Real-time job scheduling method, system and device for load balance in cloud, and storage medium
By employing fractional programming relaxation and random rounding techniques in a cloud computing environment, and collaboratively modeling job data volume and link bandwidth, a low-latency load balancing job scheduling method is designed. This solves the problems of scheduling latency and load balancing in cloud computing and achieves collaborative optimization of low latency and load balancing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-04-03
AI Technical Summary
Existing cloud computing environments struggle to simultaneously meet the dual requirements of "low latency" and "load balancing" in dynamic cloud environments. Traditional scheduling solutions suffer from excessively long scheduling delays, are unable to adapt to the real-time requirements of short job lifecycles in containerized environments, and fail to effectively control the impact of job data volume differences on scheduling delays.
By employing fractional programming relaxation and random rounding techniques, and through a latency control mechanism that coordinates job data volume and link bandwidth modeling, combined with multinomial complexity fractional programming solution and dynamic load-aware triggering strategy, a low-latency load balancing job scheduling method (RBJS) for cloud environments is designed to achieve coordinated optimization of low latency and load balancing.
It achieves collaborative optimization of low latency and load balancing in cloud cluster job scheduling, satisfies the multi-objective collaborative optimization of low scheduling latency, low decision latency and load balancing, reduces decision latency and improves resource utilization.
Smart Images
Figure CN121785770A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cloud computing resource scheduling technology, specifically to a real-time job scheduling method, system, device, and storage medium for load balancing in the cloud. Background Technology
[0002] In cloud computing environments, job scheduling optimizes resource utilization and improves the service of cloud computing devices. It requires the rational allocation of storage and network resources to ensure that tasks can be executed efficiently and stably. With the widespread application of containerization and microservice technologies, cloud workloads exhibit characteristics of transient and high-frequency scheduling—frequent job creation, termination, and migration can easily lead to server load imbalances within the cluster, thereby reducing overall system performance and resource utilization.
[0003] Existing scheduling schemes in cloud computing environments struggle to simultaneously meet the dual requirements of "low latency" and "load balancing" in dynamic cloud environments, and exhibit significant deficiencies in matching "scheduled quantity control" with "latency optimization." Limitations of heuristic and metaheuristic methods used in cloud computing environments: For example, while traditional heuristic algorithms (such as genetic algorithms and simulated annealing) and metaheuristic algorithms (such as ant colony optimization, tabu search, and sea lion optimization) can alleviate load imbalance in cloud computing environments to some extent, they have fundamental shortcomings. Existing cloud computing environments lack global scheduling planning. In pursuit of the global optimum, they ignore the current system scheduling state, easily leading to numerous unnecessary job adjustments. Furthermore, they generally do not focus on controlling the "number of scheduled jobs" or consider the impact of differences in job data volume on scheduling latency, resulting in a significant increase in latency during large-scale job scheduling and failing to meet the real-time requirements of short job lifecycles in containerized environments.
[0004] The decision latency problem of Integer Linear Programming (ILP) schemes: ILP-based scheduling schemes, by constructing models that integrate server load, job resource requirements, and scheduling latency constraints, can reduce the number of scheduled jobs and thus lower latency, but they suffer from high computational complexity. In large-scale scenarios (e.g., 400 jobs on 5 servers), the latency of generating scheduling decisions using ILP solvers (such as Gurobi) can reach hundreds of seconds. Excessive decision latency can cause the scheduling scheme to become disconnected from the actual system state. When the scheme is executed, the server load has already changed due to job completion and new job arrivals, rendering the original scheme ineffective and failing to achieve the expected load balancing goal.
[0005] The limitations of simply controlling the number of scheduled jobs: Even though some solutions attempt to reduce latency by limiting the number of scheduled jobs, they still cannot solve the core problem—scheduling latency is not only related to the number of jobs, but also closely related to the amount of job data. The cross-server transmission of large jobs requires more bandwidth and time. Simply controlling the number of scheduled jobs may cause latency to exceed the limit due to the inclusion of large jobs, and still cannot meet the real-time scheduling requirements.
[0006] To address the aforementioned issues and simultaneously meet the requirements of low scheduling latency, low decision latency, and load balancing, this invention, overcoming the limitations of traditional solutions, designs a low-latency load balancing job scheduling method (RBJS) for cloud environments based on fractional programming relaxation and random rounding techniques. Through a latency control mechanism that collaboratively models job data volume and link bandwidth, a multinomial-complexity fractional programming solution to reduce decision latency, and a dynamic load-aware triggering strategy, the invention achieves the synergistic optimization goal of low latency and load balancing in cloud cluster job scheduling. Summary of the Invention
[0007] This invention overcomes the shortcomings of existing technologies and provides a real-time job scheduling method, system, device, and storage medium for load balancing in the cloud. Based on fractional programming relaxation and random rounding techniques, it adopts a low-latency load balancing job scheduling method (RBJS) for cloud environments. Through a latency control mechanism that coordinates job data volume and link bandwidth modeling, a multinomial complexity fractional programming solution to reduce decision latency, and a dynamic load-aware triggering strategy, it achieves the synergistic optimization goal of low latency and load balancing in cloud cluster job scheduling, meets the requirements of low scheduling latency, low decision latency, and load balancing, and realizes multi-objective synergistic optimization of scheduling latency and load balancing.
[0008] To achieve the above objectives, the technical solution adopted by this invention is: a real-time job scheduling method for load balancing in the cloud, comprising: Obtain real-time operational information, real-time load status, and task allocation information of the server nodes in the corresponding cloud cluster. Based on the real-time operation information, real-time load status, and task allocation information, scheduling is performed through a pre-established load-balanced real-time job scheduling model. The scheduling method of the real-time job scheduling model includes: Based on real-time operational information, a method of real-time load monitoring and dynamic scheduling is used to determine whether to trigger load balancing job scheduling. Load balancing job scheduling includes the following steps: Based on real-time operation information, real-time load status, and task allocation information, the optimal fractional solution matrix (i.e. allocation probability matrix) of the current task set on each server in the cloud cluster is solved using linear programming. Based on the optimal fractional solution matrix, a feasible integer allocation scheme is generated through a random rounding strategy. The migration delay during the scheduling process is controlled based on the constraints of task data volume and link bandwidth, thereby achieving load-balanced job scheduling. After the load balancing job scheduling is completed, the real-time running information, real-time load status, and task allocation information of the server nodes are updated, and the next round of real-time job scheduling model scheduling method is entered.
[0009] In a preferred embodiment of the present invention, the real-time load monitoring and dynamic scheduling triggering method includes determining whether a scheduling process needs to be triggered by analyzing the load variation coefficient (CV) of the cloud cluster's current operating state; the algorithm for the load variation coefficient (CV) includes: ;in, For average load, The standard deviation of the load; When CV is below the adaptive threshold When the current cluster is in a balanced state, no scheduling needs to be triggered; when the CV exceeds the adaptive threshold... At this time, the load difference between servers is significant, triggering load balancing job scheduling.
[0010] In a preferred embodiment of the present invention, an adaptive threshold Initially, a threshold is preset. Adaptive threshold The threshold is dynamically adjusted based on the number of servers, differences in node performance, and job arrival rate to obtain an adaptive threshold. ; The threshold is raised when the number of nodes increases or resource heterogeneity increases; the threshold is lowered when the cluster size is small or homogeneity is high.
[0011] In a preferred embodiment of the present invention, a cloud cluster is defined as: a set V containing n server nodes. ; Let the cluster topology be represented by graph G=(V,E), where E is the directed communication link between servers used for job scheduling; The hardware resource parameters for each server node include: CPU capacity of... Memory capacity is ; Set of jobs to be scheduled ; The resource requirements for each job j include: CPU requirements are... Memory requirements are The identifier for the initial placement of task j on server g is... A value of 1 indicates that it was initially placed on server g, and a value of 0 indicates that it was not initially placed on server g. It is the total bandwidth of the link. It represents the number of jobs passing through the link.
[0012] In a preferred embodiment of the present invention, the linear programming method includes a delayed satisfaction task scheduling solution method, comprising: The actual scheduling delay is modeled based on the job data volume and link bandwidth. The actual scheduling delay algorithm includes: .
[0013] In a preferred embodiment of the present invention, the Delay Constraint Job Scheduling (DSJS) problem constraints include: Not exceeding T0; The scheduling delay for a single job is T0, where T0 is the maximum allowed delay, and t is the maximum allowed delay. max The maximum scheduling delay across all jobs; CPU load constraints are: This means that the total CPU requirement of all jobs on a single server does not exceed the server's CPU capacity and maximum load factor. The product; This means that the total memory requirement of all jobs on a single server does not exceed the server's memory capacity. This means that each job can only select one target server.
[0014] In a preferred embodiment of the present invention, a weighting function is used to balance the two objectives of "load balancing" and "scheduling delay," the formula of which includes: ; The weights adjust the trade-offs: smaller ones Prioritize load balancing, larger loads It helps reduce scheduling delays; It belongs to the range 0 to 1.
[0015] In a preferred embodiment of the present invention, the random rounding strategy includes: for each job j, according to The probability of randomly selecting a unique target server v satisfies the following condition. .
[0016] In a preferred embodiment of the present invention, a real-time job scheduling system for load balancing in the cloud includes: an interconnected system initialization and status awareness module, a delay-satisfaction task scheduling solution module, a random rounding scheduling module, and a real-time load monitoring and dynamic scheduling triggering module. The system initialization and status awareness module obtains real-time operating information, real-time load status, and task allocation information of the server nodes. The real-time load monitoring and dynamic scheduling triggering module determines whether to trigger load balancing job scheduling based on the real-time operation information, real-time load status, and task allocation information. Based on real-time operation information, real-time load status, and task allocation information, the linear programming method is used by the delay-satisfaction task scheduling solution module to solve for the optimal fractional solution matrix (i.e. allocation probability matrix) of the current task set on each server in the cloud cluster. Based on the optimal fractional solution matrix, a feasible integer allocation scheme is generated through the random rounding strategy of the random rounding scheduling module to achieve load-balanced job scheduling. After the load balancing job scheduling is completed, the real-time running information, real-time load status, and task allocation information of the server nodes are updated through the system initialization and status awareness module.
[0017] In a preferred embodiment of the present invention, a real-time job scheduling device for load balancing in the cloud includes a memory for storing computer programs / instructions. A processor is used to execute the computer program / instructions to implement a real-time job scheduling method for load balancing in the cloud.
[0018] In a preferred embodiment of the present invention, a storage medium for real-time job scheduling for load balancing in the cloud stores a computer program / instruction, which, when executed by a processor, implements the steps of a real-time job scheduling method for load balancing in the cloud.
[0019] This invention addresses the deficiencies in the technical background, and the beneficial technical effects of this invention are: This invention provides a real-time job scheduling method, system, device, and storage medium for load balancing in the cloud. Based on fractional programming relaxation and random rounding techniques, it adopts a low-latency load balancing job scheduling method (RBJS) for cloud environments. Through a latency control mechanism that coordinates job data volume and link bandwidth modeling, a fractional programming solution with polynomial complexity to reduce decision latency, and a dynamic load-aware triggering strategy, it achieves the synergistic optimization goal of low latency and load balancing in cloud cluster job scheduling, meets the requirements of low scheduling latency, low decision latency, and load balancing, and realizes multi-objective synergistic optimization of scheduling latency and load balancing. Attached Figure Description
[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0021] Figure 1 This is a flowchart of the overall process of real-time job scheduling for cloud load balancing, according to a preferred embodiment of the present invention. Figure 2 This is a flowchart of the system initialization and state awareness module according to a preferred embodiment of the present invention; Figure 3 This is a flowchart of the delay satisfaction task scheduling solution module of a preferred embodiment of the present invention; Figure 4 This is a flowchart of the random rounding scheduling module of a preferred embodiment of the present invention; Figure 5 This is a flowchart of the real-time load monitoring and dynamic scheduling triggering module of a preferred embodiment of the present invention; Figure 6 It is a comparison of server CPU load rate and number of tasks (T0=1 second) when different scheduling schemes are used for scheduling operations; Figure 7 This is a comparison of server CPU load ratio and scheduling latency constraints (3000 tasks) when using different scheduling schemes for scheduling operations; Figure 8 This is a graph showing the relationship between execution time and the number of tasks. Figure 9 It is the server CPU load rate and the number of tasks (T0=1 second) (random rounding and integer programming); Figure 10 This is a stable test pattern of the present invention; Figure 11 This is a trade-off diagram of the present invention. Detailed Implementation
[0022] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention and therefore only show the components relevant to the present invention.
[0023] It should be noted that if directional indicators (such as up, down, bottom, top, etc.) are involved in the embodiments of the present invention, these directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of that feature. Unless otherwise explicitly specified and limited, the terms "set," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.
[0024] Example 1, as Figures 1-3As shown, a real-time job scheduling method for load balancing in the cloud includes: acquiring real-time operating information, real-time load status, and task allocation information of server nodes in the corresponding cloud cluster. The real-time operating information includes CPU utilization, memory usage, and network bandwidth usage. Based on the real-time operating information, real-time load status, and task allocation information, scheduling is performed using a pre-established load-balancing real-time job scheduling model.
[0025] The scheduling method of the real-time job scheduling model includes: determining whether to trigger load balancing job scheduling based on real-time operation information through real-time load monitoring and dynamic scheduling triggering methods; load balancing job scheduling includes the following steps: based on real-time operation information, real-time load status, and task allocation information, solving the optimal fractional solution matrix (i.e., allocation probability matrix) of the current task set on each server in the cloud cluster using linear programming methods; generating feasible integer allocation schemes based on the optimal fractional solution matrix through random rounding strategies, and controlling the migration delay during the scheduling process based on the constraints of task data volume and link bandwidth to achieve load balancing job scheduling; after the load balancing job scheduling is completed, updating the real-time operation information, real-time load status, and task allocation information of the server nodes, and entering the next round of the real-time job scheduling model's scheduling method.
[0026] Furthermore, a cloud cluster is defined as: a set V containing n server nodes. The cluster topology is represented by graph G=(V,E), where E is the directed communication link between servers used for job scheduling (supporting job scheduling between any two servers); the hardware resource parameters of each server node include: CPU capacity... Memory capacity is Set of jobs to be scheduled The resource requirements for each job j include: CPU requirements are... Memory requirements are The identifier for the initial placement of task j on server g is... A value of 1 indicates that it was initially placed on server g, and a value of 0 indicates that it was not initially placed on server g. It is the total bandwidth of the link. It represents the number of jobs passing through the link.
[0027] Specifically, the real-time load monitoring and dynamic scheduling triggering method includes determining whether a scheduling process needs to be triggered by analyzing the load variation coefficient (CV) of the cloud cluster at the current moment; the algorithm for the load variation coefficient (CV) includes: ;in, For average load, The load standard deviation is used when the CV is below the adaptive threshold. When the current cluster is in a balanced state, no scheduling needs to be triggered; when the CV exceeds the adaptive threshold... When the load difference between servers is significant, load balancing job scheduling is triggered. Furthermore, the load calculation method for each server node includes: server node load ratio vector: the ratio of the total CPU demand of all running jobs on the server node to the total CPU capacity of the server node; obtaining the average load and load standard deviation of all server nodes based on the server node load ratio vector; and obtaining the load variation coefficient (CV) based on the average load and load standard deviation.
[0028] Specifically, adaptive threshold Initially, a threshold is preset. threshold The threshold is dynamically adjusted based on the number of servers, differences in node performance, and job arrival rate to obtain an adaptive threshold. When the number of nodes increases or resource heterogeneity increases, the threshold is raised; when the cluster size is small or homogeneity is high, the threshold is lowered. Furthermore, when the number of nodes increases or resource heterogeneity increases, the system automatically raises the threshold to avoid normal fluctuations being misjudged as imbalances; when the system size is small or homogeneity is high, the system automatically lowers the threshold to improve scheduling sensitivity, thereby ensuring the stability and rationality of the scheduling triggering logic.
[0029] Specifically, linear programming methods include delay-based task scheduling solutions, such as modeling actual scheduling delays based on job data volume and link bandwidth. The actual scheduling delay algorithm includes: .
[0030] Furthermore, the Delay-Constrained Job Scheduling (DSJS) problem constraints include: Not exceeding T0; The scheduling delay for a single job is T0, where T0 is the maximum allowed delay, and t is the maximum allowed delay. max The maximum scheduling delay across all jobs; The memory resource utilization of the j-th device / task; Benchmark values for resource demand; Total system resource capacity.
[0031] Furthermore, the CPU load constraint is as follows: This means that the total CPU requirement of all jobs on a single server does not exceed the server's CPU capacity and maximum load factor. The product; This means that the total memory requirement of all jobs on a single server does not exceed the server's memory capacity. This means that each job can only select one target server.
[0032] Furthermore, a weighted function is used to balance the two objectives of "load balancing" and "scheduling latency," and the formula includes: ; The weights adjust the trade-offs: smaller ones Prioritize load balancing, larger loads It helps reduce scheduling delays; It belongs to the range 0 to 1. Specifically, since integer linear programming is an NP-hard problem, relaxation... Binary constraints (allowing) ∈[0,1]), transforming it into fractional programming (LP1). LP1 is a multinomial-scale model with a time complexity of only Fractional solutions can be obtained quickly using a standard solver. (Operation Assigned to The probability of (the probability of) and the lower bound of the optimal load factor This significantly reduces decision-making delays.
[0033] Furthermore, the random rounding strategy includes: for each job j, based on... The probability of randomly selecting a unique target server v satisfies the following condition. To transform fractional solutions into feasible integer solutions while avoiding link congestion, a random rounding strategy is employed. Furthermore, approximation factor analysis is performed: to schedule jobs to suitable selected servers, the above analysis shows that for the DSJS problem, scheduling latency is almost negligible. If the factor is violated, the server CPU capacity constraint will not be... The factor is violated.
[0034] Example 2, based on Example 1, presents a real-time job scheduling method for load balancing in the cloud. The overall flowchart is as follows: Figure 1 As shown.
[0035] Among them, the system initialization and state awareness module is as follows: Figure 2 As shown, the specific implementation includes the following steps: S11: Start the cloud resource scheduling system and collect basic information of all server nodes, including total CPU capacity, memory capacity and network bandwidth.
[0036] S12: Read the currently running job set and record the CPU requirements, memory requirements, and initial allocation location of each job.
[0037] S13: Generate a load state diagram G(V,E) based on the collected system status information to provide input for subsequent scheduling optimization.
[0038] Among them, the delayed satisfaction task scheduling solution module is as follows: Figure 3 As shown, the specific implementation includes the following steps: S21: Receive the current system load status and job set, and initialize the delay-constrained task scheduling model.
[0039] S22: Construct a joint optimization objective function, using the load balancing factor and scheduling delay t max For optimization purposes.
[0040] S23: Generate CPU, memory, and latency constraints to form a linear programming model.
[0041] S24: Calculate the optimal fractional solution matrix using the linear programming solver module. .
[0042] S25: If the constraints are not met, adjust the weight parameters and solve again; if they are met, output the result.
[0043] S26: Pass the result to the random rounding module to generate a feasible allocation scheme.
[0044] Among them, the random rounding scheduling module is as follows: Figure 4 As shown, the specific implementation includes the following steps: S31: Receive the fractionalized result of the linear programming problem.
[0045] S32: Randomly sample based on the distribution probability of each task on each server to determine the target node.
[0046] S33: Calculate the overall CPU load and the expected scheduling latency.
[0047] S34: If there is a task migration delay exceeding the limit, perform a partial correction and adjust the task allocation scheme.
[0048] S35: Output the final task allocation result and execute job scheduling.
[0049] Among them, the real-time load monitoring and dynamic scheduling triggering module, such as Figure 5 As shown, the specific implementation includes the following steps: S41: Periodically collect real-time CPU load and status information of all servers to form a load vector.
[0050] S42: Calculate the average load and standard deviation, and obtain the load variation coefficient CV.
[0051] S43: Determine if CV exceeds the threshold CV. th If the limit is not exceeded, the status quo is maintained; if the limit is exceeded, scheduling is triggered.
[0052] S44: Execute the RBJS algorithm to reallocate jobs.
[0053] S45: Update the server status and return to the monitoring phase, forming a cyclical execution mechanism.
[0054] Example 3: A real-time job scheduling system for load balancing in the cloud, comprising: an interconnected system initialization and status awareness module, a delay-satisfaction task scheduling solution module, a random rounding scheduling module, and a real-time load monitoring and dynamic scheduling triggering module; The system initialization and status awareness module obtains real-time operating information, real-time load status, and task allocation information of the server nodes. The real-time load monitoring and dynamic scheduling triggering module determines whether to trigger load balancing job scheduling based on the real-time operation information, real-time load status, and task allocation information. Based on real-time operation information, real-time load status, and task allocation information, the linear programming method is used by the delay-satisfaction task scheduling solution module to solve the allocation probability matrix of the current task set on each server in the cloud cluster. Based on the allocation probability matrix, a feasible integer allocation scheme is generated through the random rounding strategy of the random rounding scheduling module to achieve load balancing job scheduling. After the load balancing job scheduling is completed, the real-time running information, real-time load status, and task allocation information of the server nodes are updated through the system initialization and status awareness module.
[0055] Example 4: A real-time job scheduling device for load balancing in the cloud, including a memory for storing computer programs / instructions; A processor is used to execute the computer program / instructions to implement the steps of a real-time job scheduling method for load balancing in the cloud, as described in Embodiment 1 or Embodiment 2.
[0056] Example 5: A storage medium for real-time job scheduling for load balancing in the cloud, which stores a computer program / instruction. When the computer program / instruction is executed by a processor, it implements the steps of a real-time job scheduling method for load balancing in the cloud according to Example 1 or Example 2.
[0057] Working principle: In cloud computing scenarios, jobs typically run in containerized environments, have short lifecycles, are frequently scheduled, and have varying hardware configurations across different nodes. Traditional scheduling strategies based on fixed thresholds often fail to adapt to this dynamic nature, easily leading to problems such as "over-scheduling" or "delayed response." This invention provides a real-time job scheduling method, system, device, and storage medium for load balancing in the cloud, using the CPU load coefficient of variation (CV) as the core metric for cluster load balancing. Figure 6The data shown compares server CPU load rate with the number of tasks (T0=1 second) when different scheduling schemes are used for scheduling operations. Figure 7 The data shown compares server CPU load ratios with scheduling latency constraints (3000 tasks) when different scheduling schemes are used. Unlike single-node load thresholds, the CV (Comparison Value) objectively reflects the global load distribution of the system by comparing the dispersion of load across servers with the overall mean. A larger CV value indicates a more significant difference in load between servers; conversely, a smaller CV value indicates a more balanced cluster operation.
[0058] like Figure 8 A graph showing the relationship between execution time and the number of tasks. Figure 9 Server CPU load rate and number of tasks (T0=1 second) (random rounding and integer programming) Figure 10 Stability test diagram; Figure 11 As shown in the trade-off diagram, this invention is based on fractional programming relaxation and random rounding techniques, and adopts a low-latency load balancing job scheduling method (RBJS) for cloud environments. Through a latency control mechanism that coordinates job data volume and link bandwidth modeling, a fractional programming solution with polynomial complexity to reduce decision latency, and a dynamic load-aware triggering strategy, it achieves the synergistic optimization goal of low latency and load balancing in cloud cluster job scheduling, meets the requirements of low scheduling latency, low decision latency, and load balancing, and realizes multi-objective synergistic optimization of scheduling latency and load balancing.
[0059] The above specific embodiments are specific support for the concept proposed in this invention, and should not be used to limit the scope of protection of this invention. Any equivalent changes or modifications made on the basis of this technical solution in accordance with the technical concept proposed in this invention shall still fall within the scope of protection of this invention.
Claims
1. A real-time job scheduling method for load balancing in the cloud, characterized in that, include: Obtain real-time operational information, real-time load status, and task allocation information of the server nodes in the corresponding cloud cluster. Based on the real-time operation information, real-time load status, and task allocation information, scheduling is performed through a pre-established load-balanced real-time job scheduling model. The scheduling method of the real-time job scheduling model includes: Based on real-time operational information, a method of real-time load monitoring and dynamic scheduling is used to determine whether to trigger load balancing job scheduling. Load balancing job scheduling includes the following steps: Based on real-time operation information, real-time load status, and task allocation information, the optimal fractional solution matrix of the current task set on each server in the cloud cluster is solved using linear programming. Based on the optimal fractional solution matrix, a feasible integer allocation scheme is generated through a random rounding strategy. The migration delay during the scheduling process is controlled based on the constraints of task data volume and link bandwidth, thereby achieving load-balanced job scheduling. After the load balancing job scheduling is completed, the real-time running information, real-time load status, and task allocation information of the server nodes are updated, and the next round of real-time job scheduling model scheduling method is entered.
2. The real-time job scheduling method for load balancing in the cloud according to claim 1, characterized in that: The real-time load monitoring and dynamic scheduling triggering method includes determining whether a scheduling process needs to be triggered by analyzing the load variation coefficient (CV) of the cloud cluster's current operating status. The algorithms for the coefficient of variation (CV) of the load include: ;in, For average load, The standard deviation of the load; When CV is below the adaptive threshold When the current cluster is in a balanced state, no scheduling needs to be triggered; when the CV exceeds the adaptive threshold... At this time, the load difference between servers is significant, triggering load balancing job scheduling.
3. The real-time job scheduling method for load balancing in the cloud according to claim 2, characterized in that: Adaptive threshold Initially, a threshold is preset. Adaptive threshold The threshold is dynamically adjusted based on the number of servers, differences in node performance, and job arrival rate to obtain an adaptive threshold. ; The threshold is raised when the number of nodes increases or resource heterogeneity increases; the threshold is lowered when the cluster size is small or homogeneity is high.
4. A real-time job scheduling method for load balancing in the cloud according to claim 3, characterized in that: The cloud cluster is defined as: a set V containing n server nodes. ; Let the cluster topology be represented by graph G=(V,E), where E is the directed communication link between servers used for job scheduling; The hardware resource parameters for each server node include: CPU capacity of... Memory capacity is ; Set of jobs to be scheduled ; The resource requirements for each job j include: CPU requirements are... Memory requirements are The identifier for the initial placement of task j on server g is... A value of 1 indicates that it was initially placed on server g, and a value of 0 indicates that it was not initially placed on server g. It is the total bandwidth of the link. It represents the number of jobs passing through the link.
5. A real-time job scheduling method for load balancing in the cloud according to claim 4, characterized in that: The linear programming method includes a delayed satisfaction task scheduling solution method, comprising: The actual scheduling delay is modeled based on the job data volume and link bandwidth. The actual scheduling delay algorithm includes: .
6. A real-time job scheduling method for load balancing in the cloud according to claim 5, characterized in that: Delay-constrained job scheduling problem constraints include: Not exceeding T0; The scheduling delay for a single job is T0, where T0 is the maximum allowed delay, and t is the maximum allowed delay. max The maximum scheduling delay across all jobs; CPU load constraints are: This means that the total CPU requirement of all jobs on a single server does not exceed the server's CPU capacity and maximum load factor. The product; This means that the total memory requirement of all jobs on a single server does not exceed the server's memory capacity. This means that each job can only select one target server.
7. A real-time job scheduling method for load balancing in the cloud according to claim 6, characterized in that: The weighted function balances the two objectives of "load balancing" and "scheduling latency," and the formula includes: ; The weights adjust the trade-offs: smaller ones Prioritize load balancing, larger loads It helps reduce scheduling delays; Belongs to 0 to 1; The random rounding strategy includes: for each job j, based on The probability of randomly selecting a unique target server v satisfies the following condition. .
8. A real-time job scheduling system for load balancing in the cloud, characterized in that, include: The interconnected system initialization and status awareness module, the delay satisfaction task scheduling solution module, the random rounding scheduling module, and the real-time load monitoring and dynamic scheduling triggering module; The system initialization and status awareness module obtains real-time operating information, real-time load status, and task allocation information of the server nodes. The real-time load monitoring and dynamic scheduling triggering module determines whether to trigger load balancing job scheduling based on the real-time operation information, real-time load status, and task allocation information. Based on real-time operation information, real-time load status, and task allocation information, the linear programming method is used by the delay-satisfaction task scheduling solution module to solve for the optimal fractional solution matrix of the current task set on each server in the cloud cluster. Based on the optimal fractional solution matrix, a feasible integer allocation scheme is generated through the random rounding strategy of the random rounding scheduling module to achieve load-balanced job scheduling. After the load balancing job scheduling is completed, the real-time running information, real-time load status, and task allocation information of the server nodes are updated through the system initialization and status awareness module.
9. A real-time job scheduling device for load balancing in the cloud, characterized in that: Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the real-time job scheduling method for load balancing in the cloud as described in any one of claims 1-7.
10. A storage medium for real-time job scheduling for load balancing in the cloud, wherein computer programs / instructions are stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of a real-time job scheduling method for load balancing in the cloud as described in any one of claims 1-7.