Method and system for scheduling computing power resources for model training

Abstract

The application relates to the field of big data, and discloses a computing resource scheduling method and system for model training. The cloud computing resource scheduling server receives computing resource request information, judges the legality of the computing resource request according to the computing resource authorization information in the computing resource request information, obtains model training task information in the computing resource request information if the request is legal, obtains training features in the model training task information, performs feature matching in the cloud computing resource scheduling server, and completes the model training task. Through dynamic analysis and intelligent scheduling, the application can more effectively utilize cloud computing resources, avoid resource idling or overload, and significantly improve resource utilization.

Description

Technical Field

[0001] This invention relates to the field of big data, specifically to a method and system for scheduling computing resources for model training. Background Technology

[0002] With the rapid development of artificial intelligence technology, deep learning models have been widely used in various fields, from image recognition and speech recognition to natural language processing, demonstrating their powerful data processing and pattern recognition capabilities. However, the training process of deep learning models often involves huge demands on high-performance computing resources, especially when dealing with large-scale datasets or complex models. A single computing node often cannot meet the computing power required for training, which necessitates efficient computing resource scheduling methods to optimize resource allocation and improve training efficiency.

[0003] Traditional computing resource scheduling methods are mostly based on static allocation strategies, which allocate fixed computing resources to different training tasks according to preset rules or strategies. This method is inflexible when facing changing task requirements and dynamic resource states, easily leading to uneven resource utilization, with some resources overloaded while others are idle, severely impacting overall training efficiency and resource utilization. Furthermore, with the widespread adoption of cloud computing technology, cloud computing resources have become increasingly abundant and diverse. How to efficiently and dynamically schedule these distributed computing resources to meet the needs of different training tasks has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for scheduling computing resources for model training, comprising the following steps:

[0005] Step 1: The cloud computing power scheduling server receives computing power resource request information and determines the legality of the computing power resource request based on the computing power resource authorization information in the request information. If it is a legal request, proceed to Step 2; otherwise, reject the computing power resource request and issue a warning message.

[0006] Step 2: Obtain the model training task information from the computing power resource request information, obtain the training features from the model training task information, and perform feature matching in the cloud computing power scheduling server. If a training container with the corresponding training features is matched, the model training task is a repetitive training task type, and proceed to Step 3; otherwise, it is a second training task type, and proceed to Step 4.

[0007] Step 3: Based on the matching training containers for each corresponding training feature, obtain the training container sequence for the corresponding training feature. According to the computing power requirement information in the model training task information, match the training container sequence for the corresponding training feature. If a training container with the corresponding computing power requirement information is matched, send the model training task to the task execution list of the matching training container with the corresponding computing power requirement information. Otherwise, select the training container with the smallest computing power difference with the computing power requirement in the training container sequence for the corresponding training feature according to the computing power requirement information, send the model training task to the merged training container, perform model training, and proceed to Step 7.

[0008] Step 4: The cloud computing power scheduling server establishes a training container corresponding to the second training feature based on the second training feature in the computing power resource request information, and matches the training container corresponding to the second computing power demand information in the cloud computing power scheduling server based on the second computing power demand information in the computing power resource request information. If a training container corresponding to the second computing power demand information is matched, proceed to step 5; otherwise, proceed to step 6.

[0009] Step 5: Obtain the training container sequence corresponding to the second computing power requirement information, and obtain the calling characteristics of each training container in the training container sequence corresponding to the second computing power requirement information. Based on the calling characteristics, obtain the scheduling training container, connect the scheduling training container to the training container corresponding to the second training feature, perform model training, and proceed to Step 7.

[0010] Step 6: The cloud computing power scheduling server generates a computing power container corresponding to the second computing power requirement based on the second computing power requirement information, and connects it to the training container corresponding to the second training feature to train the model, and then proceeds to Step 7.

[0011] Step 7: Complete the model training task.

[0012] Furthermore, the determination of the legality of the computing resource request based on the computing resource authorization information in the computing resource request information includes:

[0013] The computing power resource authorization information includes authorization instructions and authorized computing power usage scope; if the authorization instruction is successfully verified and the authorized computing power usage scope is within the set computing power usage scope of a single training task, then the computing power resource request is valid; otherwise, it is invalid.

[0014] Furthermore, the process of obtaining model training task information from computing resource request information, obtaining training features from the model training task information, and performing feature matching on the cloud computing power scheduling server includes:

[0015] The training features are one or more of the following: dataset size, training rounds, and model size; the training container in the cloud computing power scheduling server includes computing power size and training features, and different training containers include different computing power sizes and training features; the corresponding training container is matched according to the training features in the training task information.

[0016] Furthermore, the step of obtaining a training container sequence for each corresponding training feature based on the matched training containers, and matching the training container sequence for each corresponding training feature according to the computing power requirement information in the model training task information, includes:

[0017] The remaining time for model training task completion is obtained for each training container corresponding to a training feature. Based on the remaining time, from shortest to longest, a sequence of training containers for each training feature is obtained. The computational power difference between the computational power requirement in the model training task information and the computational power of each training container in the corresponding training feature sequence is obtained. Based on the computational power difference and the remaining time for model training task completion, the training container matching degree is obtained. The training container corresponding to the minimum matching degree is the matched training container. The training container matching degree is:

[0018] Training container matching degree = duration weight × remaining time for model training task completion + computing power difference × computing power weight.

[0019] Furthermore, the step of matching a training container corresponding to the second computing power requirement information in the cloud computing power scheduling server based on the second computing power requirement information in the computing power resource request information includes:

[0020] In the cloud computing power scheduling server, obtain the training container whose computing power is greater than the second computing power requirement information, and obtain the training container that matches the corresponding second computing power requirement information.

[0021] Furthermore, obtaining the training container sequence corresponding to the second computing power requirement information, and respectively acquiring the calling characteristics of each training container in the training container sequence corresponding to the second computing power requirement information, includes:

[0022] The calling characteristic is the network transmission latency of the training container, where the training container with the lowest network latency is the scheduled training container.

[0023] The computing resource scheduling system for model training, which applies the computing resource scheduling method for model training, includes a cloud computing resource scheduling server, a communication module, a computing resource request module, a data processing module, an early warning module, and a display module.

[0024] The computing power request module, communication module, early warning module, and display module are respectively connected to the data processing module; the cloud computing power scheduling server is communicatively connected to the communication module.

[0025] The beneficial effects of this invention are: improved resource utilization: through dynamic analysis and intelligent scheduling, this invention can more effectively utilize cloud computing resources, avoid resource idleness or overload, and significantly improve resource utilization.

[0026] Shorten model training time: By accurately matching training features and computing power requirements, and optimizing network latency, this invention can accelerate model training and shorten the training cycle.

[0027] Enhanced system flexibility: For special or newly emerging training tasks, the system can dynamically generate new computing power containers, enhancing the system's flexibility and scalability.

[0028] Ensuring resource security: Through a strict legality verification mechanism, this invention ensures the secure use of computing resources and prevents unauthorized or unauthorized access to resources. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the computing resource scheduling method used for model training.

[0030] Figure 2 This is a schematic diagram of the computing resource scheduling system used for model training. Detailed Implementation

[0031] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0032] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0033] like Figure 1 As shown, the computing resource scheduling method for model training includes the following steps:

[0034] Step 1: The cloud computing power scheduling server receives computing power resource request information and determines the legality of the computing power resource request based on the computing power resource authorization information in the request information. If it is a legal request, proceed to Step 2; otherwise, reject the computing power resource request and issue a warning message.

[0035] Step 2: Obtain the model training task information from the computing power resource request information, obtain the training features from the model training task information, and perform feature matching in the cloud computing power scheduling server. If a training container with the corresponding training features is matched, the model training task is a repetitive training task type, and proceed to Step 3; otherwise, it is a second training task type, and proceed to Step 4.

[0036] Step 3: Based on the matching training containers for each corresponding training feature, obtain the training container sequence for the corresponding training feature. According to the computing power requirement information in the model training task information, match the training container sequence for the corresponding training feature. If a training container with the corresponding computing power requirement information is matched, send the model training task to the task execution list of the matching training container with the corresponding computing power requirement information. Otherwise, select the training container with the smallest computing power difference with the computing power requirement in the training container sequence for the corresponding training feature according to the computing power requirement information, send the model training task to the merged training container, perform model training, and proceed to Step 7.

[0037] Step 4: The cloud computing power scheduling server establishes a training container corresponding to the second training feature based on the second training feature in the computing power resource request information, and matches the training container corresponding to the second computing power demand information in the cloud computing power scheduling server based on the second computing power demand information in the computing power resource request information. If a training container corresponding to the second computing power demand information is matched, proceed to step 5; otherwise, proceed to step 6.

[0038] Step 5: Obtain the training container sequence corresponding to the second computing power requirement information, and obtain the calling characteristics of each training container in the training container sequence corresponding to the second computing power requirement information. Based on the calling characteristics, obtain the scheduling training container, connect the scheduling training container to the training container corresponding to the second training feature, perform model training, and proceed to Step 7.

[0039] Step 6: The cloud computing power scheduling server generates a computing power container corresponding to the second computing power requirement based on the second computing power requirement information, and connects it to the training container corresponding to the second training feature to train the model, and then proceeds to Step 7.

[0040] Step 7: Complete the model training task.

[0041] The determination of the legality of a computing resource request based on the computing resource authorization information in the computing resource request information includes:

[0042] The computing power resource authorization information includes authorization instructions and authorized computing power usage scope; if the authorization instruction is successfully verified and the authorized computing power usage scope is within the set computing power usage scope of a single training task, then the computing power resource request is valid; otherwise, it is invalid.

[0043] The process of obtaining model training task information from computing resource request information, obtaining training features from the model training task information, and performing feature matching on the cloud computing power scheduling server includes:

[0044] The training features are one or more of the following: dataset size, training rounds, and model size; the training container in the cloud computing power scheduling server includes computing power size and training features, and different training containers include different computing power sizes and training features; the corresponding training container is matched according to the training features in the training task information.

[0045] The step of obtaining a training container sequence for each corresponding training feature based on the matched training containers, and matching the training container sequence for each corresponding training feature according to the computing power requirement information in the model training task information, includes:

[0046] The remaining time for model training task completion is obtained for each training container corresponding to a training feature. Based on the remaining time, from shortest to longest, a sequence of training containers for each training feature is obtained. The computational power difference between the computational power requirement in the model training task information and the computational power of each training container in the corresponding training feature sequence is obtained. Based on the computational power difference and the remaining time for model training task completion, the training container matching degree is obtained. The training container corresponding to the minimum matching degree is the matched training container. The training container matching degree is:

[0047] Training container matching degree = duration weight × remaining time for model training task completion + computing power difference × computing power weight.

[0048] The method of matching training containers corresponding to the second computing power requirement information in the cloud computing power scheduling server based on the second computing power requirement information in the computing power resource request information includes:

[0049] In the cloud computing power scheduling server, obtain the training container whose computing power is greater than the second computing power requirement information, and obtain the training container that matches the corresponding second computing power requirement information.

[0050] The process of obtaining the training container sequence corresponding to the second computing power requirement information, and acquiring the call characteristics of each training container in the training container sequence corresponding to the second computing power requirement information, includes:

[0051] The calling characteristic is the network transmission latency of the training container, where the training container with the lowest network latency is the scheduled training container.

[0052] like Figure 2As shown, the computing resource scheduling system for model training, which applies the computing resource scheduling method for model training, includes a cloud computing resource scheduling server, a communication module, a computing request module, a data processing module, an early warning module, and a display module.

[0053] The computing power request module, communication module, early warning module, and display module are respectively connected to the data processing module; the cloud computing power scheduling server is communicatively connected to the communication module.

[0054] Specifically, this invention provides a method for scheduling computing resources for model training, which includes the following key steps:

[0055] Step 1: Legality Verification

[0056] The cloud-based computing power scheduling server first receives computing power resource request information from users. This information includes computing power resource authorization information used to verify the legitimacy of the request. The authorization information specifically includes authorization instructions and the authorized scope of computing power usage. The server determines the legitimacy of the request by verifying the validity of the authorization instructions and ensuring that the requested computing power usage scope does not exceed the preset computing power usage limit for a single training task. If the request is legitimate, it proceeds to the next step; otherwise, the request is rejected and a warning message is issued to prevent illegal or unauthorized use of resources.

[0057] Step 2: Training Task Type Recognition

[0058] After confirming the request's legitimacy, the server extracts model training task information, particularly training features such as dataset size, training epochs, and model scale, from the computing resource request information. Internally, the server maintains a training container library, each predefined with its supported computing power and training features. Through feature matching, the server can identify whether the current task is a repetitive training task (i.e., a task with the same or similar features has already been processed). If it is a repetitive task, proceed to step three; otherwise, it is considered a second training task type and proceeds to step four.

[0059] Step 3: Repetitive Training Task Scheduling

[0060] For repetitive training tasks, the server searches for all training containers that match the current training features and generates a sequence of training containers based on the remaining time and computing power required to complete the model training task for each container. Then, based on the computing power requirements of the current task, it searches the sequence for the most suitable training container. If a perfectly matching container is found, the task is directly added to that container's task execution list. If no perfectly matching container is found, the container with the smallest difference in computing power and the shortest remaining time to complete the task is selected, and the task is merged into it to maximize resource utilization.

[0061] Step 4: Scheduling the Second Training Task Type

[0062] For the second training task type, the server needs to create a corresponding training container based on its unique training characteristics. Next, the server will search its existing resources for a training container that meets the computational power requirements of the task. If found, scheduling will proceed according to step five; otherwise, step six will be executed to dynamically generate a new computational power container.

[0063] Step 5: Schedule the connection of the training container

[0064] After finding a training container that matches the second computing power requirement, the server will further consider the container's calling characteristics, such as network transmission latency, select the container with the lowest latency as the scheduled training container, and connect it to the newly created training container to ensure efficient data transmission and model training.

[0065] Step Six: Generate a new computing power container

[0066] If the existing resources cannot meet the computing power requirements of the second training task, the server will dynamically generate a new computing power container based on the computing power requirements of the task and connect it to the corresponding training container to meet the specific requirements of the task.

[0067] Step 7: Complete the model training task

[0068] Whether it's a repetitive task or a second training task, the model training will eventually be completed in the selected training container (or combination of containers). The server will monitor the training process until the task is completed.

[0069] This invention ensures the legality and security of resource use through dual verification (authorization instructions + computing power usage scope).

[0070] By leveraging training features for precise matching, and combining computing power requirements with remaining task completion time, intelligent container selection and task scheduling are achieved. For computing power demands that cannot be met, new computing power containers can be dynamically generated, improving the system's flexibility and adaptability. When selecting and scheduling training containers, network transmission latency is considered to reduce data transmission time and improve training efficiency.

[0071] Example 1: Efficient scheduling of repetitive training tasks

[0072] Suppose a deep learning research team frequently uses the same model architecture and dataset for different training rounds to optimize model performance. They submit a computing resource request through the computing resource scheduling system of this invention, hoping to utilize cloud computing power for model training.

[0073] Legality Verification: The research team submitted a computing resource request through the computing power request module, containing authorization instructions and the scope of authorized computing power usage. Upon receiving the request, the cloud-based computing power scheduling server first verifies the validity of the authorization instructions and checks whether the requested computing power usage scope is within the preset limit for a single training task. If the verification is successful, the request is confirmed as legitimate and proceeds to the next step.

[0074] Training task type identification: The server extracts model training task information from the request information, including training features such as dataset size, training epochs, and model size. By matching features with containers in the training container library, the server identifies that this is a repetitive training task because a task with the same features has already been processed.

[0075] Repeated training task scheduling: The server searches for all training containers that match the current training features and generates a sequence of training containers based on the remaining time and computing power required to complete the model training task for each container. Suppose a container A is found with computing power similar to the current task requirement and a short remaining time to complete the task, but no perfectly matching container is found. The server selects container A and merges the current task into the task execution list of container A to maximize resource utilization.

[0076] Model training task completion: The server monitors the model training process in container A until the task is completed. The research team views the training progress and resource usage in real time through the display module and is satisfied with the scheduling efficiency.

[0077] Through the computing resource scheduling method of this invention, the research team's repetitive training tasks were efficiently scheduled, which not only shortened the model training time but also improved the utilization rate of cloud computing resources.

[0078] Example 2: Dynamic resource scheduling for the second training task type

[0079] An AI startup is developing a new deep learning model with unique training characteristics and high computational requirements. They decided to use the computational resource scheduling system of this invention to apply for cloud computing resources.

[0080] Legality Verification: The AI ​​company submits a computing resource request through the computing power request module, which includes authorization instructions and the scope of authorized computing power usage. The cloud computing power scheduling server verifies the legality of the request. Once the request is confirmed to be legitimate, it proceeds to the next step of processing.

[0081] Training task type identification: The server extracts model training task information from the request information, including unique training features such as dataset size, training epochs, and model size. By matching features with containers in the training container library, the server identifies this as a second training task type because it has not previously processed tasks with the same features.

[0082] Second training task type scheduling: Based on the unique training characteristics of the task, the server creates a corresponding training container B. Next, the server searches for a training container in the existing resources that can meet the computing power requirements of the task. Since the task's computing power requirements are high, none of the existing containers can meet them. Therefore, the server proceeds to step six, dynamically generating a new computing power container C and connecting it to the newly created training container B.

[0083] Scheduling training container connections: After generating a new computing power container C, the server considers the container's calling characteristics, such as network transmission latency, and selects the path with the least latency to connect container C to training container B.

[0084] The model training task is completed by the server monitoring the training process in training container B and computing power container C, ensuring efficient data transfer and model training. The AI ​​company uses a display module to view training progress and resource usage in real time, and highly appreciates the system's ability to dynamically generate new computing power containers to meet specific needs.

[0085] Through the computing power resource scheduling method of the present invention, the AI ​​company's second training task was successfully scheduled and executed, which not only met the high computing power requirements, but also demonstrated the system's flexibility and adaptability in the face of new tasks.

Claims

1. A method for scheduling computing resources for model training, characterized in that, Includes the following steps: Step 1: The cloud computing power scheduling server receives computing power resource request information and determines the legality of the computing power resource request based on the computing power resource authorization information in the request information. If it is a legal request, proceed to Step 2; otherwise, reject the computing power resource request and issue a warning message. Step 2: Obtain the model training task information from the computing power resource request information, obtain the training features from the model training task information, and perform feature matching in the cloud computing power scheduling server. If a training container with the corresponding training features is matched, the model training task is a repetitive training task type, and proceed to Step 3; otherwise, it is a second training task type, and proceed to Step 4. Step 3: Based on the matching training containers for each corresponding training feature, obtain the training container sequence for the corresponding training feature. According to the computing power requirement information in the model training task information, match the training container sequence for the corresponding training feature. If a training container with the corresponding computing power requirement information is matched, send the model training task to the task execution list of the matching training container with the corresponding computing power requirement information. Otherwise, select the training container with the smallest computing power difference with the computing power requirement in the training container sequence for the corresponding training feature according to the computing power requirement information, send the model training task to the merged training container, perform model training, and proceed to Step 7. Step 4: The cloud computing power scheduling server establishes a training container corresponding to the second training feature based on the second training feature in the computing power resource request information, and matches the training container corresponding to the second computing power demand information in the cloud computing power scheduling server based on the second computing power demand information in the computing power resource request information. If a training container corresponding to the second computing power demand information is matched, proceed to step 5; otherwise, proceed to step 6. Step 5: Obtain the training container sequence corresponding to the second computing power requirement information, and obtain the calling characteristics of each training container in the training container sequence corresponding to the second computing power requirement information. Based on the calling characteristics, obtain the scheduling training container, connect the scheduling training container to the training container corresponding to the second training feature, perform model training, and proceed to Step 7. Step 6: The cloud computing power scheduling server generates a computing power container corresponding to the second computing power requirement based on the second computing power requirement information, and connects it to the training container corresponding to the second training feature to train the model, and then proceeds to Step 7. Step 7: Complete the model training task.

2. The computing resource scheduling method for model training according to claim 1, characterized in that, The determination of the legality of a computing resource request based on the computing resource authorization information in the computing resource request information includes: The computing power resource authorization information includes authorization instructions and authorized computing power usage scope; if the authorization instruction is successfully verified and the authorized computing power usage scope is within the set computing power usage scope of a single training task, then the computing power resource request is valid; otherwise, it is invalid.

3. The computing resource scheduling method for model training according to claim 2, characterized in that, The process of obtaining model training task information from computing resource request information, obtaining training features from the model training task information, and performing feature matching on the cloud computing power scheduling server includes: The training features are one or more of the following: dataset size, training rounds, and model size; the training container in the cloud computing power scheduling server includes computing power size and training features, and different training containers include different computing power sizes and training features; the corresponding training container is matched according to the training features in the training task information.

4. The computing resource scheduling method for model training according to claim 3, characterized in that, The step of obtaining a training container sequence for each corresponding training feature based on the matched training containers, and matching the training container sequence for each corresponding training feature according to the computing power requirement information in the model training task information, includes: The remaining time for model training task completion is obtained for each training container corresponding to a training feature. Based on the remaining time, from shortest to longest, a sequence of training containers for each training feature is obtained. The computational power difference between the computational power requirement in the model training task information and the computational power of each training container in the corresponding training feature sequence is obtained. Based on the computational power difference and the remaining time for model training task completion, the training container matching degree is obtained. The training container corresponding to the minimum matching degree is the matched training container. The training container matching degree is: Training container matching degree = duration weight × remaining time for model training task completion + computing power difference × computing power weight.

5. The computing resource scheduling method for model training according to claim 3, characterized in that, The method of matching training containers corresponding to the second computing power requirement information in the cloud computing power scheduling server based on the second computing power requirement information in the computing power resource request information includes: In the cloud computing power scheduling server, obtain the training container whose computing power is greater than the second computing power requirement information, and obtain the training container that matches the corresponding second computing power requirement information.

6. The computing resource scheduling method for model training according to claim 5, characterized in that, The process of obtaining the training container sequence corresponding to the second computing power requirement information, and acquiring the call characteristics of each training container in the training container sequence corresponding to the second computing power requirement information, includes: The calling characteristic is the network transmission latency of the training container, where the training container with the lowest network latency is the scheduled training container.

7. A computing resource scheduling system for model training, characterized in that, The computing power resource scheduling method for model training according to any one of claims 1-6 includes a cloud computing power scheduling server, a communication module, a computing power request module, a data processing module, an early warning module, and a display module; The computing power request module, communication module, early warning module, and display module are respectively connected to the data processing module; the cloud computing power scheduling server is communicatively connected to the communication module.

Images