Deep learning-based hyper-converged system resource optimization method and system, and medium

Abstract

The invention discloses a deep learning-based hyper-fusion system resource optimization method and system and a medium, mainly relates to the technical field of hyper-fusion systems, and is used for solving the problems that an existing scheme is difficult to adapt to a real-time operation scene of a hyper-fusion system and cannot capture global relevance of a system operation state. Comprising the following steps: acquiring operation data and system operation performance of each resource node corresponding to the hyper-converged system in real time; screening out operation tasks with distributability from all the operation data, and matching the screened operation tasks with resource nodes to obtain all permutation combinations; inputting the resource node operation data under each combination as input data into a deep learning model, obtaining predicted system operation performance, and obtaining a permutation combination corresponding to the highest predicted system operation performance; and determining a final resource node of the distributable operation task based on the operation node of the distributable operation task under the permutation combination corresponding to the highest prediction system operation performance.

Description

Technical Field

[0001] This application relates to the technical field of hyper-converged systems, and particularly to a method, system, and medium for optimizing resources in a hyper-converged system based on deep learning. Background Art

[0002] A hyper-converged system is an IT infrastructure architecture that integrates resources such as computing, storage, networking, and virtualization into a unified platform. Its core feature is to achieve seamless integration and centralized management of multiple modules through software-defined technology, forming a resource pool that can be elastically expanded.

[0003] In the field of hyper-converged systems, dynamic task allocation and performance optimization of resource nodes are the core challenges in improving system efficiency. Current technologies generally rely on manual experience or static rules for resource scheduling, making it difficult to respond in real time to complex load changes and heterogeneous task requirements. For example, traditional methods usually trigger task migration based on preset thresholds (such as CPU utilization), but such rules lack in-depth exploration of the correlation between task types, running loads, and allocatability, resulting in problems such as uneven resource utilization, coexistence of node overload and idleness.

[0004] In existing solutions, some studies have attempted to optimize resource allocation through machine learning algorithms, but their model training mostly relies on offline datasets and is difficult to adapt to the real-time operation scenario of hyper-converged systems. At the same time, these methods do not fully integrate multiple resource nodes (virtual machines, distributed computing devices, etc.), resulting in the model being unable to capture the global correlation of the system operation state. For example, the mixed deployment of storage-intensive tasks and computing-intensive tasks may cause I / O contention, but existing technologies have not established a performance prediction mechanism for such scenarios and it is difficult to achieve cross-node collaborative optimization. Summary of the Invention

[0005] In view of the above deficiencies of the prior art, this application provides a method, system, and medium for optimizing resources in a hyper-converged system based on deep learning to solve the problems that existing solutions are difficult to adapt to the real-time operation scenario of hyper-converged systems and cannot capture the global correlation of the system operation state.

[0006] In a first aspect, this application provides a method for optimizing resources in a hyper-converged system based on deep learning, the method including: Obtain the operation data of the hyper-converged system within a preset historical time period; wherein, the operation data at least includes: task data of each resource node, system operation performance, and the task data at least includes: task type of the task running on the current resource node, running load of the running task, and allocatability of the running task; Use the operation data as the training data of the deep learning model, and obtain the correlation relationship between the internal operation data by using the trained deep learning model; Obtain the operation data of each resource node corresponding to the hyper-converged system and the system operation performance at intervals of a preset update time period; screen out the running tasks with distributability as distributable from all the operation data, match the screened running tasks with the resource nodes to obtain all permutations and combinations; generate the task data of the resource nodes under each combination; use the generated task data as input data to input into a deep learning model to obtain the predicted system operation performance, and obtain the permutation and combination corresponding to the highest predicted system operation performance; Determine that the allocation node of the distributable running task under the permutation and combination corresponding to the highest predicted system operation performance is the resource node of the distributable running task within the preset update time period.

[0007] In an implementation manner of the present application, obtaining the operation data of the hyper-converged system within a preset historical time period specifically includes: Deploy lightweight monitoring agents on each resource node to collect the task type of the running tasks and the running load of the running tasks on the current resource node; Obtain the calling program and calling data involved in the running task. When the calling program exists on all resource nodes and the calling data is not preset encrypted data, determine that the distributability of the current running task is distributable; otherwise, determine that the distributability of the current running task is non-distributable.

[0008] In an implementation manner of the present application, before using the generated task data as input data to input into a deep learning model to obtain the predicted system operation performance, the method further includes: Install the trained deep learning model on the resource node; Determine the splitting ratio of the permutation and combination according to the ratio between the running memories of the resource nodes, and further obtain the permutations and combinations to be measured corresponding to each resource node.

[0009] In an implementation manner of the present application, using the generated task data as input data to input into a deep learning model to obtain the predicted system operation performance specifically includes: The resource node obtains the permutations and combinations to be measured sent down, and sequentially uses a group of combinations in the permutations and combinations to be measured as the input of the trained deep learning model on the current node to obtain the output predicted system operation performance; Upload the predicted system operation performance to the hyper-converged system so that the hyper-converged system can obtain the predicted system operation performance of all combinations.

[0010] In an implementation manner of the present application, determining that the allocation node of the distributable running task under the permutation and combination corresponding to the highest predicted system operation performance is the resource node of the distributable running task within the preset update time period specifically includes: When there are distributable running tasks to be allocated outward by a resource node, the calling program name and calling data will be sent to the running node of the distributable running task under the permutation and combination corresponding to the highest predicted system running performance.

[0011] In a second aspect, the present application provides a resource optimization system for a hyper-converged system based on deep learning. The system includes: An acquisition module, configured to acquire the running data of the hyper-converged system within a preset historical time period; wherein, the running data at least includes: task data of each resource node and system running performance, and the task data at least includes: the task type of the running task of the current resource node, the running load of the running task, and the distributability of the running task; An obtaining module, configured to use the running data as the training data of a deep learning model, and obtain the correlation relationship among the running data by using the trained deep learning model; A prediction module, configured to acquire the running data of each resource node corresponding to the hyper-converged system and the system running performance at intervals of a preset update time period; screen out the running tasks with distributability as distributable from all the running data, match the screened running tasks with the resource nodes to obtain all permutations and combinations; generate the task data of the resource nodes under each combination; use the generated task data as input data to input into the deep learning model to obtain the predicted system running performance, and obtain the permutation and combination corresponding to the highest predicted system running performance; An allocation module, configured to determine that the allocation node of the distributable running task under the permutation and combination corresponding to the highest predicted system running performance is the resource node of the distributable running task within the preset update time period.

[0012] In an implementation manner of the present application, the acquisition module includes an acquisition unit, configured to deploy lightweight monitoring agents on each resource node to collect the task type of the running task of the current resource node and the running load of the running task; acquire the calling program and calling data involved in the running task, and when the calling program exists on all resource nodes and the calling data is not preset encrypted data, determine that the distributability of the current running task is distributable; otherwise, determine that the distributability of the current running task is non-distributable.

[0013] In an implementation manner of the present application, the prediction module includes an allocation unit, configured to install the trained deep learning model on the resource node; determine the segmentation ratio of the permutation and combination according to the ratio between the running memories of the resource nodes, and further obtain the permutations and combinations to be tested corresponding to each resource node.

[0014] In one implementation of the present application, the prediction module includes a prediction unit, which is used to obtain the to-be-tested permutations and combinations sent down through the resource nodes, and sequentially use a set of combinations in the to-be-tested permutations and combinations as the input of the deep learning model trained on the current node to obtain the predicted system operation performance of the output; and upload the predicted system operation performance to the hyper-converged system, so that the hyper-converged system can obtain the predicted system operation performance of all combinations.

[0015] In a third aspect, the present application provides a non-volatile computer storage medium, on which computer instructions are stored, and when the computer instructions are executed, a resource optimization method for a hyper-converged system based on deep learning as described in any one of the above is implemented.

[0016] Those skilled in the art can understand that the present application has at least the following beneficial effects: 1. Improved real-time dynamic adaptation ability: By obtaining the operation data (task type, load, allocability) of each resource node and system performance in real time, and combining with the deep learning model to predict the system performance under different task permutations and combinations, it can dynamically respond to the load changes and resource requirements of the hyper-converged system. Compared with the traditional scheduling scheme based on static rules or offline data training, it breaks through the real-time bottleneck and solves the problem of lagging resource allocation. For example, in a high-concurrency scenario, it can quickly identify overloaded nodes and trigger task migration.

[0017] 2. Achieved global correlation relationship modeling: Using the deep learning model to mine the non-linear correlation relationships among parameters such as task type, load distribution, and allocability from historical operation data, such as the implicit dependency relationship between storage-intensive tasks and network bandwidth occupancy. This process overcomes the optimization blind spots caused by traditional methods that only rely on local metrics (such as CPU utilization rate), realizes the global perception of the system operation state, and thus avoids the overall performance degradation caused by local optimization.

[0018] 3. Improved quantitative evaluation ability: By generating all permutations and combinations of allocable tasks and inputting them into the model to predict the system performance, this method quantifies the impact of different resource allocation strategies on the system. For example, when deploying compute-intensive and I / O-intensive tasks in a hybrid manner, the model can predict the increased latency caused by resource competition among nodes and select the permutation and combination with the least conflict. This mechanism solves the problem that the prior art lacks the quantitative evaluation ability and makes the optimization decision shift from experience-driven to data-driven. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the present invention, the drawings required to be used in the description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 It is a flowchart of a method for optimizing resources of a hyper-converged system based on deep learning provided by an embodiment of the present application.

[0021] Figure 2 It is a schematic diagram of the internal structure of a system for optimizing resources of a hyper-converged system based on deep learning provided by an embodiment of the present application. Detailed implementation manners

[0022] Those skilled in the art should understand that the embodiments described below are only the preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through these preferred embodiments. These preferred embodiments are only used to explain the technical principles of the present disclosure, rather than to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts should still fall within the protection scope of the present disclosure.

[0023] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

[0024] The technical solutions proposed by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0025] An embodiment provides a method for optimizing resources of a hyper-converged system based on deep learning. As Figure 1 shown, the method provided by the embodiment of the present application mainly includes the following steps: Step 110: Obtain the operation data of the hyper-converged system within a preset historical time period.

[0026] It should be noted that the operation data at least includes: task data of each resource node, system operation performance, and the task data at least includes: task type of the current resource node running tasks, operation load of the running tasks, and allocatability of the running tasks.

[0027] In some embodiments, obtaining the operation data of the hyper-converged system within a preset historical time period specifically includes: Deploy lightweight monitoring agents on each resource node to collect the task type of the running tasks and the operation load of the running tasks of the current resource node; Obtain the calling program and calling data involved in the running task. When the calling program exists in all resource nodes and the calling data is not preset encrypted data, determine that the allocatability of the current running task is allocatable; otherwise, determine that the allocatability of the current running task is non-allocatable.

[0028] Those skilled in the art can understand that by deploying a lightweight monitoring agent, core metrics such as task type and load of each node can be collected in real time, avoiding the lag of manual monitoring. For example, an enterprise discovers through the monitoring agent that a certain node has task backlogs due to sudden high load, and promptly triggers the subsequent dynamic allocation process. Additionally, through the dual verification of the calling program and encrypted data, non-migratable tasks are prevented from being misallocated. For example, a medical imaging system prevents unauthorized cross-node migration of AI diagnostic tasks by verifying the hash value of the calling program, ensuring data compliance.

[0029] Step 120: Use the running data as training data for the deep learning model, and obtain the correlation relationship within the running data by using the trained deep learning model.

[0030] It should be noted that the deep learning model is trained with multi-dimensional data and can identify the non-linear correlation between task types and performance fluctuations. For example, it discovers a strong correlation between "promotion calculation tasks" and memory load. Additionally, this application can train the deep learning model periodically, and thus the periodic model update mechanism can adapt to hardware upgrades or changes in business load.

[0031] Step 130: Obtain the running data and system running performance of each resource node corresponding to the hyper-converged system at intervals of a preset update time period; screen out the running tasks with allocatability as allocatable from all the running data, match the screened running tasks with the resource nodes to obtain all permutations and combinations; generate the task data of the resource nodes under each combination; use the generated task data as input data to input into the deep learning model to obtain the predicted system running performance, and obtain the permutation and combination corresponding to the highest predicted system running performance.

[0032] It should be noted that the specific solution for obtaining the permutations and combinations of the running tasks with allocatability as allocatable on each resource node can be: use algorithms (such as permutation and combination algorithms, greedy algorithms, heuristic algorithms, etc.) to generate all possible task permutations and combinations. Here, each permutation and combination represents a specific task allocation scheme, that is, the distribution of each task on the resource nodes.

[0033] To reduce the computing power consumption of the hyper-converged system, this application can send the deep learning model to each resource node, and then utilize the computing power of the resource nodes to achieve fast data processing. The specific implementation process can be: Before using the generated task data as input data to a deep learning model to obtain the predicted system running performance, install the trained deep learning model on the resource nodes; determine the splitting ratio of the permutation and combination according to the ratio between the running memories of the resource nodes, and then obtain the permutation and combination to be tested corresponding to each resource node.

[0034] Among them, using the generated task data as input data to a deep learning model to obtain the predicted system running performance specifically includes: The resource node obtains the permutation and combination to be tested sent down, and sequentially uses a set of combinations in the permutation and combination to be tested as the input of the trained deep learning model on the current node to obtain the predicted system running performance of the output; upload the predicted system running performance to the hyper-converged system so that the hyper-converged system can obtain the predicted system running performance of all combinations.

[0035] Those skilled in the art can understand that in this application, by distributing the deep learning model to each resource node, distributed processing of data is achieved. Each resource node independently performs prediction processing, improving the overall processing efficiency and shortening the prediction time. The distribution of the permutation and combination to be tested on the resource nodes can be flexibly adjusted according to the computing power differences of the nodes. This can ensure that the task data is reasonably distributed to each node, achieving load balancing and avoiding the situation where some nodes are overloaded while other nodes are idle. Through comprehensive prediction of the permutation and combination to be tested, the hyper-converged system can obtain the predicted system running performance under all combinations. This provides comprehensive performance evaluation data for system administrators, helping them make more accurate decisions and select the optimal task allocation plan. During actual operation, system administrators can dynamically adjust the task allocation strategy according to the prediction results and actual requirements. For example, when it is found that the prediction performance of a certain resource node is low, the task allocation can be adjusted in time, and some tasks of this node can be migrated to other nodes with higher performance, thereby optimizing the overall performance of the system.

[0036] Step 140: Determine that the allocation node of the allocable running task corresponding to the permutation and combination with the highest predicted system running performance is the resource node of the allocable running task within the preset update time period.

[0037] In some embodiments, determining that the allocation node of the allocable running task corresponding to the permutation and combination with the highest predicted system running performance is the resource node of the allocable running task within the preset update time period can specifically be: When there is an allocable running task to be distributed outward on the resource node, send the program name and call data to the running node of the allocable running task corresponding to the permutation and combination with the highest predicted system running performance.

[0038] Those skilled in the art can understand that through the prediction of the deep learning model in this step, the optimal task allocation scheme can be found, thereby improving the overall performance of the system. The prediction results are based on a large amount of historical data and real-time information, so they have high accuracy and reliability. The task reallocation process takes into account the computing power differences and task requirements of each resource node, realizing the reasonable allocation and efficient utilization of resources. It avoids the situation where some nodes are overloaded while others are idle, and improves the resource utilization rate of the system. Assigning tasks to the most suitable nodes can shorten the task execution time and improve the task execution efficiency. At the same time, because the task allocation is more reasonable, it can also reduce the conflicts and waiting time between tasks, further improving the system performance.

[0039] Based on the above content, this application provides a resource optimization method, system and medium for a hyper-converged system based on deep learning. By obtaining the running data (task type, load, allocatability) and system performance of each resource node in real time, and combining the deep learning model to predict the system performance under different task permutations and combinations, it can dynamically respond to the load changes and resource requirements of the hyper-converged system. Compared with the traditional scheduling scheme based on static rules or offline data training, it breaks through the real-time bottleneck and solves the problem of lagging resource allocation. For example, in high-concurrency scenarios, it can quickly identify overloaded nodes and trigger task migration. Using the deep learning model to mine the non-linear correlation relationships between parameters such as task type, load distribution, and allocatability from historical operation data, such as the implicit dependency relationship between storage-intensive tasks and network bandwidth occupancy. This process overcomes the optimization blind spots caused by traditional methods relying only on local metrics (such as CPU utilization), realizes the global perception of the system operation state, and thus avoids the overall performance degradation caused by local optimization. By generating all permutations and combinations of allocable tasks and inputting them into the model to predict the system performance, this method quantifies the impact of different resource allocation strategies on the system. For example, when deploying compute-intensive and I / O-intensive tasks in a hybrid manner, the model can predict the increased latency caused by resource competition between nodes and select the permutation and combination with the least conflicts. This mechanism solves the problem that the existing technology lacks the ability of quantitative evaluation, and makes the optimization decision shift from experience-driven to data-driven. It solves the problems that the existing solutions are difficult to adapt to the real-time operation scenario of the hyper-converged system and cannot capture the global relevance of the system operation state.

[0040] In addition, this application Figure 2 is a resource optimization system for a hyper-converged system based on deep learning provided by an embodiment of this application. As Figure 2 shown, the system provided by the embodiment of this application mainly includes: An acquisition module 210 is configured to acquire the operation data of the hyper-converged system within a preset historical time period; wherein, the operation data at least includes: the task data of each resource node and the system operation performance, and the task data at least includes: the task type of the running task of the current resource node, the running load of the running task, and the allocatability of the running task.

[0041] The acquisition module 210 includes an acquisition unit, which is used to deploy lightweight monitoring agents on each resource node to collect the task type of the running tasks and the running load of the running tasks of the current resource node; acquire the calling program and calling data involved in the running task, and when the calling program exists in all resource nodes and the calling data is not preset encrypted data, determine that the allocatability of the current running task is allocable; otherwise, determine that the allocatability of the current running task is non-allocable.

[0042] An obtaining module 220 is configured to use the operation data as training data for a deep learning model, and use the trained deep learning model to obtain the correlation relationship among the operation data.

[0043] A prediction module 230 is configured to acquire the operation data of each resource node corresponding to the hyper-converged system and the system operation performance at intervals of a preset update time period; screen out the running tasks with allocatability as allocable from all the operation data, match the screened running tasks with the resource nodes to obtain all permutations and combinations; generate the task data of the resource nodes under each combination; use the generated task data as input data to input into the deep learning model to obtain the predicted system operation performance, and obtain the permutation and combination corresponding to the highest predicted system operation performance.

[0044] The prediction module 230 includes an allocation unit, which is used to install the trained deep learning model on the resource node; determine the splitting ratio of the permutation and combination according to the ratio between the running memories of the resource nodes, and further obtain the permutations and combinations to be tested corresponding to each resource node.

[0045] The prediction module 230 includes a prediction unit, which is used to obtain the permutations and combinations to be tested sent down through the resource nodes, and sequentially use a group of combinations in the permutations and combinations to be tested as the input of the trained deep learning model on the current node to obtain the output predicted system operation performance; upload the predicted system operation performance to the hyper-converged system so that the hyper-converged system can obtain the predicted system operation performance of all combinations.

[0046] Those skilled in the art can understand that the present application realizes the distributed processing of data by sending the deep learning model to each resource node. Each resource node independently performs prediction processing, improving the overall processing efficiency and shortening the prediction time. The allocation of the permutations and combinations to be measured on the resource nodes can be flexibly adjusted according to the computing power differences of the nodes. This can ensure that the task data is reasonably allocated to each node, achieving load balancing and avoiding the situation where some nodes are overloaded while other nodes are idle. Through the comprehensive prediction of the permutations and combinations to be measured, the hyper-converged system can obtain the predicted system operation performance under all combinations. This provides comprehensive performance evaluation data for system administrators, helping them make more accurate decisions and select the optimal task allocation scheme. During the actual operation process, the system administrator can dynamically adjust the task allocation strategy according to the prediction results and actual requirements. For example, when it is found that the prediction performance of a certain resource node is low, the task allocation can be adjusted in a timely manner, and some tasks of this node can be migrated to other nodes with higher performance, thereby optimizing the overall performance of the system.

[0047] The allocation module 240 is used to determine that the allocation node of the allocable running tasks under the permutation and combination corresponding to the highest predicted system operation performance is the resource node of the allocable running tasks within the preset update time period.

[0048] Those skilled in the art can understand that through the prediction of the deep learning model, the allocation module 240 can find the optimal task allocation scheme, thereby improving the overall performance of the system. The prediction results are based on a large amount of historical data and real-time information, so they have high accuracy and reliability. The task reallocation process takes into account the computing power differences and task requirements of each resource node, realizing the reasonable allocation and efficient utilization of resources. It avoids the situation where some nodes are overloaded while other nodes are idle, and improves the resource utilization rate of the system. Allocating tasks to the most suitable nodes can shorten the task execution time and improve the task execution efficiency. At the same time, due to the more reasonable task allocation, it can also reduce the conflicts and waiting times between tasks, further improving the system performance.

[0049] In addition, the embodiment of the present application also provides a non-volatile computer storage medium, on which executable instructions are stored. When the executable instructions are executed, the above-mentioned resource optimization method of a deep learning-based hyper-converged system is realized.

[0050] So far, the technical solutions of the present disclosure have been described in combination with multiple embodiments of the foregoing text. However, it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principle of the present disclosure, those skilled in the art can split and combine the technical solutions in the above various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.

Claims

1. A resource optimization method for a hyper-converged system based on deep learning, characterized in that The method includes: Obtaining the operation data of the hyper-converged system within a preset historical time period; wherein, the operation data at least includes: task data of each resource node and system operation performance, and the task data at least includes: the task type of the operation task running on the current resource node, the operation load of the operation task, and the allocatability of the operation task; Taking the operation data as the training data of the deep learning model, and obtaining the correlation relationship among the internal operation data by using the trained deep learning model; Obtaining the operation data of each resource node corresponding to the hyper-converged system and the system operation performance at intervals of a preset update time period; screening out the operation tasks with allocatability as allocable from all the operation data, matching the screened operation tasks with the resource nodes to obtain all permutations and combinations; generating the task data of the resource nodes under each combination; taking the generated task data as the input data and inputting it into the deep learning model to obtain the predicted system operation performance, and obtaining the permutation and combination corresponding to the highest predicted system operation performance; Determining the allocation node of the allocable operation task corresponding to the permutation and combination with the highest predicted system operation performance as the resource node of the allocable operation task within the preset update time period.

2. The resource optimization method of the hyper-converged system based on deep learning according to claim 1, characterized in that, Obtaining the operation data of the hyper-converged system within a preset historical time period, specifically including: Deploying lightweight monitoring agents on each resource node to collect the task type of the operation task running on the current resource node and the operation load of the operation task; Obtaining the calling program and calling data involved in the operation task, and when the calling program exists on all resource nodes and the calling data is not preset encrypted data, determining that the allocatability of the current operation task is allocable; otherwise, determining that the allocatability of the current operation task is non-allocable.

3. The resource optimization method for the hyper-converged system based on deep learning according to claim 1, wherein Before taking the generated task data as the input data and inputting it into the deep learning model to obtain the predicted system operation performance, the method further includes: Installing the trained deep learning model on the resource node; Determining the segmentation ratio of the permutation and combination according to the ratio between the operation memories of the resource nodes, and further obtaining the permutations and combinations to be tested corresponding to each resource node.

4. The resource optimization method of the hyper-converged system based on deep learning according to claim 3, characterized in that Taking the generated task data as the input data and inputting it into the deep learning model to obtain the predicted system operation performance, specifically including: The resource node obtains the permutations and combinations to be tested sent down, and sequentially takes a group of combinations in the permutations and combinations to be tested as the input of the trained deep learning model on the current node to obtain the output predicted system operation performance; Uploading the predicted system operation performance to the hyper-converged system so that the hyper-converged system can obtain the predicted system operation performance of all combinations.

5. The resource optimization method for the hyper-converged system based on deep learning according to claim 1, wherein Determining the allocation node of the allocable operation task corresponding to the permutation and combination with the highest predicted system operation performance as the resource node of the allocable operation task within the preset update time period, specifically including: When there are allocable operation tasks to be allocated outward on the resource node, sending the calling program name and calling data to the operation node of the allocable operation task corresponding to the permutation and combination with the highest predicted system operation performance.

6. A resource optimization system for a hyper-converged system based on deep learning, characterized in that, The system includes: An acquisition module, configured to acquire the operation data of the hyper-converged system within a preset historical time period; wherein, the operation data at least includes: the task data of each resource node and the system operation performance, and the task data at least includes: the task type of the operation task running on the current resource node, the operation load of the operation task, and the allocatability of the operation task; A obtaining module, configured to use the operation data as the training data of the deep learning model, and obtain the correlation relationship among the operation data by using the trained deep learning model; A prediction module, configured to acquire the operation data and the system operation performance of each resource node corresponding to the hyper-converged system at intervals of a preset update time period; screen out the operation tasks with allocatability as allocable from all the operation data, match the screened operation tasks with the resource nodes to obtain all permutations and combinations; generate the task data of the resource nodes under each combination; use the generated task data as input data to input into the deep learning model to obtain the predicted system operation performance, and obtain the permutation and combination corresponding to the highest predicted system operation performance; An allocation module, configured to determine that the allocation node of the allocable operation task under the permutation and combination corresponding to the highest predicted system operation performance is the resource node of the allocable operation task within the preset update time period.

7. The resource optimization system of the hyper-converged system based on deep learning according to claim 6, wherein The acquisition module includes an acquisition unit, configured to deploy a lightweight monitoring agent on each resource node to collect the task type of the operation task running on the current resource node and the operation load of the operation task; Acquire the calling program and the calling data involved in the operation task. When the calling program exists on all resource nodes and the calling data is not preset encrypted data, determine that the allocatability of the current operation task is allocable; otherwise, determine that the allocatability of the current operation task is non-allocable.

8. The resource optimization system of the hyper-converged system based on deep learning according to claim 6, characterized in that The prediction module includes an allocation unit, configured to install the trained deep learning model on the resource node; Determine the splitting ratio of the permutation and combination according to the ratio between the operation memories of the resource nodes, and further obtain the permutations and combinations to be tested corresponding to each resource node.

9. The resource optimization system for a hyper-converged system based on deep learning according to claim 8, wherein The prediction module includes a prediction unit, configured to obtain the permutations and combinations to be tested sent down through the resource node, and sequentially use a group of combinations in the permutations and combinations to be tested as the input of the trained deep learning model on the current node to obtain the output predicted system operation performance; Upload the predicted system operation performance to the hyper-converged system so that the hyper-converged system can obtain the predicted system operation performance of all combinations.

10. A non-volatile computer storage medium, characterized in that, It stores computer instructions, and the computer instructions, when executed, implement a method for optimizing the resources of a hyper-converged system based on deep learning as described in any one of claims 1-5.

Images