Resource scheduling system, method and device and computer equipment

Abstract

The invention discloses a resource scheduling system, method and device and computer equipment. The system comprises a server cluster module, a service mirror image warehouse module and a service management module which are arranged in a medical institution. The service mirror image warehouse module comprises a first service mirror image and a second service mirror image which are used for medical auxiliary diagnosis; the service management module is used for scheduling resources from first computing power resources to start instances corresponding to the first target service mirror images if the first computing power resources in the server cluster module meet the resource requirements of the first target service mirror images; and the service management module is further used for scheduling resources from the second computing power resources to start instances corresponding to the second target service mirror images if the second computing power resources meet the resource requirements of the second target service mirror images. Therefore, cross-scene reuse of server resources supporting medical diagnosis service computing power in a medical institution is realized, and the efficiency of medical auxiliary diagnosis is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of server resource scheduling technology, and in particular to a resource scheduling system, method, apparatus and computer equipment. Background Technology

[0002] AI services for medical auxiliary diagnosis can help doctors make quick diagnoses. With the promotion and application of the cloud platform + AI service model for medical auxiliary diagnosis, it is of great significance to balance the efficiency and compliance of medical auxiliary diagnosis.

[0003] Among the related technologies, a solution of decentralizing computing power is adopted, deploying AI service environments within hospitals to ensure closed-loop management of the entire data lifecycle within the hospital, thereby meeting the compliance requirement of "data not leaving the hospital." However, the efficiency of medical auxiliary diagnosis in these technologies needs to be improved. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. To this end, this invention proposes a resource scheduling system, method, apparatus, and computer equipment to enable cross-scenario reuse of server resources supporting medical diagnostic services within medical institutions, thereby improving the utilization rate of server resources and effectively enhancing the efficiency of medical auxiliary diagnosis.

[0005] To achieve the above objectives, a first aspect of the present invention provides a resource scheduling system, comprising a server cluster module, a service image repository module, and a service management module deployed within a medical institution; the service image repository module includes a first service image and a second service image for medical auxiliary diagnosis; wherein the first service image is obtained from the server cluster module; and the second service image is obtained from an external platform; the service management module is configured to determine a first target service image from the first service image based on a first diagnostic task request; if the first computing power resources in the server cluster module meet the resource requirements of the first target service image, the module schedules resources from the first computing power resources to launch an instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the first diagnostic task request based on the instance of the first target service image; the service management module is further configured to, if a second target service image corresponding to the first diagnostic task request exists in the second service image, and the second computing power resources in the server cluster module meet the resource requirements of the second target service image, schedule resources from the second computing power resources to launch an instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein the resource requirements of the first target service image have a higher priority than the resource requirements of the second target service image.

[0006] According to one embodiment of the present invention, the service management module is specifically used to, when starting an instance corresponding to the first target service image, if the second target service image exists in the service image repository and the second computing power resources meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start the instance corresponding to the second target service image.

[0007] According to one embodiment of the present invention, the service management module is further configured to release the resources used to start the instance corresponding to the second target service image if a second diagnostic task request is obtained when starting the instance corresponding to the second target service image.

[0008] According to one embodiment of the present invention, the server cluster module is built based on the domain server nodes used for medical auxiliary diagnosis within the medical institution; the domain server nodes are server nodes registered in the server registration module.

[0009] According to one embodiment of the present invention, the second service image is specifically obtained from the external platform through an encrypted transmission channel.

[0010] According to one embodiment of the present invention, the server cluster module includes a first domain server node and a second domain server node; the service management module is used to upgrade the environment version of the first domain server node to the environment version of the second domain server node when the first domain server node and the second domain server node have the same hardware configuration, but the environment version of the first domain server node is lower than the environment version of the second domain server node.

[0011] According to one embodiment of the present invention, the service management module is further configured to, when starting instances corresponding to the first target service image and the second target service image respectively, compare the calculation results generated by the instance of the first target service image with the calculation results generated by the instance of the second target service image to obtain a comparison result.

[0012] To achieve the above objectives, a second aspect of the present invention provides a resource scheduling method, applied to a service management module in a resource scheduling system described in any of the foregoing embodiments. The method includes: obtaining a diagnostic task request; determining a first target service image from a first service image in a service image repository module based on the first diagnostic task request; if a first computing power resource in a server cluster module meets the resource requirements of the first target service image, scheduling resources from the first computing power resource to start an instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image; if a second target service image corresponding to the first diagnostic task request exists in a second service image in the service image repository, and a second computing power resource in the server cluster module meets the resource requirements of the second target service image, scheduling resources from the second computing power resource to start an instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

[0013] To achieve the above objectives, a third aspect of the present invention provides a resource scheduling apparatus, applied to the service management module of the resource scheduling system described in any of the preceding embodiments. The apparatus includes: a task request acquisition module for acquiring a diagnostic task request; a service image determination module for determining a first target service image from a first service image in a service image repository module based on the first diagnostic task request; a first instance startup module for scheduling resources from the first computing resources to start an instance corresponding to the first target service image if the first computing resources in the server cluster module meet the resource requirements of the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image; and a second instance startup module for scheduling resources from the second computing resources to start an instance corresponding to the second target service image if a second target service image corresponding to the first diagnostic task request exists in a second service image in the service image repository, and the second computing resources in the server cluster module meet the resource requirements of the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

[0014] To achieve the above objectives, a fourth aspect of the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the resource scheduling method in the foregoing embodiments.

[0015] According to multiple embodiments provided by this invention, by establishing a server cluster module and a service image repository module within the medical institution, and by using a service management module deployed within the medical institution to dynamically determine whether to invoke resources to start service instances based on the resource requirements of diagnostic tasks and server load, this system can achieve reasonable utilization of external service resources while meeting the core requirement of "data not leaving the hospital," thus significantly improving the resource utilization rate of locally deployed servers within the medical institution. Compared to the local service environment deployment mode in related technologies, this system can achieve cross-scenario reuse of server resources in medical scenarios and ensure the priority operation of core services, effectively improving the efficiency of medical auxiliary diagnosis.

[0016] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of a resource scheduling system according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of an edge node resource management architecture for medical data compliance scenarios provided by an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the registration and management process for edge nodes according to an embodiment of the present invention.

[0020] Figure 4 This is a flowchart illustrating a task processing method according to an embodiment of the present invention.

[0021] Figure 5 This is a flowchart illustrating a resource scheduling method according to an embodiment of the present invention.

[0022] Figure 6 This is a structural block diagram of a resource scheduling device according to an embodiment of the present invention.

[0023] In the diagram, 100: Resource scheduling system; 110: Server cluster module; 120: Service image repository module; 130: Service management module; 122: First service image; 124: Second service image; 600: Resource scheduling device; 610: Task request acquisition module; 620: Service image determination module; 630: First instance startup module; 640: Second instance startup module. Detailed Implementation

[0024] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0025] AI (Artificial Intelligence) products for medical auxiliary diagnosis can help doctors perform quantitative and qualitative analysis, greatly improving the accuracy of their diagnoses. The cloud platform + AI diagnostic model for medical auxiliary diagnosis is gradually being promoted and applied, gaining widespread user recognition. Users can upload their data to the cloud platform, where AI processes the data. The cloud platform then provides the results back to the user, who only needs a client PC to obtain the results obtained through complex calculations.

[0026] The cloud platform + AI-based collaborative model for healthcare can effectively reduce the cost of assisted diagnostic applications, but it still faces two major challenges. Firstly, the massive amounts of medical data place stringent demands on cloud network bandwidth resources and server cluster stability, potentially causing transmission delays during concurrent processing of multimodal images. Secondly, this model needs to adapt to healthcare institutions' localized control policies for sensitive data. For example, the "data not leaving the hospital" requirement explicitly stipulates that patient image data can only be circulated within the hospital's dedicated network and is prohibited from being transmitted over the public internet. Therefore, to balance the efficiency and compliance of assisted medical diagnosis, related technologies commonly adopt a "computing power decentralization" solution. This involves deploying a private AI environment within the hospital, purchasing closed-loop AI systems from various vendors to achieve localized independent computation, ensuring closed-loop management of the data throughout its entire lifecycle within the hospital.

[0027] However, private AI deployment faces significant efficiency and cost challenges. First, in the independent deployment model, the computing resources of each system cannot be reused across scenarios, resulting in insufficient resource utilization and redundant configuration. It is difficult to achieve intensive scheduling and dynamic allocation of computing resources, which limits the efficiency of medical auxiliary diagnosis and makes it difficult to control the overall TCO (Total Cost of Ownership). Second, the procurement cost of high-performance servers supporting medical AI computing power is high, especially for the diagnostic needs of multiple diseases in hospitals, which require the configuration of multiple dedicated servers, leading to a significant increase in the pressure of initial hardware investment and subsequent operation and maintenance expenses.

[0028] To improve the utilization rate of server resources supporting medical AI computing power within medical institutions, achieve cross-scenario reuse of server resources, and enhance the efficiency of medical auxiliary diagnosis, it is necessary to propose a resource scheduling system, method, device, and computer equipment. This system includes a server cluster module deployed locally within the medical institution, a service image repository module, and a service management module. The service image repository module includes a first service image obtained from the server cluster module and a second service image obtained from an external platform. The first service image may include images of core services deployed locally within the medical institution, while the second service image may include images of supplementary, auxiliary, or backup services. When the service management module receives a diagnostic task request, it first determines a first target service image from the first service image that matches the task request. If the idle computing power resources in the server cluster module can meet the resource requirements of the first target service image, it then calls upon resources to start an instance of the first target service image.

[0029] If a second target service image exists in the second service image that can be used to execute the diagnostic task corresponding to the diagnostic task request, the service management module determines whether the idle second computing resources in the server cluster module meet the resource requirements of the second target service image. If they do, resources are allocated from the second computing resources to start the instance of the second target service image. For idle resources in the server cluster module, priority is given to determining whether they meet the resource requirements of the first target service image, so that the image of the local core service can preferentially call upon computing resources.

[0030] Therefore, by establishing a server cluster module and a service image repository module within the medical institution, and by using a service management module deployed within the medical institution to dynamically decide whether to call resources to start service instances based on the resource requirements of diagnostic tasks and server load, the resource utilization of each server is improved. Compared with the local private AI environment deployment mode that does not adopt cluster deployment in related technologies, this system can realize cross-scenario reuse of server resources in medical scenarios and ensure the priority operation of core services, effectively improving the efficiency of medical auxiliary diagnosis.

[0031] Furthermore, the system can leverage a second service mirror to expand its service capabilities. While meeting the core requirement of "data not leaving the hospital," it achieves the rational utilization of external service resources, further improving the resource utilization rate of locally deployed servers within medical institutions. Compared to the local private AI environment deployment model in related technologies, the resource scheduling system provided in this manual demonstrates significant advantages in resource utilization, response speed, scalability, and system stability, meeting the needs of modern medical institutions for intelligent, efficient, and secure diagnosis and treatment.

[0032] This specification provides a resource scheduling system, which is described in the embodiments below. Figure 1 As shown, the resource scheduling system 100 includes a server cluster module 110, a service image repository module 120, and a service management module 130, all deployed within the medical institution.

[0033] The service image repository module 120 includes a first service image 122 and a second service image 124 for medical auxiliary diagnosis; wherein, the first service image 122 is obtained from the server cluster module 110; and the second service image 124 is obtained from an external platform.

[0034] The service management module 130 is used to determine the first target service image in the first service image 122 according to the first diagnostic task request. If the first computing power resources in the server cluster module 110 meet the resource requirements of the first target service image, the module schedules resources from the first computing power resources to start the instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the first diagnostic task request based on the instance of the first target service image.

[0035] The service management module 130 is further configured to, if there is a second target service image corresponding to the first diagnostic task request in the second service image 124, and the second computing power resources in the server cluster module 110 meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start the instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

[0036] Among them, the server cluster module 110, the service image repository module 120, and the service management module 130 are local modules within the medical institution.

[0037] The server cluster module 110 includes local server computing power nodes used for medical auxiliary diagnosis within the medical institution. The first service image 122 is built based on the services in the local server computing power nodes, including images of core services used for medical auxiliary diagnosis within the medical institution, specifically AI service images.

[0038] The second service image 124 includes images of supplementary, auxiliary, or backup diagnostic services, specifically AI service images. External platforms can be third-party systems or service providers outside of medical institutions, such as cloud service providers, service marketplaces built by medical technology companies, or industry-shared algorithm platforms.

[0039] The first target service image corresponds to the first target service, and the second target service image corresponds to the second target service. Both the first and second target services have the ability to execute the diagnostic task corresponding to the first diagnostic task request. The resource requirements of the first and second target service images can be the same or different. The second target service image can be called a homogeneous service image of the first target service image (i.e., an image of the same type of application that processes the same computing task).

[0040] Specifically, in response to the service deployment needs of primary healthcare institutions under the "data not leaving the hospital" policy, this manual provides a resource scheduling system 100, including a server cluster module 110, a service image repository module 120, and a service management module 130 deployed within the healthcare institution.

[0041] Server cluster module 110 provides local computing resources for running various diagnostic service instances. Service image repository module 120 stores various service images used for medical diagnosis, mainly including two types of service images: first service image 122 and second service image 124. First service image 122 originates from the local server cluster and may include images of diagnostic services purchased by medical institutions from various vendors, images of diagnostic services developed by medical institutions themselves or used long-term, etc.; second service image 124 originates from external platforms (such as cloud service providers, service marketplaces, etc.) and may include images of diagnostic services used for supplementary diagnosis, auxiliary diagnosis, or backup.

[0042] Service Management Module 130 is the core control module, responsible for task scheduling, resource allocation, and service startup. It selects appropriate service images based on diagnostic task requests and determines whether to execute them based on resource availability. Service Management Module 130 can also be referred to as the service orchestration engine or service orchestration manager.

[0043] The core scheduling logic of the service management module 130 is as follows: upon receiving a first diagnostic task request, it determines the corresponding first target service image from the first service image 122 and firstly determines whether the idle resources (i.e., the first computing power resources) in the current server cluster module 110 meet the resource requirements of the first target service image. If not, the diagnostic task corresponding to the first diagnostic task request is placed in a queue to wait; if it meets the requirements, resources are scheduled from the first computing power resources to start the instance corresponding to the first target service image to execute the diagnostic task corresponding to the first diagnostic task request.

[0044] The service management module 130 also searches within the second service image 124 to determine if a matching second target service image can be found to execute the diagnostic task. If such a second target service image exists, it determines whether the idle resources (i.e., the second computing power resources) in the current server cluster module 110 meet the resource requirements of the second target service image. If so, resources are scheduled from the second computing power resources to start the instance corresponding to the second target service image to execute the diagnostic task corresponding to the first diagnostic task request.

[0045] In some embodiments, the second computing power resource is the same as the first computing power resource. That is, after receiving the first diagnostic task request, the service management module 130 can determine the corresponding first target service image from the first service image 122, and determine whether there is a corresponding second target service image in the second service image 124. If there is, the service management module 130 can determine whether the first computing power resource simultaneously meets the resource requirements of the first target service image and the second target service image.

[0046] If both conditions are met, the service management module 130 can allocate resources from the first computing power resources to launch the instances corresponding to the first target service image and the second target service image, respectively.

[0047] If the first computing power resource can only meet the resource requirements of the first target service image, or if the first computing power resource can meet the resource requirements of the first target service image and the second target service image, but cannot meet the resource requirements of both the first and second target service images at the same time, then the service management module 130 shall prioritize scheduling resources from the first computing power resource to start the instance corresponding to the first target service image.

[0048] In some cases, if the first computing power resource can only meet the resource requirements of the second target service image, the service management module 130 can schedule resources from the first computing power resource to start the instance corresponding to the second target service image.

[0049] In other embodiments, the second computing power resource is the idle resource in the server cluster module 110 after the instance corresponding to the first target service image is started. That is, after receiving the first diagnostic task request, the service management module 130 determines the corresponding first target service image from the first service image 122, and if the currently idle resources (i.e., the first computing power resource) of the server cluster module 110 meet the resource requirements of the first target service image, it calls the resources to start the instance corresponding to the first target service image.

[0050] After the instance corresponding to the first target service image is started, the idle resources in the server cluster module 110 are called the second computing resources. The service management module 130 determines whether the second target service image exists in the second service image 124. If it exists, it determines whether the second computing resources meet the resource requirements of the second target service image to determine whether to start the instance corresponding to the second target service image.

[0051] Furthermore, in some embodiments, a resource scheduling system 100 can be constructed based on an edge computing processing model. Specifically, the server cluster module 110 can be an edge node cluster deployed locally in the medical institution, and the external platform can be a cloud-based resource management platform, including an image repository. The second service image 124 can be obtained by downloading the image from the cloud-based image repository.

[0052] It is understandable that if the first computing power resource meets the resource requirements of the first target service image, it indicates that there are idle server nodes in server cluster module 110 that meet the resource requirements of the first target service image; similarly, if the second computing power resource meets the resource requirements of the second target service image, it indicates that there are idle server nodes in server cluster module 110 that meet the resource requirements of the second target service image. Starting the instance corresponding to the first target service image is equivalent to starting the first target service; starting the instance corresponding to the second target service image is equivalent to starting the second target service.

[0053] In the above embodiments, by establishing a server cluster module and a service image repository module within the medical institution, and by using a service management module deployed within the medical institution to dynamically determine whether to call resources to start service instances based on the resource requirements of diagnostic tasks and server load, the resource utilization of each server is improved, ensuring the timeliness of diagnosis. The resource scheduling system in this specification enables a clustered AI environment deployment mode. Compared to the local private AI environment deployment mode in related technologies, this system can achieve cross-scenario reuse of server resources and ensure the priority operation of core services, effectively improving the efficiency of medical auxiliary diagnosis.

[0054] Meanwhile, by coordinating and scheduling the server resources deployed locally within medical institutions, the system can allocate more server resources when faced with a sudden surge in diagnostic task requests, avoiding service crashes due to insufficient local resources and enhancing the stability of diagnostic services.

[0055] Furthermore, the system can leverage a second service mirror to expand its service capabilities, eliminating the need for medical institutions to invest heavily in expanding their local server clusters. This further improves the resource utilization of locally deployed servers within medical institutions and the efficiency of medical auxiliary diagnosis, while meeting the core requirement of "data not leaving the hospital." Under the premise of ensuring data security, it achieves the rational use of external resources, complying with regulatory requirements in the medical industry. Compared to the local private service environment deployment model in related technologies, the resource scheduling system provided in this specification demonstrates significant advantages in resource utilization, response speed, scalability, and system stability, meeting the needs of modern medical institutions for intelligent, efficient, and secure diagnosis and treatment.

[0056] In some implementations, the service management module 130 is specifically used to, when starting an instance corresponding to the first target service image, if a second target service image exists in the service image repository and the second computing power resources meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start an instance corresponding to the second target service image.

[0057] Specifically, after starting the instance corresponding to the first target service image, the service management module 130 can check whether there is a homogeneous service image in the second service image 124, that is, whether there is a second target service image corresponding to the first diagnostic task request. If it exists, it determines whether the idle resources (i.e., the second computing power resources) in the current server cluster module 110 meet the resource requirements of the second target service image. If they do, it schedules resources from them to start the instance corresponding to the second target service image.

[0058] Furthermore, the task calculation results generated by the instance of the first target service image and the task calculation results generated by the instance of the second target service image can both be returned to the client. Alternatively, the task calculation results generated by the instance of the second target service image can be used to perform quality control on the task calculation results generated by the instance of the first target service image, serving as a reference for evaluating the accuracy of the task calculation results generated by the instance of the first target service image.

[0059] For example, a user submits a request for a lung CT (Computed Tomography) image diagnosis. Upon receiving this request, the service management module identifies the lung screening model image as the first target service image from the first service images and checks whether there are suitable computing resources in the current server cluster module, i.e., whether the first computing power resource meets the resource requirements of the lung screening model image. If so, it calls upon the computing resources to start an instance of the model image.

[0060] After launching an instance of the lung screening model image, the service management module also checks whether a homogeneous service image (i.e., a second target service image) exists in the second service image. If a mirror image of a 3D nodule analysis model suitable for lung screening tasks exists in the second service image, it is used as the second target service image. The module then determines whether there are suitable idle computing resources in the server cluster module, i.e., whether the second computing power meets the resource requirements of the 3D nodule analysis model image. If so, the computing resources are used to launch an instance of the 3D nodule analysis model image.

[0061] The diagnostic results generated from the two instances can be presented to doctors simultaneously to assist in clinical decision-making. If the two results are consistent, it can enhance the diagnostic reliability of the lung screening model; if there is a significant difference, it can alert the doctor that the diagnostic results may be abnormal and prompt the doctor to conduct a manual review.

[0062] It should be noted that after starting the instance corresponding to the first target service image, the second computing resources can be other resources in the first computing resources besides the resources used to start the instance corresponding to the first target service image, or they can include the other resources and the resources released after the execution of other diagnostic tasks.

[0063] In the above implementation, when launching an instance corresponding to the first target service image, idle computing resources in the server cluster module are used to launch an instance corresponding to a homogeneous service image. This allows the system to dynamically introduce homogeneous services based on resource availability while ensuring the operation of local critical services, further improving the effective utilization of server computing resources and avoiding resource idleness and waste. Simultaneously, it enables multi-service collaborative diagnosis, ensuring timely diagnosis while improving the reliability of diagnostic results. The system possesses strong intelligence and scalability.

[0064] In some implementations, the service management module 130 is also used to release the resources used to start the instance corresponding to the second target service image if a second diagnostic task request is obtained when starting the instance corresponding to the second target service image.

[0065] Among them, the service images in the first service image 122 are called critical service images, and the service images in the second service image 124 are called non-critical images. The priority of critical service images in executing tasks (or scheduling resources) is higher than that of non-critical service images.

[0066] Specifically, if the instance corresponding to the second target service image has already been started, and the service management module 130 receives a second diagnostic task request, it needs to determine the target service image corresponding to the second diagnostic task request in the first service image 122, and check whether there are suitable computing resources in the server cluster module 110 to start the instance corresponding to the target service image. Since the service image in the first service image 122 has a higher priority for executing tasks or scheduling resources, the service management module 130 will release the resources used to start the instance corresponding to the second target service image, so as to allocate the computing resources to the new diagnostic task.

[0067] In some embodiments, the target service image corresponding to the second diagnostic task request in the first service image 122 is referred to as the third target service image. The service management module 130 can also be used to release the resources used to start the instance corresponding to the second target service image when the second diagnostic task request is obtained and the third computing power resources in the server cluster module 110 do not meet the resource requirements of the third target service image.

[0068] In the above implementation, by establishing a resource reclamation mechanism, the service management module dynamically adjusts service operation and resource allocation based on changes in task priority and resource demand. This avoids critical services waiting and failing to run in a timely manner due to non-critical services occupying resources. Thus, by dynamically releasing resources, an efficient resource scheduling and priority guarantee mechanism is achieved. This makes the entire resource scheduling system more intelligent and flexible, and able to adapt to the complex situation of multi-task concurrency in medical scenarios.

[0069] In some implementations, the server cluster module 110 is built based on the domain server nodes used for medical auxiliary diagnosis within the medical institution; the domain server nodes are the server nodes registered in the server registration module.

[0070] Among them, the intra-domain server node is the server node within the network security domain of the medical institution.

[0071] Specifically, the server registration module registers, verifies, and manages nodes that can join the server cluster. Based on the domain server nodes deployed within the medical institution, reviewed and registered by the server registration module, and used for medical auxiliary diagnosis, a cluster is built, resulting in server cluster module 110. Thus, all domain server nodes are registered in the server registration module for management, facilitating unified monitoring, maintenance, and scheduling.

[0072] For example, server nodes within the domain can download and install an authentication program to initiate authentication requests to the server registration module. Once the server registration module successfully authenticates the node, it is registered and included in the server cluster management scope. By building a node management hub based on the service registration and discovery mechanism, server node identity authentication and digital certificate binding registration can be achieved. Furthermore, a multi-dimensional resource monitoring system can be built to collect real-time node heartbeat status, GPU memory utilization (memory usage / total memory), computing unit load (CUDA (Compute Unified Device Architecture) core utilization), and operation logs, providing a visual operation and maintenance dashboard and an anomaly alarm triggering mechanism.

[0073] In some embodiments, the server registration module may be deployed within a medical institution.

[0074] In other embodiments, the server registration module may be deployed on an external platform.

[0075] It should be noted that the specific method for registering server nodes within the domain in the server registration module can be determined based on actual application requirements, and is not specifically limited in this specification. In this specification, the server registration module may also be referred to as the node registration module or the node registration center.

[0076] In some implementations, the second service image 124 is specifically obtained from an external platform via an encrypted transmission channel.

[0077] Specifically, when downloading the second service image 124 from an external platform, the image file is transmitted end-to-end through an encrypted transmission channel, protecting data privacy and integrity, ensuring that the service image content is not leaked, which helps to achieve secure, compliant, and reliable collaboration of the medical resource scheduling system 100, and solves the pain point of localization capability expansion.

[0078] For example, the encrypted transport channel can be a TLS (Transport Layer Security) / SSL (Secure Sockets Layer) encrypted channel.

[0079] Furthermore, the service image repository module can include an on-site service image repository and a private image repository. The on-site service image repository is used to store the first service image, and the private image repository is used to encrypt and store the second service image. Since the second service image can be obtained from the cloud image repository, the private image repository can also be called a private cloud synchronization repository.

[0080] In some cases, only medical institutions authorized by the external platform can obtain the second service image 124 from the external platform through an encrypted transmission channel, thereby achieving hierarchical control of access permissions.

[0081] For example, to address the AI ​​service deployment needs of primary-level hospitals under the "data stays within the hospital" policy, an intelligent solution based on edge computing is proposed. Specifically, by building an AI server cluster locally within the hospital, it supports both new computing facilities and the integration of existing heterogeneous AI servers to form a unified resource pool. The resource scheduling system provided in this specification, based on the edge computing processing model, constructs a data processing environment primarily for downlink data transmission, and innovatively designs an edge node resource management platform architecture for cloud deployment. (Reference) Figure 2 As shown, the platform comprises three core modules: a node registration center, a service orchestration manager (corresponding to service management module 130 in this manual), and a standardized service image library.

[0082] The service orchestration engine is deployed locally at the hospital level, serving as a local cloud, while the node registry and standardized service image library can be deployed on an external cloud. The domain-specific server nodes used to build the server cluster module are designated as edge nodes, and the server cluster module itself is the edge node cluster. Edge nodes register (or enroll) with the node registry, enabling service discovery and status monitoring. The platform can perform operational operations such as service start / stop and hot version updates on authorized nodes, while also supporting intelligent cross-node distribution and elastic deployment of AI service images. This solution ensures a closed-loop medical data flow throughout the hospital, achieving both centralized resource scheduling of AI services from multiple vendors and eliminating network transmission latency through localized computing. This allows primary healthcare institutions to achieve near-cloud-level AI service response efficiency, while simultaneously fostering a sustainable and iterative intelligent service ecosystem.

[0083] Edge nodes, as heterogeneous computing nodes deployed in primary healthcare institutions, are composed of high-performance computing servers equipped with GPUs. They support the automated deployment and hot updates of AI services through a resource management platform, forming a hospital-level distributed intelligent computing unit.

[0084] Continue to refer to Figure 2 As shown, the AI ​​service image repository deployed within the hospital (corresponding to service image repository module 120 in this specification) includes a hospital-side repository and a private cloud synchronization repository. The hospital-side repository integrates images of AI service modules already procured by the hospital. The private cloud synchronization repository stores standardized service images downloaded from the platform's standardized image repository with authorization from the resource management platform. It implements hierarchical access control, uses TLS / SSL encrypted channels to securely distribute image data, ensures image integrity through digital signature verification, and only responds to platform commands to activate and update encrypted images.

[0085] The node registration center is built upon a service registration and discovery mechanism to form a central hub for node management, enabling edge node identity authentication and digital certificate binding registration. Simultaneously, a multi-dimensional resource monitoring system is constructed to collect real-time data such as node heartbeat status, GPU memory utilization (memory usage / total memory), computing unit load (CUDA core utilization), and operation logs, providing a visual operation and maintenance dashboard and an anomaly alarm triggering mechanism.

[0086] The Service Orchestration Manager is responsible for managing the deployment of application services on edge nodes. Domain server nodes at the institute's end are registered and integrated into a server cluster. Servers no longer deploy a single, fixed service; server resources are dynamically scheduled by the Service Orchestration Manager. Based on the idle status of server resources, the Service Orchestration Manager dynamically deploys and uses AI service images from the AI ​​service image repository.

[0087] The aforementioned intelligent management architecture for edge node resources in medical data compliance scenarios achieves closed-loop management of data throughout the entire process by building a hospital-level AI service cluster. Under the premise of ensuring compliance with the "data does not leave the hospital" policy, it supports the localized deployment and collaborative computing of multimodal medical AI services.

[0088] It should be noted that the specific encryption protocol used in the encrypted transmission channel can be determined according to actual application requirements, and this manual does not impose specific limitations.

[0089] In some implementations, the server cluster module 110 includes a first domain server node and a second domain server node. The service management module 130 is used to upgrade the environment version of the first domain server node to the environment version of the second domain server node when the first domain server node and the second domain server node have the same hardware configuration, but the environment version of the first domain server node is lower than that of the second domain server node.

[0090] Hardware configuration refers to the physical resources of the server nodes, such as CPU model, number of GPUs, memory capacity, etc. Server nodes in the first domain are the same as those in the second domain in terms of hardware configuration.

[0091] The environment version refers to the software stack version required to run the service image, which may include, for example, the versions of the operating system kernel, containers, AI framework dependencies, etc.

[0092] In some cases, for any two intra-domain server nodes with identical hardware configurations (i.e., the first intra-domain server node and the second intra-domain server node), differences in software environments may cause certain applications or services to fail to run properly on all available server nodes. That is, intra-domain server nodes with higher environment versions can run services requiring high environment configurations and services requiring only low environment configurations, while intra-domain server nodes with lower environment versions can only run services requiring low environment configurations and are incompatible with services requiring high environment configurations. Therefore, the environment version of intra-domain server nodes with lower environment versions can be upgraded to support services requiring high environment configurations. This optimizes resource allocation and further improves server resource utilization.

[0093] Specifically, the service management module 130 can check the hardware and environment configurations of the domain server nodes in the server cluster module 110. For a first domain server node and a second domain server node with the same hardware configuration, if the service management module 130 detects that the environment version of the first domain server node is lower than that of the second domain server node, it will upgrade the environment version of the first domain server node to the environment version of the second domain server node. This ensures that the first domain server node can also support the services supported by the second domain server node after the upgrade, thus achieving cross-scenario reuse of the computing resources of the first domain server node.

[0094] Furthermore, both the server nodes in the first domain and the server nodes in the second domain are registered in the server registration module. After successful registration, each server node in the domain is managed. The service management module 130 checks the hardware configuration of each server node in the domain and installs the software environment version according to the hardware configuration.

[0095] For example, the service management module checks the hardware configuration of server nodes within each domain and installs the GPU driver and CUDA versions accordingly. Assume a hospital has purchased three servers from different AI vendors, denoted as Server A, Server B, and Server C. Server A and Server B have identical hardware configurations, while Server C has lower hardware configurations than Server A and Server B, requiring these three servers to be configured as a server cluster. After Server A is managed, when managing Server B, if the module finds that Server B's environment (GPU driver and CUDA) versions are lower than those of the managed Server A, it will upgrade Server B's environment. Generally, higher versions are backward compatible. In actual resource scheduling, both Server A and Server B can run each other's AI services.

[0096] When managing server C, if it is found that its hardware configuration is too low to use a higher version environment, then server C's environment version does not need to be upgraded. If a lower-configuration server D with the same hardware configuration as server C is added at this time, and server D's environment version is higher than server C's environment version, then server C's environment version needs to be upgraded to server D's environment version.

[0097] For example, in the foregoing Figure 2 Based on the architecture and deployment of the resource scheduling system shown, refer to Figure 3 As shown, the registration and management process for edge nodes (i.e., server nodes within the domain) in the resource scheduling system can mainly include the following steps:

[0098] (1) Download and install the authentication program on the edge node (server side). The edge node initiates a registration authentication request to the node registration center through the authentication program.

[0099] (2) The node registration center verifies whether the edge node has passed authentication. If it has, the edge node is registered in the node registration center, the edge node is added to the server cluster, and the edge node is managed. If not, the process ends.

[0100] (3) The service orchestration manager checks the hardware configuration of the edge nodes in the server cluster and verifies the environment version of each edge node to confirm whether an upgrade is needed. If so, the corresponding edge node is upgraded and the process proceeds to the next step; otherwise, the process proceeds directly to the next step.

[0101] (4) The standardized image repository in the cloud synchronizes the standardized service image to the private cloud synchronous repository deployed in the hospital through an encrypted transmission channel according to the authorization, so as to complete the service image downlink.

[0102] This enables a dynamic scheduling mechanism for heterogeneous computing power, integrates servers within the institute to build a service cluster, and uses a service management module to rationally utilize and orchestrate computing resources, thereby improving the utilization rate of computing resources.

[0103] In some implementations, the service management module 130 is further configured to compare the calculation results generated by the instance of the first target service image with the calculation results generated by the instance of the second target service image when the instances corresponding to the first target service image and the second target service image are started, and obtain a comparison result.

[0104] The comparison results can serve as a reference for evaluating the quality or accuracy of the computation results generated by the instance of the first target service image.

[0105] Specifically, after starting the instance corresponding to the first target service image and executing the diagnostic task corresponding to the first diagnostic task request, a calculation result is generated. After starting the instance corresponding to the second target service image and executing the same diagnostic task, another calculation result is generated. These two calculation results are compared to obtain a comparison result. The comparison result can serve as a reference for users to evaluate the quality or accuracy of the calculation result generated by the instance of the first target service image. If the comparison result indicates that the quality or accuracy of the calculation result is low, the user can be alerted that the result may be abnormal.

[0106] In some cases, the diagnostic task performed by the instance that initiates the second target service image in this embodiment can be called a quality control task, and the calculation result generated by the instance of the second target service image can be called the quality control task calculation result.

[0107] For example, this specification provides a task processing method based on a resource scheduling system. (See reference) Figure 4 As shown, the main steps of this method include:

[0108] (1) The user initiates a task processing request, which is sent to the service orchestration manager.

[0109] (2) The service orchestration manager checks whether there are suitable computing resources in the current server cluster. If there are, it starts the corresponding service image (i.e. the first target service image) to perform task computing and obtain the diagnostic task computing results; if not, it puts the computing task into the queue to wait.

[0110] (3) After the task calculation begins, the service orchestration manager checks whether there is a homogeneous service image (an image of the same type of application that handles the same disease, i.e., the second target service image) in the private cloud synchronization repository. If there are suitable idle computing resources in the server cluster at this time, the homogeneous service image can be started to perform task calculation in order to start the quality control task. If a new calculation task is added during the quality control task calculation, the service orchestration manager will interrupt the quality control task and allocate computing resources to the new task; if no new calculation task is added, the quality control task calculation result is obtained and used for quality reference of the diagnostic task calculation result obtained in the above steps.

[0111] (4) The service orchestration manager compares the calculation results of the diagnostic task with the calculation results of the quality control task, and returns the comparison results and the diagnostic task calculation results to the user, thus ending the process. The comparison results can remind the user whether there are any abnormalities in the diagnostic task calculation results, and if there are abnormalities, the user can be reminded to conduct a manual review.

[0112] This specification provides a resource scheduling method, applied to the service management module of the resource scheduling system described in any of the foregoing embodiments, referencing... Figure 5 As shown, the resource scheduling method includes the following steps.

[0113] S510, Obtain diagnostic task request.

[0114] S520. Based on the first diagnostic task request, determine the first target service image from the first service image in the service image repository module.

[0115] S530. If the first computing power resource in the server cluster module meets the resource requirements of the first target service image, resources are scheduled from the first computing power resource to start the instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image.

[0116] S540. If the second service image in the service image repository contains a second target service image corresponding to the first diagnostic task request, and the second computing power resources in the server cluster module meet the resource requirements of the second target service image, resources are scheduled from the second computing power resources to start the instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

[0117] It should be noted that for a detailed description of the resource scheduling method in the above embodiments, please refer to the detailed description of the service management module 130 in the resource scheduling system 100 in this specification, which will not be repeated here.

[0118] This specification also provides a resource scheduling device, applied to the service management module of the resource scheduling system described in any of the foregoing embodiments, referencing... Figure 6 As shown, the resource scheduling device 600 may include: a task request acquisition module 610, a service image determination module 620, a first instance startup module 630, and a second instance startup module 640.

[0119] The task request acquisition module 610 is used to acquire diagnostic task requests.

[0120] The service image determination module 620 is used to determine the first target service image from the first service image in the service image repository module according to the first diagnostic task request.

[0121] The first instance startup module 630 is used to schedule resources from the first computing resources to start the instance corresponding to the first target service image if the first computing resources in the server cluster module meet the resource requirements of the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image.

[0122] The second instance startup module 640 is used to schedule resources from the second computing resources to start an instance corresponding to the second target service image if there is a second target service image corresponding to the first diagnostic task request in the second service image in the service image repository, and the second computing resources in the server cluster module meet the resource requirements of the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

[0123] Specific limitations regarding the resource scheduling device can be found in the limitations of the resource scheduling method described above, and will not be repeated here. Each module in the aforementioned resource scheduling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0124] This specification also provides a computer device, which may include a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned resource scheduling method.

[0125] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned resource scheduling method.

[0126] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0127] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0128] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0129] Furthermore, 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0130] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0131] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A resource scheduling system, characterized in that, The system includes a server cluster module, a service image repository module, and a service management module deployed within the medical institution. The service image repository module includes a first service image and a second service image for medical auxiliary diagnosis; wherein, the first service image is obtained from the server cluster module; and the second service image is obtained from an external platform. The service management module is used to determine the first target service image in the first service image according to the first diagnostic task request. If the first computing power resources in the server cluster module meet the resource requirements of the first target service image, resources are scheduled from the first computing power resources to start the instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the first diagnostic task request based on the instance of the first target service image. The service management module is further configured to, if the second service image contains a second target service image corresponding to the first diagnostic task request, and the second computing power resources in the server cluster module meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start an instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

2. The system according to claim 1, characterized in that, Specifically, the service management module is used to, when starting an instance corresponding to the first target service image, if the second target service image exists in the service image repository and the second computing power resources meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start the instance corresponding to the second target service image.

3. The system according to claim 1, characterized in that, The service management module is also used to release the resources used to start the instance corresponding to the second target service image if a second diagnostic task request is obtained when starting the instance corresponding to the second target service image.

4. The system according to claim 1, characterized in that, The server cluster module is built based on the domain server nodes used for medical auxiliary diagnosis within the medical institution. The server nodes within the domain are the server nodes registered in the server registration module.

5. The system according to claim 1, characterized in that, The second service image is obtained from the external platform via an encrypted transmission channel.

6. The system according to claim 1, characterized in that, The server cluster module includes server nodes within a first domain and server nodes within a second domain. The service management module is used to upgrade the environment version of the first domain server node to the environment version of the second domain server node when the first domain server node and the second domain server node have the same hardware configuration, but the environment version of the first domain server node is lower than that of the second domain server node.

7. The system according to any one of claims 1 to 6, characterized in that, The service management module is further configured to, when starting instances corresponding to the first target service image and the second target service image, compare the calculation results generated by the instance of the first target service image with the calculation results generated by the instance of the second target service image to obtain a comparison result.

8. A resource scheduling method, characterized in that, The method, applied to a service management module in a resource scheduling system according to any one of claims 1 to 7, comprises: Obtain a diagnostic task request; Based on the first diagnostic task request, determine the first target service image from the first service image in the service image repository module; If the first computing power resource in the server cluster module meets the resource requirements of the first target service image, resources are scheduled from the first computing power resource to start the instance corresponding to the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image. If the second service image in the service image repository contains the second target service image corresponding to the first diagnostic task request, and the second computing power resources in the server cluster module meet the resource requirements of the second target service image, resources are scheduled from the second computing power resources to start the instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the resource requirements of the first target service image have a higher priority than the resource requirements of the second target service image.

9. A resource scheduling device, characterized in that, A service management module applied in the resource scheduling system according to any one of claims 1 to 7, the device comprising: The task request acquisition module is used to acquire diagnostic task requests; The service image determination module is used to determine the first target service image from the first service images in the service image repository module according to the first diagnostic task request. The first instance startup module is used to schedule resources from the first computing resources to start the instance corresponding to the first target service image if the first computing resources in the server cluster module meet the resource requirements of the first target service image, so as to execute the diagnostic task corresponding to the diagnostic task request based on the instance of the first target service image. The second instance startup module is configured to, if the second service image in the service image repository contains a second target service image corresponding to the first diagnostic task request, and the second computing power resources in the server cluster module meet the resource requirements of the second target service image, schedule resources from the second computing power resources to start an instance corresponding to the second target service image, so as to execute the diagnostic task based on the instance of the second target service image; wherein, the priority of the resource requirements of the first target service image is higher than the priority of the resource requirements of the second target service image.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method described in claim 8.

Images