualization technologies, campus information infrastructure is gradually evolving from traditional distributed server architectures to centralized, platform-based campus cloud platforms. Currently, most universities rely on private or hybrid clouds to host various business workloads such as teaching and experimental systems, scientific research computing tasks, virtual simulation environments, and teaching management applications on a unified cloud platform. Existing industrial cloud platform resource scheduling methods mostly follow the elastic scheduling approach in general cloud computing environments, primarily triggering resource scaling operations based on static or instantaneous indicators such as CPU utilization, memory usage, or storage capacity thresholds.

[0003] At the storage resource scheduling level, existing campus cloud platform resource scheduling solutions generally focus on monitoring single indicators such as storage capacity utilization or throughput. On the one hand, storage resource scheduling decisions lack joint analysis of program execution status, failing to incorporate operational characteristics such as thread blocking, I / O access concurrency, and storage access latency into a unified scheduling judgment framework, making it difficult to reflect the actual changes in storage access pressure. On the other hand, storage resource scheduling lacks phased and continuous constraints, making it prone to over-adjustment of storage resources during program execution, thereby reducing the overall stability of teaching and research tasks. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for elastic resource scheduling in a campus cloud platform to solve the problems of difficulty in achieving state collaborative analysis and lack of phased continuous scheduling constraints in storage resource scheduling.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for elastic resource scheduling on a campus cloud platform, comprising: collecting execution status information and storing resource status, and solidifying them into execution status evidence after stabilization processing; performing execution status co-evolution discrimination on the execution status evidence, locating the stage switching boundary in the program execution process and dividing the execution stage interval; performing stage-based screening on the execution status evidence according to the execution stage interval to generate execution status constraints; based on the execution status constraints, using a continuous deviation judgment algorithm to perform attenuation update processing on continuous out-of-bounds situations to generate an execution scheduling trigger state; and a scheduling update module, based on the execution scheduling trigger state, performing progressive resource scheduling control according to the execution stage interval, and updating the execution status evidence through cooling constraint markers.

[0007] As a preferred embodiment of the campus cloud platform elastic resource scheduling method of the present invention, the execution status information includes thread scheduling status, blocking duration, I / O access occupancy ratio and request response latency information. Storage resource status includes available storage capacity status, storage access concurrency status, storage access latency status, and storage access queue status.

[0008] As a preferred embodiment of the elastic resource scheduling method for the campus cloud platform described in this invention, the specific steps for generating execution status evidence are as follows: The execution status information is processed by a sliding window in chronological order to remove transient data that only appears at a single sampling point. Data that maintains a consistent trend across multiple consecutive sampling periods in the execution status information processed by the sliding window is locked and stored to generate execution status evidence.

[0009] As a preferred embodiment of the elastic resource scheduling method for the campus cloud platform described in this invention, the specific steps for performing execution state co-evolution discrimination on the execution state evidence and locating the stage switching boundaries during program execution are as follows. The execution status evidence is timestamped and trend-marked to generate an execution status mark trajectory. Based on the execution state marker trajectory, a sequence of execution state linkage directions is generated by judging the direction of change of execution state at adjacent time points; In the execution state linkage direction sequence, the linkage direction order comparison method is used to compare the arrangement order of linkage directions in adjacent time periods to obtain the stage switching boundary.

[0010] As a preferred embodiment of the campus cloud platform elastic resource scheduling method of the present invention, the division of execution stage intervals refers to segmenting the time range of execution status evidence according to the stage switching boundary, and merging the execution status evidence to form execution stage intervals.

[0011] As a preferred embodiment of the elastic resource scheduling method for the campus cloud platform described in this invention, the specific steps for performing stage-based filtering of execution status evidence according to execution stage intervals are as follows. The execution status evidence is sliced ​​and grouped according to each execution stage interval to obtain the execution status evidence fragment set corresponding to each stage; Based on the execution state evidence fragment set, the intra-stage pattern compression algorithm is used to merge the execution state marker paths that appear repeatedly in the fragments into the shortest representative path to obtain a representative evolution path set. The path coverage priority algorithm is used to select the path subset with the most covered segments from the representative evolution path set to obtain the normal execution path coverage set.

[0012] As a preferred embodiment of the campus cloud platform elastic resource scheduling method of the present invention, the generation of execution state constraints refers to extracting the time sequence state transition relationship from the normal execution path coverage set, encapsulating and solidifying it to generate execution state constraints.

[0013] As a preferred embodiment of the elastic resource scheduling method for the campus cloud platform described in this invention, the step of using a continuous deviation judgment algorithm to perform decay update processing on continuous out-of-bounds situations based on execution state constraints to generate an execution scheduling trigger state is as follows. The execution status evidence is mapped to the execution status constraints to identify whether a cross-stage switching boundary has occurred and to record the corresponding crossing event. The time decay memory method is used to update the time decay memory of crossover events in adjacent time periods to generate a deviation evolution sequence. Based on the deviation evolution sequence, a continuous deviation judgment algorithm is used to determine the temporal distribution pattern of the crossing event and generate the execution scheduling trigger state.

[0014] As a preferred embodiment of the elastic resource scheduling method for the campus cloud platform described in this invention, the step of performing progressive resource scheduling control according to the execution phase interval based on the execution scheduling trigger state, and updating the execution state evidence through cooling constraint flags, is as follows: Read the execution stage interval corresponding to the execution scheduling trigger state, and use the stage-constrained progressive scheduling method to perform progressive resource scheduling control on the storage resource state to generate updated execution state evidence; Based on the updated execution status evidence, calculate the degree of convergence of state deviation within the phase to determine whether the current scheduling behavior has generated feedback on the execution status, and obtain the deviation regression discriminant quantity within the phase. Based on the deviation regression discriminant within the stage, the corresponding execution stage interval is marked with a cooling constraint and merged with the updated execution status evidence to generate scheduling continuation execution status evidence.

[0015] Secondly, this invention provides a campus cloud platform elastic resource scheduling system, comprising: a state evidence module, which collects execution state information and stores resource status, and solidifies it after stabilization processing to generate execution state evidence; a stage identification module, which performs execution state co-evolution discrimination on the execution state evidence, locates the stage switching boundary in the program execution process, and divides the execution stage interval; a state constraint module, which performs stage-based screening of the execution state evidence according to the execution stage interval to generate execution state constraint conditions; a trigger determination module, which, based on the execution state constraint conditions, uses a continuous deviation determination algorithm to perform attenuation update processing on continuous out-of-bounds situations to generate an execution scheduling trigger state; and a scheduling update module, which, based on the execution scheduling trigger state, performs progressive resource scheduling control according to the execution stage interval, and updates the execution state evidence through cooling constraint markers.

[0016] The beneficial effects of this invention are as follows: by constructing evidence of execution status, identifying execution phases, determining continuous deviations, and scheduling cooling constraints, the resource scheduling decision-making is transformed from resource result-driven to execution process-driven; it can effectively improve the operational stability and resource scheduling accuracy of industrial cloud platforms in complex business scenarios without relying on complex hardware modifications. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the elastic resource scheduling method for the campus cloud platform.

[0019] Figure 2 This is a schematic diagram of the campus cloud platform's elastic resource scheduling system.

[0020] Figure 3 A flowchart for generating execution state constraints.

[0021] Figure 4 This is a flowchart for incremental resource scheduling.

[0022] Figure 5 This is a comparison chart of request and response latency.

[0023] Figure 6 This is a comparison chart of scheduling capacity changes. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for elastic resource scheduling of a campus cloud platform, including the following steps: S1: Collect execution status information and storage resource status, and solidify them into execution status evidence after stabilization processing.

[0028] S1.1: Execution status information includes thread scheduling status, blocking duration, I / O access occupancy ratio, and request response latency information.

[0029] Specifically, data is collected through processor scheduling status sensors. During program execution, statistical information of threads in different scheduling states is periodically read. At each sampling moment, the changes in the number of threads in the running, ready, and waiting states are obtained, thereby forming a thread scheduling status that reflects the characteristics of program thread scheduling occupancy.

[0030] By using a thread state time sensor to record the time when a thread enters and exits a blocked state, and by statistically analyzing the length of time the same thread stays in a blocked state continuously, the blocking duration, which reflects the degree of blocking behavior during program execution, can be obtained.

[0031] The number of storage access operations initiated by the program is counted during continuous sampling by the I / O access status sensor, and the total number of operations executed by the program in the same time period is counted simultaneously. By calculating the ratio of the number of storage access operations to the total number of operations, the I / O access occupancy ratio in the corresponding time period is obtained.

[0032] By using a request-response time sensor, the initiation time of a request operation and the completion time of the corresponding response are recorded during program execution. The time difference between the initiation time of the request operation and the completion time of the corresponding response is calculated to obtain the request-response latency, which reflects the changes in the program's request processing efficiency.

[0033] S1.2: Storage resource status includes available storage capacity status, storage access concurrency status, storage access latency status, and storage access queue status.

[0034] Specifically, during program execution, the allocated capacity and remaining available capacity information of storage resources are periodically read, and the read remaining available capacity data are recorded in chronological order to form the available storage capacity status reflecting the changes in available storage space.

[0035] The system counts the number of access requests that are simultaneously in the storage access process within a continuous sampling period, and records the changes in the number of concurrent access requests at each sampling moment, thereby obtaining the storage access concurrency status that reflects the concurrent access pressure on storage resources.

[0036] Record the initiation time and corresponding completion time of storage access requests, and calculate the time interval between the initiation time and the corresponding completion time to form a storage access latency status arranged in chronological order.

[0037] During the sampling process, the number of storage access requests in the waiting state is counted, and the cumulative changes of storage access requests in the queue are recorded, thereby forming storage access queue status information that reflects the changes in the storage access queuing degree.

[0038] S1.3: Perform sliding window processing on the execution status information in chronological order to remove transient change data that only appears at a single sampling point.

[0039] Specifically, the execution status information is arranged in the order of collection time to construct a continuous sliding window. Within each sliding window, the change range of the execution status information corresponding to adjacent time positions is compared sequentially. When the execution status information at a certain time position deviates suddenly relative to the time positions before and after it, for example, when the request response latency is 120 milliseconds, 460 milliseconds and 130 milliseconds in three consecutive time positions, the request response latency corresponding to the middle time position deviates relative to the time positions before and after it, which can be determined as a sudden deviation.

[0040] The execution status information corresponding to the time position is determined to be transient mutation data that only appears at a single sampling point, and the execution status information that is determined to be transient mutation data is removed from the execution status information sequence corresponding to the sliding window, so as to obtain the execution status information sequence after time sequence smoothing.

[0041] It should be noted that transient mutation data refers to data that shows an abnormal change only at a single sampling moment in the time series of execution status information, and does not show the same direction of change or continuous trend of change at adjacent sampling moments. For example, thread scheduling status, blocking duration, I / O access occupancy ratio, or request response latency information may show a sudden increase or decrease at a certain time position.

[0042] S1.4: Lock and store the data in the execution status information that maintains a consistent trend across multiple consecutive sampling periods after the sliding window processing, and generate execution status evidence.

[0043] Specifically, the execution status information after sliding window processing is compared sequentially in chronological order with the changing direction of thread scheduling status, blocking duration, I / O access ratio, and request response latency within adjacent sampling periods. When the values ​​corresponding to thread scheduling status, blocking duration, I / O access ratio, and request response latency increase, the execution status information within the corresponding sampling period is determined to be data with a consistent changing trend.

[0044] Data that is determined to maintain a consistent trend of change will be locked and stored in the corresponding time sequence, and retained as a reliable representation of continuous execution behavior to generate evidence of execution status.

[0045] S2: Perform co-evolutionary judgment on execution status evidence, locate the stage switching boundary in the program execution process, and divide the execution stage interval.

[0046] S2.1: Timestamp and trend mark the execution status evidence to generate an execution status mark trajectory.

[0047] Specifically, the sampling time sequence position is recorded for each execution status evidence, so that each execution status evidence has a unique time sequence identifier; the execution status evidence (thread scheduling status, blocking duration, I / O access occupancy ratio, and request response latency information) corresponding to adjacent time sequence positions are compared in turn by comparing the changes in the values ​​at the previous and subsequent timestamp positions; when the difference between the values ​​at the subsequent and previous timestamp positions is negative, the change is marked as decreasing; when the difference between the values ​​at the subsequent and previous timestamp positions is positive, the change is marked as increasing; when the values ​​at the previous and subsequent timestamp positions are the same, the change is marked as remaining unchanged.

[0048] The execution status evidence, which includes time sequence identifiers and change status markers, is arranged and connected in chronological order to generate an execution status marker trajectory.

[0049] It should be noted that trend marking refers to the discretization of the direction of change of execution status evidence between adjacent time positions, used to characterize the upward, downward or unchanged trend of execution status as it evolves over time; it can transform continuous numerical changes into a comparable and combinable sequence of changes, providing clear and stable evidence of changes for subsequent determination of execution status linkage direction and identification of stage switching boundaries.

[0050] S2.2: Based on the execution state marker trajectory, generate an execution state linkage direction sequence by judging the direction of change of execution state at adjacent time points.

[0051] Specifically, the execution status markers corresponding to two adjacent time positions are read sequentially according to time order. For each type of execution status information, the changes in values ​​between the previous and next time positions are compared. When the value at the next time position is greater than that at the previous time position, it is recorded as an upward change direction; when the value at the next time position is less than that at the previous time position, it is recorded as a downward change direction; and when the value remains unchanged, it is recorded as a stable change direction.

[0052] The changes in thread scheduling status, blocking duration, I / O access occupancy ratio, and request-response latency within the same time interval are recorded and continuously arranged along the time axis to generate a sequence of execution status linkage directions.

[0053] S2.3: In the execution state linkage direction sequence, the linkage direction order is compared with the previous and next time periods by the linkage direction order comparison method to obtain the stage switching boundary.

[0054] Specifically, the execution state linkage direction sequence is read in chronological order. Each time position in the execution state linkage direction sequence corresponds to a set of linkage direction records consisting of changes in thread scheduling state, blocking duration, I / O access occupancy ratio, and request response latency information. The linkage direction records are arranged in a fixed order in the execution state linkage direction sequence.

[0055] Based on the linkage direction records arranged in chronological order, a sliding time window is used to sequentially extract the linkage direction sequence of the execution state, so that each sliding time window contains linkage direction records corresponding to multiple consecutive time positions.

[0056] Within each sliding time window, the execution state linkage direction is arranged in chronological order, and the execution state linkage direction arrangement order in adjacent sliding time windows is compared item by item. When the execution state linkage direction arrangement order in the later sliding time window changes as a whole compared to the earlier sliding time window, for example, if the execution state linkage direction arrangement corresponding to the continuous time position in the earlier sliding time window is upward-upward-stable, and the execution state linkage direction arrangement corresponding to the later sliding time window becomes downward-downward-upward, the time position corresponding to the execution state linkage direction arrangement order is determined as the stage switching boundary.

[0057] It should be noted that the stage switching boundary refers to the dividing position in the time dimension where the execution state evidence changes from one relatively stable form of coordinated change to another during the continuous operation of the program. It is used to characterize the specific time position of the program execution process transitioning from one running stage to the next.

[0058] The linkage direction sequence comparison and discrimination method identifies the position where the execution stage changes in the linkage direction sequence of the execution state, compares the arrangement order of each execution state linkage direction in adjacent time periods, and determines the corresponding time position as the stage switching boundary when the overall arrangement order of the linkage direction changes from a stable combination to a different combination, thus achieving precise positioning of the changes in the program execution stage.

[0059] S2.4: Divide the time range of execution status evidence into segments based on the stage switching boundary, and merge the execution status evidence to form execution stage intervals.

[0060] Specifically, the time positions corresponding to the phase switching boundaries are read in chronological order, and the continuous time range covered between two adjacent phase switching boundaries is determined as an independent time period. The independent time period is used as a filtering condition to extract all execution status evidence whose timestamps fall within the independent time period from the execution status evidence.

[0061] The execution status evidence within the same independent time period is arranged chronologically to form a set of execution status evidence that is temporally continuous and has clear boundaries. Each set of execution status evidence is then used as an execution phase interval.

[0062] It should be noted that the execution phase interval refers to the set of time intervals formed by continuously segmenting the time range corresponding to the execution state evidence according to the phase switching boundary. It is used to characterize the time period when the program is under the same type of execution state evolution characteristics during continuous operation, and provides a basic unit of division for implementing differentiated constraint analysis and resource scheduling for different execution phases.

[0063] S3: Based on the execution stage interval, perform stage-based screening of execution status evidence to generate execution status constraints.

[0064] S3.1: Slice and group the execution status evidence according to each execution stage interval to obtain the execution status evidence fragment set corresponding to each stage.

[0065] Specifically, the start and end times of each execution stage interval in the execution stage interval sequence are read in chronological order, and the continuous time range formed by the start and end times is used as the slicing condition.

[0066] Based on the slicing conditions, when the timestamp in the execution status evidence falls within the continuous time range corresponding to a certain execution stage interval, the execution status evidence is extracted and grouped into the group set corresponding to the execution stage interval, forming a set of execution status evidence fragments that correspond one-to-one with each execution stage interval and have a continuous internal time sequence.

[0067] S3.2: Based on the set of execution state evidence fragments, the intra-stage pattern compression algorithm is used to merge the execution state marker paths that appear repeatedly in the fragments into the shortest representative path to obtain a set of representative evolution paths.

[0068] Specifically, each piece of execution status evidence in the execution status evidence fragment set is read in chronological order, and the execution status marks corresponding to consecutive time positions within the same execution status evidence fragment are concatenated in chronological order according to the execution status mark trajectory to form an execution status mark path; within the same execution stage interval, when multiple execution status mark paths in the execution status evidence fragment set successively show consecutive repetition of execution status mark paths corresponding to different execution status evidence fragments in chronological order and the mark arrangement is completely consistent, they are identified as mergeable paths.

[0069] Multiple execution state marker paths containing the same mergeable path are compressed and merged into a single shortest representative path (the execution state marker path containing the fewest execution state markers); all execution state marker paths within the execution state evidence fragment set are counted to obtain a representative evolutionary path set composed of multiple shortest execution state marker paths.

[0070] It should be noted that the intra-stage pattern compression algorithm refers to comparing and merging the execution state marker paths formed over time in the execution state evidence within the same execution stage interval. By identifying recurring state evolution paths and merging them into the shortest state evolution path that can represent the main operational characteristics of the stage, it provides a compact and representative path basis for the extraction of subsequent execution state constraints.

[0071] S3.3: Use the path coverage priority algorithm to select the path subset with the most covered segments in the representative evolution path set to obtain the normal execution path coverage set.

[0072] Specifically, each representative evolution path in the set of representative evolution paths is read in chronological order, and the number of execution state evidence fragments that can be completely matched (the start time position and the end time position are consistent) for each representative evolution path is counted. The representative evolution paths are sorted in descending order of the number of covered fragments, and the representative evolution path with the largest number of covered fragments is selected as the first path. The execution state evidence fragments that have been covered by the representative evolution path are marked simultaneously.

[0073] The matching relationships between the uncovered execution status evidence fragments and the remaining representative evolutionary paths are re-statistically analyzed, and the representative evolutionary path with the largest number of covered fragments is selected again within the scope of the uncovered execution status evidence fragments; until all execution status evidence fragments in the execution status evidence fragment set are covered by representative evolutionary paths, and all representative evolutionary paths are statistically analyzed to obtain the normal execution path coverage set (e.g., multiple representative evolutionary paths composed of execution status marker sequences, and the path set structure that forms a coverage relationship with the execution status evidence fragments).

[0074] It should be noted that the path coverage priority algorithm refers to counting the number of execution state evidence fragments that each representative evolution path can cover, sorting them according to the coverage range from large to small, prioritizing the selection of evolution paths that can cover more execution state evidence fragments, and gradually forming a set of paths for the main execution behaviors within the coverage stage, providing a stable and representative path foundation for the construction of execution state constraints.

[0075] S3.4: Extract the temporal state transition relationships from the normal execution path coverage set, encapsulate and solidify them, and generate execution state constraints.

[0076] Specifically, each representative evolution path in the normal execution path coverage set is read, and two adjacent execution state markers in the representative evolution path are read in chronological order. The continuous connection process between the previous execution state marker and the next execution state marker on the time axis is recorded as a time-sequential state transition relationship.

[0077] The temporal state transition relationships of all representative evolution paths in the normal execution path coverage set are statistically analyzed to obtain the set of allowed state change processes. For the set of allowed state change processes, state change processes that occur repeatedly in multiple representative evolution paths are retained only once, while keeping the order of state changes unchanged, to obtain the sorted set of allowed state change processes. The fields of the sorted set of allowed state change processes are aligned to generate execution state constraints.

[0078] S4: Based on the execution state constraints, a continuous deviation judgment algorithm is used to perform decay update processing on continuous out-of-bounds situations to generate the execution scheduling trigger state.

[0079] like Figure 5 As shown, the comparison results of request response latency over time under different resource scheduling strategies are illustrated. During the experiment, a request load sequence containing sudden requests, phased business load switching, and instantaneous abnormal disturbances was constructed to continuously monitor the execution status of the campus cloud platform. The collected request response latency data was stabilized before recording. Control group 1 adopted a resource scheduling scheme driven by resource results, using request response results or resource utilization exceeding limits as direct trigger conditions, and performing resource adjustments after detecting abnormal performance indicators. Control group 2 adopted a scheduling scheme without introducing cooling constraints, and did not set time or state interval restrictions for adjacent scheduling behaviors when responding to changes in performance indicators. The experimental group adopted the elastic resource scheduling method based on execution status evidence and execution stage intervals proposed in this invention, comprehensively considering the evolution and continuous deviation characteristics of the execution stage during the scheduling decision process. In contrast, the experimental group maintained a relatively stable request response latency during high load and stage switching, with a reduced peak amplitude. By combining the local magnified area, it can be further observed that the method of the present invention effectively avoids over-scheduling caused by instantaneous disturbances by suppressing continuous out-of-bounds behavior, thus demonstrating the technical effect of the execution process-driven resource scheduling method in improving operational stability.

[0080] like Figure 5The results show a comparison of request response latency over time under different resource scheduling strategies. During the experiment, a request load sequence including sudden requests, phased business load switching, and instantaneous abnormal disturbances was constructed to continuously monitor the platform's execution status and record the request response latency information after stabilization. When using the traditional resource result-driven scheduling strategy, the request response latency exhibits a significant increase in peak value and aggravated fluctuations during high-load phases. While the scheduling method without cooling constraints can alleviate the latency increase to some extent, multiple scheduling triggers still occur during continuous abnormal phases, leading to frequent fluctuations in the latency curve. In contrast, the elastic resource scheduling method based on execution status evidence and execution phase intervals maintains a stable request response latency with reduced peak amplitude during high load and phase switching. Further observation using a magnified local area reveals that the method of this invention can effectively suppress latency spikes caused by continuous boundary violations and avoid over-scheduling due to instantaneous disturbances, thus demonstrating the technical effectiveness of execution process-driven resource scheduling in improving operational stability.

[0081] S4.1: Map execution state evidence to execution state constraints, identify whether a cross-stage switching boundary has occurred, and record the corresponding cross-stage event.

[0082] Specifically, execution status evidence is selected from the execution phase interval. For each piece of execution status evidence, the process of changing adjacent execution status markers is compared item by item with the allowed state change process in the execution status constraint. When the state change process recorded in the execution status evidence is consistent with any allowed state change process in the execution status constraint, the state change process is determined to be within the allowed range.

[0083] When the state change process recorded in the execution state evidence is not included in the allowed state change process defined by the execution state constraints, the state change process is determined to have crossed the stage switching boundary, and the corresponding time position, the execution state mark before the change and the execution state mark after the change are recorded to form a crossing event.

[0084] S4.2: Using the time decay memory method, the time decay memory is updated for crossover events in adjacent time periods to generate a deviation evolution sequence.

[0085] Specifically, the system sequentially reads the time-based event records corresponding to each time position from the generated time-based events, and updates the system based on the interval between adjacent time-based events on the time axis (whether the time positions are consecutive and whether there are any un-occurred time-based events between two time-based events). The system uses a time decay memory method: when both adjacent time positions have recorded time-based events and the time positions are consecutive, the corresponding time-based events are continuously retained; when neither adjacent time position has recorded time-based events and the time positions are not consecutive, the count value of the time-based event corresponding to the previous time position is reduced by one count unit.

[0086] Calculate the deviation evolution representation of the execution traversal event at each time position in the traversal event sequence, and statistically analyze the deviation evolution representation of all time positions to generate a deviation evolution sequence.

[0087] The expression for calculating the deviation evolutionary characterization is: ; in, To deviate from evolutionary characterization, For time location The count value of time travel events. This is the index of the sequence number of the consecutive traversed segments. For time location index, This is the time position index between the last time no time travel event occurred and the current time position. For time location Whether to record the binary indicator of the traversal event.

[0088] It should be noted that when there is a state change process that crosses the stage switching boundary in the execution status evidence, the binary indicator is 1; when there is no state change process that crosses the stage switching boundary in the execution status evidence, the binary indicator is 0.

[0089] The expression for calculating the deviation evolutionary characterization consists of the time position index and the event count. The deviation evolutionary characterization is obtained through the same dimension operation, and the overall dimensions remain consistent.

[0090] The time decay memory method refers to recording each crossover event in chronological order and updating the recorded crossover events one by one as time progresses, so that the influence of crossover events that have occurred and have not recurred gradually weakens, forming a deviation evolution sequence that can reflect the persistence and decay of deviation evolution characteristics.

[0091] S4.3: Based on the deviation evolution sequence, the continuous deviation judgment algorithm is used to determine the time distribution pattern of the crossing event and generate the execution scheduling trigger state.

[0092] Specifically, the deviation evolution characteristics corresponding to each time position are read sequentially from the deviation evolution sequence, and the ratio of the deviation evolution characteristics between adjacent time positions is used as the degree of deviation in the continuous deviation judgment algorithm. If the degree of deviation in the deviation evolution sequence continues to increase at multiple adjacent time positions and the time interval between adjacent crossing events is the same, the corresponding deviation evolution sequence is judged as a continuous deviation progression pattern.

[0093] If the degree of deviation in the statistical deviation evolution sequence gradually decreases at multiple adjacent time positions, or if the time interval between adjacent crossing events shows a continuous increase, the corresponding deviation evolution sequence is determined to be a deviation convergence pattern.

[0094] When the deviation from the evolutionary sequence is determined to be a deviation from the convergence pattern, no execution scheduling trigger state is generated; when the deviation from the evolutionary sequence is determined to be a deviation from the continuous advancement pattern within the execution phase interval, an execution scheduling trigger state is generated.

[0095] It should be noted that the continuous deviation judgment algorithm refers to the sequential analysis of the deviation changes at each time position in the deviation evolution sequence. By comparing the direction and magnitude of the change of the deviation evolution characteristics at adjacent time positions, it determines whether the deviation evolution characteristics show a continuously increasing evolution trend or a gradually weakening regression trend in the time dimension, thereby distinguishing between occasional deviations and continuous deviations.

[0096] S5: Based on the execution scheduling trigger state, perform progressive resource scheduling control according to the execution phase interval, and update the execution status evidence through cooling constraint markers.

[0097] like Figure 6The figure shows a comparison of scheduling capacity changes over time for different resource scheduling strategies. In the experiment, scheduling capacity is used to characterize the magnitude of resource adjustment when load changes or abnormal states are detected. Control group 1 uses a resource result-driven resource scheduling scheme, using outcome indicators such as request response latency or resource utilization as the basis for scheduling triggers, and directly performs resource scaling operations after the indicators exceed the limits. Control group 2 uses a scheduling scheme without introducing cooling constraints. When performance indicators change, no time or execution phase interval is set for adjacent scheduling behaviors, which may trigger scheduling multiple times during the period of abnormality. The experimental group uses the progressive resource scheduling control method proposed in this invention, which is based on execution phase intervals and combined with cooling constraints. It comprehensively utilizes execution status evidence in the scheduling decision-making process to continuously limit the magnitude of resource adjustment. As can be seen from the figure, control group 1 often triggers large capacity adjustments when abnormalities occur, resulting in jumps and overshoots in the scheduling capacity curve. Although control group 2 can respond to abnormal states faster, due to the lack of cooling constraints, it is prone to multiple repeated scheduling during the abnormal phase, causing frequent fluctuations in scheduling capacity. In contrast, the experimental group mainly exhibited a small-step, continuous capacity adjustment process during the abnormal phase, and effectively avoided repeated scheduling in a short period of time under the effect of cooling constraints, thus demonstrating the technical effect of the present invention in improving the accuracy of resource scheduling.

[0098] S5.1: Read the execution stage interval corresponding to the execution scheduling trigger state, and use the stage-constrained progressive scheduling method to perform progressive resource scheduling control on the storage resource state to generate updated execution state evidence.

[0099] Specifically, the execution phase intervals recorded in the execution scheduling trigger state are identified, and the corresponding storage resource status (including available storage capacity status, storage access concurrency status, storage access latency status, and storage access queue status) is read within the time position covered by the execution phase interval.

[0100] After determining the execution phase interval corresponding to the execution scheduling trigger state, select the storage resource items that actually participate in the access from the storage resource status associated with the execution phase interval, and perform a single-step adjustment operation on the storage scheduling parameters based on the existing storage configuration values. For example, increase or decrease the concurrency limit corresponding to the storage access concurrency state, and adjust the queue depth corresponding to the storage access queue state. After the adjustment is completed, stop further modification immediately and record the adjusted storage resource status.

[0101] The execution status information at the corresponding time position within the execution phase interval is re-collected, and updated execution status evidence is generated through stabilization processing.

[0102] It should be noted that the phase-constrained incremental scheduling method refers to adjusting the storage resource status of the current execution phase in a single step within the divided execution phase interval. By limiting the resource adjustment behavior to a single execution phase interval, the phase-constrained incremental scheduling method ensures that each resource adjustment corresponds to a clear execution phase background, avoids excessive resource changes caused by continuous scheduling across phases, and provides a clear and comparable scheduling basis for subsequent feedback judgment based on execution status evidence.

[0103] S5.2: Based on the updated execution status evidence, calculate the degree of convergence of state deviation within the phase to determine whether the current scheduling behavior has generated feedback on the execution status, and obtain the deviation regression discriminant quantity within the phase.

[0104] Specifically, the updated execution state evidence extracts execution state evidence fragments within the execution phase interval corresponding to the execution scheduling trigger state. These fragments are then compared item by item with the execution state evidence prior to the resource adjustment operation. For example, thread scheduling state, blocking duration, I / O access occupancy ratio, and request-response latency information are compared. The degree of state deviation convergence within the phase is calculated using the following expression: ; in, The degree of deviation of the state from convergence within a certain stage. To determine the number of dimensions for execution status information, To perform state information dimension indexing, Before resource adjustment The range of change of each execution status information within the execution phase interval. The change range of execution status information within the execution phase interval after resource adjustment is represented by sgn, which is the sign function.

[0105] It should be noted that the expressions for the degree of convergence deviation of the state during the calculation phase are all composed of the change amplitude of the same type of execution state information, and the dimensions are dimensionless quantities.

[0106] When the degree of convergence of the state deviation within a stage shows a decreasing trend compared to before the resource adjustment, the state evidence is merged and updated to output the corresponding in-stage deviation regression discriminant, and the output is solidified as the in-stage deviation regression discriminant.

[0107] S5.3: Based on the deviation regression discriminant within the stage, the corresponding execution stage interval is marked with a cooling constraint and merged with the updated execution status evidence to generate scheduling continuation execution status evidence.

[0108] Specifically, based on the deviation regression discriminant within a stage, the corresponding execution stage interval is assigned a cooling constraint label and recorded using the execution stage interval as an index.

[0109] The cooling constraint marker is added as attribute information to the updated execution state evidence, while keeping the original time order, execution state marker trajectory and deviation evolution representation unchanged. The updated execution state evidence and the execution stage interval identifier carrying the cooling constraint marker are continuously spliced ​​on the time axis.

[0110] The execution status evidence with added cooling constraint markings is merged with the historical execution status evidence within the same execution phase interval in chronological order to form scheduling continuation execution status evidence that includes the execution status evolution process, scheduling feedback results, and phase cooling information.

[0111] It should be noted that the cooling constraint mark refers to the time constraint mark attached to the corresponding execution stage interval after completing a progressive resource scheduling control and confirming that the execution status has generated effective feedback. It is used to record the scheduling adjustments that have occurred in the execution stage within the time range, so as to avoid repeated scheduling or over-scheduling and ensure the continuity, controllability and consistency of the resource adjustment process in the time dimension and the evolution of the execution status.

[0112] This embodiment also provides a campus cloud platform elastic resource scheduling system, including: a status evidence module, which collects execution status information and stores resource status, and solidifies it after stabilization processing to generate execution status evidence; a stage identification module, which performs execution status co-evolution discrimination on the execution status evidence, locates the stage switching boundary in the program execution process, and divides the execution stage interval; a status constraint module, which performs stage-based screening of the execution status evidence according to the execution stage interval and generates execution status constraint conditions; a trigger judgment module, which, based on the execution status constraint conditions, uses a continuous deviation judgment algorithm to perform attenuation update processing on continuous out-of-bounds situations and generates an execution scheduling trigger state; and a scheduling update module, which, based on the execution scheduling trigger state, performs progressive resource scheduling control according to the execution stage interval and updates the execution status evidence through cooling constraint markers.

[0113] In summary, this invention achieves a shift in resource scheduling decisions from resource outcome-driven to execution process-driven by: constructing execution status evidence, identifying execution phases, determining continuous deviations, and scheduling with cooling constraints; it can effectively improve the operational stability and resource scheduling accuracy of industrial cloud platforms in complex business scenarios without relying on complex hardware modifications.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for elastic resource scheduling on a campus cloud platform, characterized in that: include, Collect execution status information and storage resource status, and solidify them into execution status evidence after stabilization processing; Perform co-evolution judgment on execution status evidence to locate the stage switching boundary in the program execution process and divide the execution stage interval; Based on the execution stage interval, the execution status evidence is screened in stages to generate execution status constraints. Based on the execution state constraints, a continuous deviation judgment algorithm is used to perform decay update processing on continuous out-of-bounds cases to generate the execution scheduling trigger state; Based on the execution scheduling trigger state, progressive resource scheduling control is performed according to the execution phase interval, and the execution status evidence is updated through cooling constraint markers.

2. The campus cloud platform elastic resource scheduling method as described in claim 1, characterized in that: The execution status information includes thread scheduling status, blocking duration, I / O access occupancy ratio, and request response latency information. Storage resource status includes available storage capacity status, storage access concurrency status, storage access latency status, and storage access queue status.

3. The campus cloud platform elastic resource scheduling method as described in claim 2, characterized in that: The specific steps for generating execution status evidence are as follows: The execution status information is processed by a sliding window in chronological order to remove transient data that only appears at a single sampling point. Data that maintains a consistent trend across multiple consecutive sampling periods in the execution status information processed by the sliding window is locked and stored to generate execution status evidence.

4. The campus cloud platform elastic resource scheduling method as described in claim 3, characterized in that: The specific steps for performing co-evolutionary discrimination of execution status evidence to locate stage switching boundaries during program execution are as follows: The execution status evidence is timestamped and trend-marked to generate an execution status mark trajectory. Based on the execution state marker trajectory, a sequence of execution state linkage directions is generated by judging the direction of change of execution state at adjacent time points; In the execution state linkage direction sequence, the linkage direction order comparison method is used to compare the arrangement order of linkage directions in adjacent time periods to obtain the stage switching boundary.

5. The campus cloud platform elastic resource scheduling method as described in claim 4, characterized in that: The division of execution phase intervals refers to segmenting the time range of execution status evidence according to the phase switching boundary, and merging the execution status evidence to form execution phase intervals.

6. The campus cloud platform elastic resource scheduling method as described in claim 5, characterized in that: The specific steps for performing phased screening of execution status evidence based on execution phase intervals are as follows. The execution status evidence is sliced ​​and grouped according to each execution stage interval to obtain the execution status evidence fragment set corresponding to each stage; Based on the execution state evidence fragment set, the intra-stage pattern compression algorithm is used to merge the execution state marker paths that appear repeatedly in the fragments into the shortest representative path to obtain a representative evolution path set. The path coverage priority algorithm is used to select the path subset with the most covered segments from the representative evolution path set to obtain the normal execution path coverage set.

7. The campus cloud platform elastic resource scheduling method as described in claim 6, characterized in that: The generation of execution state constraints refers to extracting the temporal sequence state transition relationships from the normal execution path coverage set, encapsulating and solidifying them, and generating execution state constraints.

8. The campus cloud platform elastic resource scheduling method as described in claim 7, characterized in that: Based on the execution state constraints, a continuous deviation judgment algorithm is used to decay and update continuous out-of-bounds situations to generate an execution scheduling trigger state. The specific steps are as follows. The execution status evidence is mapped to the execution status constraints to identify whether a cross-stage switching boundary has occurred and to record the corresponding crossing event. The time decay memory method is used to update the time decay memory of crossover events in adjacent time periods to generate a deviation evolution sequence. Based on the deviation evolution sequence, a continuous deviation judgment algorithm is used to determine the temporal distribution pattern of the crossing event and generate the execution scheduling trigger state.

9. The campus cloud platform elastic resource scheduling method as described in claim 8, characterized in that: Based on the execution scheduling trigger state, progressive resource scheduling control is performed according to the execution phase interval, and the execution status evidence is updated through cooling constraint markers. The specific steps are as follows. Read the execution stage interval corresponding to the execution scheduling trigger state, and use the stage-constrained progressive scheduling method to perform progressive resource scheduling control on the storage resource state to generate updated execution state evidence; Based on the updated execution status evidence, calculate the degree of convergence of state deviation within the phase to determine whether the current scheduling behavior has generated feedback on the execution status, and obtain the deviation regression discriminant quantity within the phase. Based on the deviation regression discriminant within the stage, the corresponding execution stage interval is marked with a cooling constraint and merged with the updated execution status evidence to generate scheduling continuation execution status evidence.

10. A campus cloud platform elastic resource scheduling system, based on the campus cloud platform elastic resource scheduling method according to any one of claims 1 to 9, characterized in that: include, The status evidence module collects execution status information and storage resource status, and solidifies them into execution status evidence after stabilization processing. The stage identification module performs collaborative evolution judgment on execution status evidence, locates stage switching boundaries in the program execution process, and divides execution stage intervals. The state constraint module performs phased filtering of execution state evidence based on the execution stage interval, and generates execution state constraint conditions. The trigger determination module, based on the execution state constraints, uses a continuous deviation determination algorithm to perform decay update processing on continuous out-of-bounds situations and generate the execution scheduling trigger state; The scheduling update module performs progressive resource scheduling control according to the execution stage interval based on the execution scheduling trigger state, and updates the execution status evidence through cooling constraint markers.

Images