Virtualized resource management platform

Abstract

The invention relates to the technical field of resource management systems, in particular to a virtualized resource management platform which comprises an infrastructure layer, and the infrastructure layer converts scattered hardware equipment into a resource pool managed in a unified mode and is composed of a physical server cluster, network equipment and storage equipment. According to the virtualization resource management platform, accurate mapping from physical resources to business services is achieved through perception, decision making, execution and optimization of full-link automation, each link from bottom layer hardware perception to top layer business guarantee of the virtualization resource management platform comprises precise automatic control and is also provided with a flexible manual intervention interface, and the automation degree of the whole link is improved. Finally, double improvement of the resource utilization rate and the service quality is realized, the operation and maintenance cost is reduced, the resource delivery efficiency is improved, rapid business innovation is supported, the level-of-guarantee three-level compliance requirement is met, and the key operation can support audit tracing.

Description

Technical Field

[0001] The present invention relates to the technical field of resource management systems, and in particular to a virtualized resource management platform. Background Art

[0002] A virtualization resource management platform is a software system used to manage and monitor virtualized resources. It uses abstraction technology to convert physical hardware resources such as CPU, memory, and storage into dynamically allocated virtual machine instances, thereby achieving on-demand allocation, flexible scheduling, and efficient utilization of resources.

[0003] Current data center resource management technologies coexist with hybrid architectures and diverse virtualization technologies, lacking a unified management solution. Operation and maintenance costs are high, and the traditional static allocation model leads to insufficient resource utilization. Scheduling decisions rely heavily on manual experience and are unable to automatically respond to load fluctuations in real time.

[0004] In order to solve the above problems, we make improvements and propose a virtualized resource management platform. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: The present invention provides a virtualized resource management platform, comprising an infrastructure layer, which converts dispersed hardware devices into a resource pool under unified management and is composed of a physical server cluster, network devices, and storage devices; The infrastructure layer is subsequently connected to the data perception layer, which consists of a resource discovery module, a status monitoring module, and an abnormality warning system. The resource discovery module continuously scans and discovers devices in the resource pool, and the status monitoring module continuously monitors the real-time status of the physical server. When the status is abnormal, the abnormality warning system is activated. The data perception layer subsequently connects to the intelligent decision-making layer, which sequentially performs demand analysis, optimizes decisions, and generates scheduling strategies. After receiving a user request, the demand analysis engine first performs semantic parsing, using NLP to process and analyze the user's demand text to identify keywords. It then analyzes historical data. If a similar task experienced a computing peak at night last week, the system automatically reserves 20% redundant resources and uses the improved NSGA-III algorithm to find the optimal solution among all deployment options. The intelligent decision-making layer subsequently connects to the execution control layer, which is mainly composed of a dynamic scheduling center. After receiving instructions, it will perform host machine pre-checks and configure the database; The execution control layer subsequently connects to the evaluation and optimization layer, which will perform resource audits, energy efficiency analysis, and SLA evaluations in turn to verify system effectiveness and continuously improve it.

[0006] As a preferred technical solution of the present invention, when a new physical server is connected to the data center, it will first automatically send a network request through the out-of-band management interface such as iLO and iDRAC. The resource discovery module will scan the pre-set IP address segment and immediately start the authentication process after capturing the device's "heartbeat signal". The device supports the Redfish protocol, and the system obtains its firmware information, hardware configuration and other key data through the HTTPS encrypted channel.

[0007] As a preferred technical solution of the present invention, when the resource discovery module discovers a server, the module will obtain its detailed configuration through the Redfish API, from the CPU model to the memory slot usage, and even the SMART health status of the hard disk. After cleaning, this data will generate a standardized JSON description file and store it in the database.

[0008] As a preferred technical solution of the present invention, the status monitoring module collects the temperature and power consumption data of the physical server once a second through the IPMI interface; for virtual machines, the vCPU utilization and memory balloon status are monitored through QEMU's LibvirtAPI; the monitoring of the container environment relies on the cAdvisor tool to accurately count the CPU cycles and memory page usage of each container. When the CPU utilization of any node exceeds 80% for 5 consecutive minutes, the system will trigger a level 3 alarm, first notifying the administrator by email. If there is no response within 10 minutes, the emergency isolation process will be automatically started.

[0009] As a preferred technical solution of the present invention, the intelligent decision-making layer will give priority to nodes equipped with liquid cooling systems, ensure that there are no more than two switch hops between servers, and avoid deploying the primary and backup nodes in the same cabinet. When encountering sudden high-priority tasks, it will initiate "preemptive scheduling" to gently evict a virtual machine running a batch job to other nodes, freeing up resources for urgent real-time analysis tasks.

[0010] As a preferred technical solution of the present invention, after receiving the instruction, the dynamic scheduling center of the execution control layer will first check whether the NVIDIAvGPU license of the target host machine is sufficient, verify the remaining number of VFs of the SR-IOV network card, and ensure that the Ceph storage pool has sufficient IOPS capacity; then, through the Ansible execution engine, create an XML format virtual machine definition file on KVM, specify NUMA affinity, and bind the A100 graphics card to the virtual machine through PCIe transparent transmission, and finally configure the GID global identifier and QoS policy of the RDMA network card.

[0011] As a preferred technical solution of the present invention, if it is detected that a host machine has insufficient memory during the creation process, the scheduling center will immediately start the "balloon driver" to compress the memory usage of other low-priority virtual machines to ensure the smooth completion of the current task.

[0012] As a preferred technical solution of the present invention, after each task is completed, the evaluation and optimization layer will initiate an SLA compliance audit, i.e., comparing the actual response time with the promised value; an energy efficiency assessment, i.e., calculating the kilowatt-hours consumed for each TFLOPS operation completed; and a resource fragmentation analysis, i.e., counting the small blocks of unavailable resources generated by the allocation strategy.

[0013] The beneficial effects of the present invention are: a virtualized resource management platform, which realizes the precise mapping from physical resources to business services through full-link automation of perception, decision-making, execution and optimization. The virtualized resource management platform includes both precise automated control and flexible manual intervention interfaces in every link from bottom-level hardware perception to top-level business assurance, ultimately achieving a dual improvement in resource utilization and service quality, reducing operation and maintenance costs, improving resource delivery efficiency, supporting rapid business innovation, meeting the Level 3 security compliance requirements, and key operations can support audit traceability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a system flow diagram of a virtualized resource management platform of the present invention; Figure 2 This is a schematic diagram of a system subdivision process of a virtualized resource management platform of the present invention; Figure 3 This is a schematic diagram of a system subdivision process of a virtualized resource management platform of the present invention; Figure 4 This is a schematic diagram of a system subdivision process of a virtualized resource management platform of the present invention; Figure 5 This is a schematic diagram of a system subdivision process of a virtualized resource management platform of the present invention; DETAILED DESCRIPTION

[0015] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0016] Example: Figure 1-Figure 5As shown, a virtualized resource management platform includes an infrastructure layer, which converts scattered hardware devices into a unified managed resource pool, consisting of a physical server cluster, network devices, and storage devices; The infrastructure layer is subsequently connected to the data perception layer, which consists of a resource discovery module, a status monitoring module, and an abnormality warning system. The resource discovery module continuously scans and discovers devices in the resource pool, and the status monitoring module continuously monitors the real-time status of the physical server. When the status is abnormal, the abnormality warning system is activated. The data perception layer subsequently connects to the intelligent decision-making layer, which sequentially performs demand analysis, optimizes decisions, and generates scheduling strategies. After receiving a user request, the demand analysis engine first performs semantic parsing, using NLP to process and analyze the user's demand text to identify keywords. It then analyzes historical data. If a similar task experienced a computing peak at night last week, the system automatically reserves 20% redundant resources and uses the improved NSGA-III algorithm to find the optimal solution among all deployment options. The intelligent decision-making layer subsequently connects to the execution control layer, which is mainly composed of a dynamic scheduling center. After receiving instructions, it will perform host machine pre-checks and configure the database; The execution control layer subsequently connects to the evaluation and optimization layer, which will perform resource audits, energy efficiency analysis, and SLA evaluations in turn to verify system effectiveness and continuously improve it.

[0017] When a new physical server is connected to the data center, it will first automatically send a network request through the out-of-band management interface such as iLO and iDRAC. The resource discovery module will scan the pre-set IP address segment and immediately start the authentication process after capturing the device's "heartbeat signal". The device supports the Redfish protocol, and the system obtains its firmware information, hardware configuration and other key data through the HTTPS encrypted channel.

[0018] When the resource discovery module discovers a server, it obtains its detailed configuration through the Redfish API, ranging from the CPU model to the memory slot usage and even the SMART health status of the hard disk. After cleaning this data, a standardized JSON description file is generated and stored in the database.

[0019] The status monitoring module collects the temperature and power consumption data of the physical server once a second through the IPMI interface; for virtual machines, it monitors the vCPU utilization and memory balloon status through QEMU's LibvirtAPI; the monitoring of the container environment relies on the cAdvisor tool to accurately count the CPU cycles and memory page usage of each container. When the CPU utilization of any node exceeds 80% for 5 consecutive minutes, the system will trigger a level 3 alarm, first notifying the administrator by email. If there is no response within 10 minutes, the emergency isolation process will be automatically initiated.

[0020] The intelligent decision-making layer prioritizes nodes equipped with liquid cooling systems, ensuring that servers span no more than two switch hops and avoiding deploying active and standby nodes in the same cabinet. When encountering sudden high-priority tasks, it initiates "preemptive scheduling" to gently evict a virtual machine running a batch job to other nodes, freeing up resources for urgent real-time analysis tasks.

[0021] After receiving the instruction, the dynamic scheduling center of the execution control layer will first check whether the NVIDIA vGPU license of the target host is sufficient, verify the remaining number of VFs of the SR-IOV network card, and ensure that the Ceph storage pool has sufficient IOPS capacity; then, it will create an XML format virtual machine definition file on KVM through the Ansible execution engine, specify NUMA affinity, and bind the A100 graphics card to the virtual machine through PCIe transparent transmission. Finally, it will configure the GID global identifier and QoS policy of the RDMA network card.

[0022] If it is detected that a host machine has insufficient memory during the creation process, the scheduling center will immediately start the "balloon driver" to compress the memory usage of other low-priority virtual machines to ensure the smooth completion of the current task.

[0023] After each task is completed, the evaluation and optimization layer will initiate an SLA compliance audit, which compares the actual response time with the promised value; an energy efficiency assessment, which calculates the kilowatt-hours consumed for each TFLOPS of computing completed; and a resource fragmentation analysis, which counts the small blocks of unavailable resources generated by the allocation strategy.

[0024] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A virtualized resource management platform, comprising an infrastructure layer, characterized in that: The infrastructure layer transforms dispersed hardware devices into a centrally managed resource pool, consisting of physical server clusters, network devices, and storage devices. The infrastructure layer is subsequently connected to the data perception layer, which consists of a resource discovery module, a status monitoring module, and an abnormality warning system. The resource discovery module continuously scans and discovers devices in the resource pool, and the status monitoring module continuously monitors the real-time status of the physical server. When the status is abnormal, the abnormality warning system is activated. The data perception layer subsequently connects to the intelligent decision-making layer, which sequentially performs demand analysis, optimizes decisions, and generates scheduling strategies. After receiving a user request, the demand analysis engine first performs semantic parsing, using NLP to process and analyze the user's demand text to identify keywords. It then analyzes historical data. If a similar task experienced a computing peak at night last week, the system automatically reserves 20% redundant resources and uses the improved NSGA-III algorithm to find the optimal solution among all deployment options. The intelligent decision-making layer subsequently connects to the execution control layer, which is mainly composed of a dynamic scheduling center. After receiving instructions, it will perform host machine pre-checks and configure the database; The execution control layer subsequently connects to the evaluation and optimization layer, which will perform resource audits, energy efficiency analysis, and SLA evaluations in turn to verify system effectiveness and continuously improve it.

2. A virtualized resource management platform according to claim 1, characterized in that: When a new physical server is connected to the data center, it will first automatically send a network request through the out-of-band management interface such as iLO and iDRAC. The resource discovery module will scan the pre-set IP address segment and immediately start the authentication process after capturing the device's "heartbeat signal". The device supports the Redfish protocol, and the system obtains its firmware information, hardware configuration and other key data through the HTTPS encrypted channel.

3. A virtualized resource management platform according to claim 1, characterized in that: When the resource discovery module discovers a server, it obtains its detailed configuration through the Redfish API, ranging from the CPU model to the memory slot usage and even the SMART health status of the hard disk. After cleaning this data, a standardized JSON description file is generated and stored in the database.

4. A virtualized resource management platform according to claim 3, characterized in that: The status monitoring module collects the temperature and power consumption data of the physical server once a second through the IPMI interface; for virtual machines, it monitors the vCPU utilization and memory balloon status through QEMU's LibvirtAPI; the monitoring of the container environment relies on the cAdvisor tool to accurately count the CPU cycles and memory page usage of each container. When the CPU utilization of any node exceeds 80% for 5 consecutive minutes, the system will trigger a level 3 alarm, first notifying the administrator by email. If there is no response within 10 minutes, the emergency isolation process will be automatically initiated.

5. A virtualized resource management platform according to claim 1, characterized in that: The intelligent decision-making layer prioritizes nodes equipped with liquid cooling systems, ensuring that servers span no more than two switch hops and avoiding deploying active and standby nodes in the same cabinet. When encountering sudden high-priority tasks, it initiates "preemptive scheduling" to gently evict a virtual machine running a batch job to other nodes, freeing up resources for urgent real-time analysis tasks.

6. A virtualized resource management platform according to claim 1, characterized in that: After receiving the instruction, the dynamic scheduling center of the execution control layer will first check whether the NVIDIA vGPU license of the target host is sufficient, verify the remaining number of VFs of the SR-IOV network card, and ensure that the Ceph storage pool has sufficient IOPS capacity; then, it will create an XML format virtual machine definition file on KVM through the Ansible execution engine, specify NUMA affinity, and bind the A100 graphics card to the virtual machine through PCIe transparent transmission. Finally, it will configure the GID global identifier and QoS policy of the RDMA network card.

7. A virtualized resource management platform according to claim 6, characterized in that: If it is detected that a host machine has insufficient memory during the creation process, the scheduling center will immediately start the "balloon driver" to compress the memory usage of other low-priority virtual machines to ensure the smooth completion of the current task.

8. A virtualized resource management platform according to claim 1, characterized in that: After each task is completed, the evaluation and optimization layer will initiate an SLA compliance audit, which compares the actual response time with the promised value; an energy efficiency assessment, which calculates the kilowatt-hours consumed for each TFLOPS of computing completed; and a resource fragmentation analysis, which counts the small blocks of unavailable resources generated by the allocation strategy.

Images