Virtual machine thermal migration performance optimization method and device, electronic equipment and storage medium
By acquiring multidimensional migration correlation information of virtual machines and optimizing migration parameters using a migration performance prediction model, the problems of high latency and service interruption in traditional virtual machine hot migration methods are solved, achieving higher migration efficiency and success rate.
Patent Information
- Application Number
- CN202511590952.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-02-17
AI Technical Summary
Traditional virtual machine hot migration methods fail to combine the virtual machine, host machine, and network status for performance prediction, resulting in fixed migration parameters, which can easily lead to high latency or service interruption. Furthermore, the selection of the target host machine depends on local information, which may cause target machine overload or link congestion. This makes it difficult to adapt to dynamic, distributed resource environments, affecting business continuity and resource utilization.
By acquiring multidimensional migration correlation information of the target virtual machine, and using a pre-trained migration performance prediction model, the optimal combination of migration parameters is determined, and the migration strategy is optimized, including memory data transfer strategy, CPU scheduling priority, storage migration mode and bandwidth allocation ratio, etc., and dynamic resource adaptation is performed by combining a distributed awareness architecture and a compute-first network router.
It improves the efficiency and success rate of virtual machine hot migration, avoids blind trial and error during migration, ensures that the migration process meets actual business needs, and improves resource utilization and business continuity.
Smart Images
Figure CN121541972A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and in particular to a virtual machine live migration performance optimization method and device, electronic equipment and storage medium. BACKGROUND
[0002] With the development of cloud computing and computing power network, virtual machine live migration has become a core technology for realizing load balancing, resource scheduling and fault recovery. The traditional virtual machine live migration method has the following technical defects: first, it does not combine the virtual machine, host and network state to make performance prediction, and fixed migration parameters are easy to cause high latency or long service interruption; second, the target host selection relies on local information, and does not dynamically adapt to the whole network resources, which may cause target machine overload or link congestion. These problems make it difficult for the traditional live migration method to adapt to the dynamic and distributed resource environment in the computing power network, affecting business continuity and resource utilization. SUMMARY
[0003] The present application provides a virtual machine live migration performance optimization method, device, electronic equipment and storage medium to improve migration efficiency and success rate.
[0004] In one aspect of the present application, a virtual machine live migration performance optimization method is provided, comprising:
[0005] In response to a virtual machine live migration task, determining a target virtual machine of the virtual machine live migration task, and obtaining multi-dimensional migration associated information of the target virtual machine;
[0006] According to the virtual machine live migration task, at least one group of migration parameter combinations is determined, and the performance prediction result of each group of migration parameter combinations is obtained based on a pre-trained migration performance prediction model and the multi-dimensional migration associated information;
[0007] According to a preset optimization target and each performance prediction result, an optimal migration parameter is determined in each migration parameter combination;
[0008] According to the optimal migration parameter, the target virtual machine is live migrated.
[0009] In one aspect of the present application, a virtual machine live migration performance optimization device is provided, comprising:
[0010] An information collection module is configured to, in response to a virtual machine live migration task, determine a target virtual machine of the virtual machine live migration task, and obtain multi-dimensional migration associated information of the target virtual machine;
[0011] A performance prediction module is configured to, according to the virtual machine live migration task, determine at least one group of migration parameter combinations, and obtain the performance prediction result of each group of migration parameter combinations based on a pre-trained migration performance prediction model and the multi-dimensional migration associated information.
[0012] determining optimal migration parameters in each of the migration parameter combinations according to a preset optimization target and each of the performance prediction results;
[0013] performing hot migration of the target virtual machine according to the optimal migration parameters.
[0014] In another aspect of the embodiments of the present application, an electronic device is provided, comprising:
[0015] at least one processor; and
[0016] a memory in communication with the at least one processor;
[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the virtual machine hot migration performance optimization method of any of the embodiments of the present application.
[0018] In another aspect of the embodiments of the present application, a computer readable storage medium is provided, comprising: computer instructions for enabling a processor to perform the virtual machine hot migration performance optimization method of any of the embodiments of the present application when executed by the processor.
[0019] The present application determines a target virtual machine of a virtual machine hot migration task in response to the virtual machine hot migration task, and obtains multi-dimensional migration correlation information of the target virtual machine, determines at least one group of migration parameter combinations according to the virtual machine hot migration task, obtains performance prediction results of each of the migration parameter combinations based on a pre-trained migration performance prediction model and the multi-dimensional migration correlation information, determines optimal migration parameters in each of the migration parameter combinations according to a preset optimization target and each of the performance prediction results, and performs hot migration of the target virtual machine according to the optimal migration parameters. The present application utilizes the migration performance prediction model to predict the performance of different migration parameter combinations in advance, avoids blind trial and error, saves migration time, and improves migration efficiency. The optimal migration parameters are selected according to the preset optimization target, which can make the hot migration more suitable for actual business requirements, and the hot migration according to the optimal migration parameters can further improve migration efficiency and success rate.
[0020] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS
[0021] In order to make the technical solution in the embodiments of the present application clearer, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.
[0022] Figure 1 is a virtual machine live migration performance optimization method flowchart provided according to the first embodiment of the present application;
[0023] Figure 2 is another virtual machine live migration performance optimization method flowchart provided according to the second embodiment of the present application;
[0024] Figure 3 is a virtual machine live migration performance optimization device structure diagram provided according to the fourth embodiment of the present application;
[0025] Figure 4 is an electronic device block diagram for executing the virtual machine live migration performance optimization method provided according to the fifth embodiment of the present application. DETAILED DESCRIPTION
[0026] In order to make the technical solution in the embodiments of the present application clearer, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.
[0027] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or apparatus that includes a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or apparatus.
[0028] Figure 1A flowchart of a virtual machine live migration performance optimization method is provided for the embodiments of the present application. The embodiments of the present application can be applied to scenarios for improving semantic connection integrity in a communication graph structure. The method can be executed by a virtual machine live migration performance optimization device, which can be implemented in the form of hardware and / or software. The virtual machine live migration performance optimization device can be configured in a semiconductor test dedicated server or a distributed computing cluster composed of multiple servers, and the like. As shown in FIG. 6, the method comprises the following steps. Figure 1
[0029] S110, in response to a virtual machine live migration task, determining a target virtual machine of the virtual machine live migration task, and obtaining multi-dimensional migration association information of the target virtual machine.
[0030] The virtual machine live migration task refers to a task of migrating the target virtual machine from a source host to a target host. By executing the virtual machine live migration task, flexible deployment, network resource optimization, load balancing, and / or fault recovery of the target virtual machine can be achieved. For example, the triggering scenarios of the virtual machine live migration task can include user manual triggering, system load balancing triggering, or fault early warning triggering, and the like.
[0031] The target virtual machine refers to a virtual machine that needs to be migrated from a source host to a target host in the virtual machine live migration task. As an operation object of the virtual machine live migration task, the migration process of the target virtual machine can be constrained by the running state of the target virtual machine, the resource situation of the host, and / or the network link situation between the hosts.
[0032] The multi-dimensional migration association information can be understood as a set containing multiple dimensions of information, which is used to reflect the migration environment of the target virtual machine. For example, the multi-dimensional migration association information at least includes virtual machine running state information, host resource information, and network-storage link information. The virtual machine running state information reflects the running state of the target virtual machine, and the virtual machine running state information includes at least one of the following: memory dirty page rate, central processing unit (CPU) load, and disk input / output operations per second. The host resource information reflects the resource situation of the source host and / or the target host, and includes at least one of the following: remaining central processing unit core number, available memory capacity, and storage throughput. The network-storage link information reflects the network link situation between the hosts, and includes at least one of the following: link bandwidth, transmission delay, packet loss rate, and storage image synchronization rate.
[0033] Specifically, when the virtual machine live migration performance optimization apparatus receives a virtual machine live migration task, the target virtual machine to be migrated in the virtual machine live migration task can be determined first. The running state information of the virtual machine can be collected through a built-in agent component of the host, such as Libvirt or Hyper-V. The host resource information and network-storage link information can be collected by calling a computing priority network router. The information set including the running state information of the target virtual machine, the host resource information, and the network-storage link information can be used as multi-dimensional migration correlation information.
[0034] In an example, in the embodiment of the present application, the running state information of the virtual machine and the resource information of the host can also be captured in real time through a data collection function module deployed by the edge computing node. The network-storage link information between the hosts can also be collected through a software-defined network controller. The embodiment of the present application does not limit the way of obtaining the multi-dimensional migration correlation information of the target virtual machine.
[0035] In S120, at least one set of migration parameter combinations is determined according to the virtual machine live migration task, and the performance prediction result of each set of migration parameter combinations is obtained based on the pre-trained migration performance prediction model and the multi-dimensional migration correlation information.
[0036] The migration parameter combination can be understood as a set of parameters describing the migration performance, which is used to assist the migration performance prediction model to make more accurate performance prediction results. For example, the migration parameter combination can include specific strategies for memory data transmission, such as the number of pre-copy iterations, the dirty page threshold, the memory compression algorithm, the CPU scheduling priority, the storage migration mode, and / or the bandwidth allocation ratio.
[0037] The migration performance prediction model can be understood as a mathematical model for predicting the migration performance based on the multi-dimensional migration correlation information. The migration performance prediction model can be trained by using a long short-term memory-random forest fusion algorithm or a convolutional neural network-support vector regression fusion algorithm. In the training process of the migration performance prediction model, the migration performance prediction model learns the relationship between the historical multi-dimensional migration correlation information and the actual migration performance, so as to predict the migration performance under different migration parameter combinations.
[0038] The performance prediction result refers to the prediction data about the migration performance of the target virtual machine output by the migration performance prediction model, which is a prediction of the migration performance under a specific migration parameter combination. For example, the performance prediction result can include the total migration time delay, the service interruption time, and the resource occupancy rate. Further, in the embodiment of the present application, a screening standard for a migration parameter combination can be pre-set, and the optimal migration parameter combination can be screened out from the performance prediction result. The target virtual machine is migrated based on the optimal migration parameter combination.
[0039] Specifically, when a virtual machine hot migration task is received, a series of migration parameters can be configured for the task, including specific strategies for memory data transfer, CPU scheduling priorities, storage migration modes, and / or bandwidth allocation ratios. It can be understood that the specific strategies for memory data transfer include various sub-strategies, CPU scheduling priorities are divided into different levels, storage migration modes can include full image transfer, incremental block synchronization, and storage tiered migration, and the bandwidth allocation ratio can be flexibly set to specific values, such as 20% or 30%. By combining the specific options of the above parameters, at least one set of migration parameter combinations can be formed. These migration parameter combinations and multi-dimensional migration correlation information can be input into a pre-trained migration performance prediction model. The pre-trained migration performance prediction model can predict migration performance under different migration parameter combinations based on the learned multi-dimensional migration correlation information and the relationship between actual migration performance, and output the performance prediction results corresponding to different migration parameter combinations.
[0040] S130. Determine the optimal migration parameters within each combination of migration parameters based on the preset optimization objectives and performance prediction results.
[0041] Among them, preset optimization goals refer to the criteria set in advance according to business needs, which are used to select the optimal migration parameters and ensure that the migration results of the target virtual machine meet business requirements. For example, preset optimization goals may include goals such as low latency priority or low resource consumption priority.
[0042] The optimal migration parameters can be understood as a set of migration parameters that best meet business requirements, used to determine the specific execution strategy for hot migration of the target virtual machine. Furthermore, different preset optimization goals result in different optimal migration parameters being selected from various performance prediction results. For example, if low latency is prioritized as the preset optimization goal, then parameter combinations with total migration latency < 500ms and service interruption time < 50ms can be prioritized in the performance prediction results; if low resource consumption is prioritized as the preset optimization goal, then parameter combinations with host CPU utilization < 60% and link bandwidth utilization < 70% can be prioritized in the performance prediction results.
[0043] Specifically, the system obtains preset optimization goals based on business needs. Based on these goals, it evaluates each performance prediction result and selects the migration parameter from the combination of migration parameters that best matches the preset optimization goal as the optimal migration parameter. For example...
[0044] Preset optimization objectives may include prioritizing low latency or low resource consumption. If the preset optimization objective is prioritizing low latency, then in the performance prediction results, the corresponding migration parameter combinations that meet the latency requirements for the total migration latency are selected, and the migration parameters in these migration parameter combinations are taken as the optimal migration parameters. If the preset optimization objective is prioritizing low resource consumption, then the parameter combinations that meet the resource consumption requirements are selected as the optimal migration parameters.
[0045] S140. Perform hot migration of the target virtual machine according to the optimal migration parameters.
[0046] Specifically, the hot migration process of the target virtual machine is initiated according to the optimal migration parameters.
[0047] In this embodiment of the invention, in response to a virtual machine hot migration task, the target virtual machine for the hot migration task is determined, and multi-dimensional migration association information of the target virtual machine is obtained. At least one set of migration parameter combinations is determined based on the virtual machine hot migration task. Performance prediction results for each set of migration parameter combinations are obtained based on a pre-trained migration performance prediction model and the multi-dimensional migration association information. Optimal migration parameters are determined within each migration parameter combination based on a preset optimization objective and each performance prediction result. Hot migration of the target virtual machine is then performed using the optimal migration parameters. This embodiment of the invention utilizes a migration performance prediction model to predict the performance of different migration parameter combinations in advance, avoiding blind trial and error, saving migration time, and improving migration efficiency. Selecting the optimal migration parameters based on a preset optimization objective makes hot migration more aligned with actual business needs. Performing hot migration according to the optimal migration parameters further improves migration efficiency and success rate.
[0048] Optionally, this embodiment of the invention also optimizes a virtual machine hot migration performance optimization method. Specifically, it supplements the specific steps of the training process of the migration performance prediction model, including: collecting historical migration data of different virtual machines; training a long short-term memory-random forest fusion algorithm model according to the historical migration data; and using the long short-term memory-random forest fusion algorithm model as the migration performance prediction model when the prediction error of the long short-term memory-random forest fusion algorithm model is determined to be lower than a threshold.
[0049] The Long Short-Term Memory-Random Forest fusion algorithm model is a hybrid machine learning model that combines Long Short-Term Memory networks and random forest algorithms to improve prediction accuracy and robustness. Long Short-Term Memory networks can be used to capture temporal features such as dirty page rate and link bandwidth, while random forests are used to fit the nonlinear relationship between transfer parameters and performance metrics. Five-fold cross-validation can also be used during model training to reduce prediction errors.
[0050] Historical migration data can be understood as migration records of virtual machines during previous hot migrations. It is used to train the Long Short-Term Memory-Random Forest fusion algorithm model. For example, historical migration data includes multi-dimensional migration association information samples, migration parameter combination samples, and actual performance indicators of migration parameter combination samples.
[0051] Specifically, historical migration data from previous hot migrations of different virtual machines is collected. This data may include multi-dimensional migration association information samples, migration parameter combination samples, and actual performance indicators. Using the historical migration data, a long short-term memory-random forest fusion algorithm model is trained. When the prediction error of the model is lower than a preset threshold, the model is determined as the final migration performance prediction model.
[0052] Optionally, this embodiment of the invention also optimizes a virtual machine hot migration performance optimization method. Specifically, it adds specific steps for monitoring the target virtual machine during the hot migration process, including:
[0053] B1. Monitor the hot migration process indicators of the target virtual machine.
[0054] In this context, hot migration process metrics can be understood as a series of indicators monitored in real time during the hot migration of the target virtual machine. These metrics reflect the real-time status of the target virtual machine during the hot migration process and are used to assess whether the hot migration process is normal. For example, hot migration process metrics may include: dirty page rate, link bandwidth utilization, and target host CPU utilization.
[0055] Specifically, during the migration of the target virtual machine, the migration process can be monitored in real time to obtain indicators of the hot migration process of the target virtual machine, such as: dirty page rate, link bandwidth utilization and / or CPU utilization of the target host machine.
[0056] B2. Adjust the optimal migration parameters based on the thermal migration process indicators and preset indicator thresholds.
[0057] The preset indicator threshold can be understood as a critical value used to classify whether the indicators in the hot migration process are abnormal. It is used to measure whether the indicators in the current hot migration process are within an acceptable range. When the actual monitored indicators exceed the preset indicator threshold, intervention or adjustment is required. For example, the preset indicator threshold may include: a first preset threshold, a second preset threshold, and a third preset threshold, each corresponding to different hot migration process indicators. The first preset threshold can be understood as an anomaly judgment standard for the dirty page rate of memory, which is the maximum acceptable value for the dirty page rate; the second preset threshold can be understood as an anomaly judgment standard for the link bandwidth utilization, which is the maximum acceptable value for the link bandwidth utilization; and the third preset threshold can be understood as an anomaly judgment standard for the target host's CPU utilization, which is the maximum acceptable value for the target host's CPU utilization.
[0058] Specifically, preset threshold values for classifying whether the hot migration process indicators are abnormal can be determined based on expert experience or numerical experiments. The hot migration process indicators can be compared with the preset threshold values to determine whether the optimal migration parameters need to be adjusted. For example, when the hot migration process indicators exceed the preset threshold values, the optimal migration parameters can be adjusted to ensure the normal hot migration of the target virtual machine.
[0059] Optionally, this embodiment of the invention further refines step A2. Specifically, it refines the specific steps for adjusting the optimal migration parameters based on the hot migration process indicators and preset indicator thresholds, including: if the memory dirty page rate is greater than a first preset threshold of the preset indicator threshold, then increase the number of pre-copy iterations within the optimal migration parameters or restart the memory deduplication algorithm of the optimal migration parameters; if the link bandwidth utilization is greater than a second preset threshold of the preset indicator threshold, then adjust the bandwidth allocation ratio of the optimal migration parameters; if the target host CPU utilization is greater than a third preset threshold of the preset indicator threshold, then reduce the virtual machine CPU scheduling priority of the optimal migration parameters.
[0060] Specifically, a first preset threshold for judging whether the dirty page rate is abnormal, a second preset threshold for judging whether the link bandwidth utilization is abnormal, and a third preset threshold for judging whether the target host CPU utilization is abnormal can be extracted from the preset index thresholds. The dirty page rate, link bandwidth utilization, and target host CPU utilization are compared with the first preset threshold, the second preset threshold, and the third preset threshold, respectively. If the dirty page rate is greater than the first preset threshold of the preset index thresholds, the number of pre-copy iterations in the optimal migration parameters is increased or the memory deduplication algorithm of the optimal migration parameters is restarted to shorten the total migration time. If the link bandwidth utilization is greater than the second preset threshold of the preset index thresholds, the bandwidth allocation ratio of the optimal migration parameters is adjusted to alleviate network congestion. If the target host CPU utilization is greater than the third preset threshold of the preset index thresholds, the virtual machine CPU scheduling priority of the optimal migration parameters is reduced to avoid host overload.
[0061] In this embodiment of the invention, the migration strategy is dynamically corrected by a real-time monitoring module, which can effectively avoid unexpected delays or failures in the migration process caused by the lack of real-time feedback in traditional methods, thereby improving the success rate and efficiency of migration.
[0062] Example 2
[0063] Figure 2 This invention provides a flowchart of another virtual machine hot migration performance optimization method in Embodiment 2. This embodiment is a refinement of the above embodiment, specifically refining the specific steps of how to determine the target virtual machine, how to obtain the multi-dimensional migration association information of the target virtual machine, and how to perform hot migration.
[0064] like Figure 2 As shown, another method for optimizing virtual machine hot migration performance may include the following specific steps:
[0065] S210. In response to the virtual machine hot migration task, extract the virtual machine identifier carried by the virtual machine hot migration task and determine the target virtual machine corresponding to the virtual machine identifier.
[0066] The virtual machine identifier can be understood as a code or string used to distinguish different virtual machines. For example, the virtual machine identifier may include the virtual machine MAC address or IP address.
[0067] Specifically, when the virtual machine hot migration performance optimization device receives a virtual machine hot migration task, it can extract the virtual machine identifier from the virtual machine hot migration task to identify the target virtual machine that needs to be migrated in the virtual machine hot migration task.
[0068] S220: Collect multi-dimensional migration association information of the target virtual machine based on a distributed perception architecture.
[0069] The distributed perception architecture can be understood as an information collection system for comprehensively collecting migration-related information. The functional components in the distributed perception architecture can be deployed in different locations. For example, the perception agent component of the distributed perception architecture is deployed on the host machine of the target virtual machine to collect multi-dimensional migration-related information such as the virtual machine's running status information. The compute priority network router of the distributed perception architecture is deployed on the communication network of the target virtual machine to collect multi-dimensional migration-related information such as host machine resource information and network-storage link information. The distributed perception architecture also integrates at least one of the following services: PXE service, DHCP service, TFTP service, and MySQL service.
[0070] Specifically, a distributed perception architecture can be built based on components such as the perception agent component and the compute-first network router to collect multi-dimensional migration-related information of the target virtual machine. For example, the perception agent component of the distributed perception architecture can be built into the host machine of the target virtual machine to collect multi-dimensional migration-related information such as the virtual machine's running status information. Multi-dimensional migration-related information such as host machine resource information and network-storage link information can be collected by calling the compute-first network router.
[0071] S230. Save the multidimensional migration association information to the data storage space of the central deployment node of the target virtual machine.
[0072] The central deployment node can be understood as a node in a distributed architecture that centrally manages and stores data. For example, the central deployment node can centrally store multi-dimensional migration-related information and can also perform operations such as unified scheduling of migration tasks and calling models for performance prediction.
[0073] Data storage space can be understood as a data storage area deployed on a central deployment node, used to store migration-related data such as multidimensional migration association information or historical migration data.
[0074] Specifically, the collected multidimensional migration correlation information can be synchronously or asynchronously stored in the data storage space of the central deployment node of the target virtual machine, which can provide data support for subsequent model training.
[0075] S240. Determine at least one set of migration parameter combinations based on the virtual machine hot migration task, and obtain the performance prediction results of each set of migration parameter combinations based on the pre-trained migration performance prediction model and multi-dimensional migration correlation information.
[0076] S250. Determine the optimal migration parameters within each combination of migration parameters based on the preset optimization objectives and performance prediction results.
[0077] S260: Call the computation priority network router to collect network resource registration information or periodically collect information.
[0078] The network-wide resource registration information can be understood as a series of data describing the resource status of all host machines on the network. Periodic collection information can be understood as network-wide host machine resource information collected at fixed time intervals. For example, a computation-priority network router can actively collect resource information from each host machine at preset intervals, such as 10 seconds or 30 seconds. For example, both network-wide resource registration information and periodic collection information can include: the remaining CPU cores, available memory capacity, storage throughput, current load rate, and other resource information for each host machine, reflecting the latest resource status of all host machines on the network.
[0079] Specifically, the network resource registration information or periodic collection information of the network where the target virtual machine is located is collected by calling the compute-priority network router. Furthermore, in this embodiment of the invention, a subscription-push method can also be used to obtain the network resource registration information or periodic collection information. For example, each host machine subscribes to the push service for network resource registration information or periodic collection information from the compute-priority network router to obtain the network resource registration information or periodic collection information.
[0080] S270. Filter the network resource registration information or periodically collect information according to the optimal migration parameters of the target virtual machine to determine the candidate host.
[0081] Specifically, the process involves obtaining full network resource registration information or periodically collected information for the communication network where the target virtual machine resides. This information identifies all host machines registered on that network. Host machines that do not meet the optimal migration parameters are then removed, and the remaining hosts are selected as candidate hosts. It is understood that the candidate host hosts do not include the source host machine where the target virtual machine resides.
[0082] S280. Determine the network-storage link cost of each candidate host according to the service routing table of the effective network router, and select the candidate host with the lowest network-storage link cost as the target host.
[0083] The service routing table can be understood as a routing configuration table that records network link topology and routing rules. For example, the service routing table may contain the link connection relationship between the source host and each candidate host, the bandwidth, latency and other parameters of each link, as well as the routing strategy for data transmission, which can clearly reflect the network topology and link characteristics of the entire network.
[0084] The target host can be understood as a set of host machines that initially meet the resource requirements, and all host machines in this set have the resource foundation to support the target virtual machine.
[0085] Network-to-storage link cost can be understood as an evaluation metric for assessing the quality of the link between the source host and the candidate host, and can be used to determine the efficiency and / or stability of link transmission. For example, the calculation of network-to-storage link cost may at least include: network-to-storage link cost equals the product of transmission latency and bandwidth utilization. A lower network-to-storage link cost indicates shorter transmission latency and / or lower bandwidth utilization, indicating higher link transmission efficiency and better stability.
[0086] Specifically, the service routing table of the effective network router is obtained, and parameters such as transmission latency and bandwidth utilization between the source host and candidate host of the target virtual machine are extracted from the service routing table. The network-storage link cost from the target virtual machine to each candidate host can be obtained by product or weighted product of transmission latency and bandwidth utilization. The candidate host with the lowest network-storage link cost can be determined as the target host.
[0087] S290. Hot-migrate the target virtual machine to the target host machine.
[0088] Specifically, the target virtual machine is hot-migrated from the source host to the target host based on the optimal migration parameters.
[0089] In response to a virtual machine hot migration task, this embodiment of the invention extracts the virtual machine identifier carried by the virtual machine hot migration task and determines the target virtual machine corresponding to the virtual machine identifier. Based on a distributed awareness architecture, it collects multi-dimensional migration association information of the target virtual machine and saves this information to the data storage space of the central deployment node of the target virtual machine. It determines at least one set of migration parameter combinations based on the virtual machine hot migration task, obtains the performance prediction results for each set of migration parameter combinations based on a pre-trained migration performance prediction model and the multi-dimensional migration association information, determines the optimal migration parameters within each migration parameter combination based on a preset optimization objective and each performance prediction result, calls a computation-priority network router to collect network-wide resource registration information or periodically collected information, filters the network-wide resource registration information or periodically collected information according to the optimal migration parameters of the target virtual machine to determine candidate host hosts, determines the network-storage link cost of each candidate host host according to the service routing table of the computation-effective network router, and selects the candidate host host with the lowest network-storage link cost as the target host host, and then hot-migrates the target virtual machine to the target host host. This invention ensures the uniqueness and accuracy of target virtual machine location results by extracting virtual machine identifiers during virtual machine hot migration tasks; it shortens the collection latency of multi-dimensional migration association information and improves the accuracy of performance prediction results by acquiring multi-dimensional migration association information through a distributed awareness architecture; and it avoids potential resource bottlenecks during hot migration by selecting candidate host machines with the lowest network-storage link costs, thereby improving the success rate of hot migration.
[0090] Example 3
[0091] The embodiments of the present invention are optimizations of the above embodiments. Specifically, they supplement the virtual machine hot migration performance optimization method with a migration environment.
[0092] This invention provides a method for optimizing virtual machine hot migration performance, aiming to solve the technical problems of high latency, high risk of service interruption, and poor resource adaptability in traditional virtual machine hot migration, and to improve hot migration efficiency and computing power network compatibility. The method includes the following steps:
[0093] Step 1: Multi-dimensional migration-related information collection
[0094] When a virtual machine hot migration task is triggered (e.g., manually triggered by the user, triggered by system load balancing, or triggered by a fault warning), multi-dimensional migration-related information is collected through a distributed awareness architecture. This multi-dimensional migration-related information specifically includes: virtual machine running status information, source / target host resource information, and cross-node network-storage link information. Among these, the virtual machine running status information is collected through a built-in agent component on the host machine (such as Libvirt), including the dirty page rate (unit: MB / s), the central processing unit (CPU), and other parameters. Unit (CPU) load (unit: %), disk I / O operations per second (unit: times / second), virtual machine image size and hot data percentage; source host information includes the number of CPU cores used, the amount of memory used, and the storage utilization rate; target host candidate set information is collected through a distributed architecture CFN router, including the number of remaining CPU cores, the amount of available memory, and the storage throughput (unit: MB / s); cross-node network-storage link information is based on the CFN protocol and is collected through the CFN router, including the link bandwidth (unit: Gbps), transmission latency (unit: ms), packet loss rate (unit: %), and storage image synchronization rate (unit: MB / s) between the source host and the target host.
[0095] The distributed awareness architecture integrates PXE, DHCP, TFTP, and MySQL services. The PXE service provides network booting for the target host, avoiding local boot disk configuration and shortening host initialization time (reducing preparation time by more than 50%). The DHCP service automatically matches the target host's MAC address and assigns IP addresses, ensuring the connectivity of the migration network (such as VLANs). The TFTP service efficiently transmits the boot files (such as virtual machine image metadata and driver files) and configuration parameters required for migration, with a transmission rate 30% higher than traditional FTP services. The MySQL service stores multi-dimensional migration-related information and historical migration data (such as migration duration and interruption time), providing data support for subsequent model training and policy optimization.
[0096] Step 2: Inference of the migration performance prediction model
[0097] The multi-dimensional migration correlation information collected in step 1 is input into the pre-trained migration performance prediction model to obtain the performance prediction results corresponding to different migration parameter combinations (e.g., total migration latency, service interruption time, resource utilization).
[0098] Furthermore, in the embodiments of the invention, the process of constructing and optimizing the migration performance prediction model includes:
[0099] Data preparation: Collect at least 1000 sets of historical migration data. Each set of data includes "multi-dimensional migration association information + migration parameter combination + actual performance indicators". The migration parameter combination covers memory transfer strategy (pre-copy iteration number: 1-5 times, memory dirty page rate threshold: 10-50MB), CPU scheduling priority (high / medium / low), and storage migration mode (full / incremental / tiered).
[0100] Model training: A fusion algorithm of Long Short-Term Memory (LSTM) and Random Forest is used. The LSTM layer is used to capture temporal features such as dirty page rate and link bandwidth, while the random forest layer is used to fit the nonlinear relationship between transfer parameters and performance indicators. Five-fold cross-validation is used during model training to ensure that the prediction error is ≤5%.
[0101] Iterative optimization: After every 100 transfer tasks are completed, new transfer data is added to the training set to incrementally train the model and avoid performance degradation.
[0102] Step 3: Optimal migration parameters and path adjustment
[0103] Based on preset optimization goals (users can choose "low latency priority" or "low resource consumption priority"), the optimal migration parameters are selected from the performance prediction results, and the target host and migration path are dynamically adjusted in conjunction with the network-wide computing power resource information obtained from the CFN protocol.
[0104] Optimal migration parameter filtering: If "low latency priority" is selected, parameter combinations with total migration latency < 500ms and service interruption time < 50ms will be selected first; if "low resource consumption priority" is selected, parameter combinations with host CPU utilization < 60% and link bandwidth utilization < 70% will be selected first.
[0105] Target host adjustment: Obtain real-time resource registration information of all hosts in the network through the CFN protocol, exclude hosts with resource utilization > 80% from the candidate set, and select hosts with "remaining memory ≥ 1.2 times virtual machine memory + storage throughput ≥ virtual machine IO requirements" as targets;
[0106] Migration path adjustment: Based on the service routing table of the CFN protocol, calculate the link cost between the source and target host (link cost = transmission delay × bandwidth utilization), and select the path with the lowest link cost as the migration path; if the current path is congested (packet loss rate > 3%), then dynamically switch to the suboptimal path through the CFN protocol.
[0107] Step 4: Perform live migration and dynamically monitor and correct.
[0108] Based on the optimal migration parameters, target host, and migration path determined in step 3, a hot migration is performed. The migration process metrics (memory dirty rate, link bandwidth utilization, and target host CPU utilization) are tracked in real time through the monitoring and feedback module. If the metrics exceed the preset threshold, the migration parameters are dynamically adjusted.
[0109] If the memory dirty rate is greater than 30%, increase the number of pre-copy iterations (e.g., from 2 to 3) or enable the LZ4 memory compression algorithm to reduce the amount of data that is repeatedly transmitted.
[0110] If the link bandwidth utilization rate is >90%, adjust the bandwidth allocation ratio (e.g., reduce the migration bandwidth ratio from 80% to 60%) to prioritize the transmission of business data.
[0111] If the target host's CPU utilization is >80%, reduce the virtual machine's CPU scheduling priority (e.g., from "high" to "medium") and enable CPU resource limits (e.g., reduce the maximum number of cores used by 1).
[0112] This invention, using an OpenStack-Train cluster environment as an example, further describes a method for optimizing virtual machine hot migration performance, including the following steps:
[0113] 1. Architecture Deployment
[0114] Distributed awareness architecture deployment: Awareness agent components (integrating Libvirt and PrometheusExporter) are deployed on each host machine (based on KVM virtualization) to collect information about virtual machines and host machines; an information collection module is deployed on the CFN router (based on the OpenDaylight controller) to collect cross-node link information; PXE, DHCP, TFTP and MySQL services are deployed on the central deployment node, where the MySQL database is used to store historical migration data (the table structure includes migration ID, multi-dimensional information fields, parameter combination fields and performance index fields).
[0115] Deployment of the transfer performance prediction model: The trained LSTM-random forest model is encapsulated as an API service (based on the Flask framework) and deployed on the central control node. The transfer control module (integrated in Nova-scheduler) can call the model inference interface via HTTP requests.
[0116] 2. Execution of migration tasks (taking hot migration triggered by load balancing as an example)
[0117] Task Trigger: When the CPU utilization of a host machine (source host machine) in the OpenStack cluster is >85%, Nova-scheduler triggers a virtual machine hot migration task, specifying the virtual machine to be migrated (running AI inference business, 8GB memory, 4 CPU cores).
[0118] Information Collection: The perception agent component collects virtual machine information (memory dirty rate 25MB / s, CPU load 70%, disk I / O operations 500 times / second) and source host information (CPU utilization 88%, memory utilization 82%); the CFN router collects candidate target host information (3 hosts, of which host B has 6 remaining CPU cores, 10GB of memory, and storage throughput 200MB / s) and link information (source-host B link bandwidth 10Gbps, latency 15ms, packet loss rate 0.5%).
[0119] Model inference: The transfer control module inputs multi-dimensional information into the prediction model to obtain performance prediction results for three sets of parameter combinations.
[0120] Combination 1: Pre-copy twice + full storage migration + high CPU priority → migration latency 480ms, interrupt time 45ms, CPU utilization 75%;
[0121] Combination 2: 3 pre-copy operations + incremental storage migration + medium CPU priority → migration latency 520ms, interrupt time 35ms, CPU utilization 65%;
[0122] Combination 3: Pre-copy twice + tiered storage migration + high CPU priority → migration latency 450ms, interrupt time 55ms, CPU utilization 80%;
[0123] Parameter and path selection: The user presets "low latency priority" and selects combination 3 as the optimal parameters; combined with the CFN network-wide resource information, host B is selected as the target host and the source-host B link is the migration path;
[0124] Migration Execution: The migration execution module performs hot migration based on a combination of 3 parameters.
[0125] Memory migration: Two pre-copy operations are used for tiered storage migration (hot data (30%) is transferred first, followed by cold data).
[0126] Real-time monitoring: The monitoring module detected that the link bandwidth utilization rate rose to 92% during the migration, and dynamically reduced the migration bandwidth ratio from 80% to 60%.
[0127] Migration complete: The final total migration latency was 460ms, the service interruption time was 52ms, and the target host CPU utilization was 72%, which meets the business requirements (AI inference business latency requirement <100ms).
[0128] The virtual machine hot migration performance optimization method proposed in this invention reduces the total migration latency by 30%-50% and controls service interruption time to within 50ms through multi-dimensional information perception and prediction models, meeting the needs of latency-sensitive businesses. By combining the CFN protocol to obtain the computing resources of the entire network, the target host and path selection are optimized, and the overload rate of the target host is reduced by 60% after migration. Through real-time monitoring and parameter correction mechanisms, it can cope with sudden situations such as memory dirty rate fluctuations and link congestion, and the migration success rate is improved to over 99%. Based on OpenStack, CFN protocol, and Kolla deployment architecture, it can be seamlessly integrated into existing computing network systems without large-scale modification.
[0129] Example 4
[0130] Figure 3 This is a schematic diagram of another virtual machine hot migration performance optimization device provided in Embodiment 4 of the present invention. Figure 3 As shown, the device includes: an information acquisition module 310, used to determine the target virtual machine for the virtual machine hot migration task in response to the virtual machine hot migration task, and acquire multi-dimensional migration association information of the target virtual machine; a performance prediction module 320, used to determine at least one set of migration parameter combinations according to the virtual machine hot migration task, and acquire the performance prediction results of each set of migration parameter combinations based on a pre-trained migration performance prediction model and multi-dimensional migration association information; an optimal parameter determination module 330, used to determine the optimal migration parameters within each set of migration parameter combinations according to a preset optimization objective and each performance prediction result; and a migration execution module 340, used to perform hot migration on the target virtual machine according to the optimal migration parameters.
[0131] Optionally, the virtual machine hot migration performance optimization device further includes: a model training module, used to collect historical migration data of different virtual machines, wherein the historical migration data includes multi-dimensional migration association information samples, migration parameter combination samples, and actual performance indicators of migration parameter combination samples; training a long short-term memory-random forest fusion algorithm model according to the historical migration data; and using the long short-term memory-random forest fusion algorithm model as the migration performance prediction model when the prediction error of the long short-term memory-random forest fusion algorithm model is determined to be lower than the threshold.
[0132] Optionally, the virtual machine hot migration performance optimization device further includes: a virtual machine monitoring module, comprising: an indicator monitoring unit for monitoring the hot migration process indicators of the target virtual machine performing hot migration; and a parameter adjustment unit for adjusting the optimal migration parameters according to the hot migration process indicators and preset indicator thresholds.
[0133] Optionally, a parameter adjustment unit is used for the following indicators during the hot migration process: 1) Dirty page rate: If the dirty page rate exceeds a first preset threshold, the number of pre-copy iterations within the optimal migration parameters is increased, or the memory deduplication algorithm of the optimal migration parameters is restarted. 2) Link bandwidth utilization: If the link bandwidth utilization exceeds a second preset threshold, the bandwidth allocation ratio of the optimal migration parameters is adjusted. 3) Target host CPU utilization: If the target host CPU utilization exceeds a third preset threshold, the virtual machine CPU scheduling priority of the optimal migration parameters is reduced.
[0134] Optionally, the information acquisition module 310 is specifically used to extract the virtual machine identifier carried by the virtual machine hot migration task and determine the target virtual machine corresponding to the virtual machine identifier; collect multi-dimensional migration association information of the target virtual machine based on the distributed perception architecture; and save the multi-dimensional migration association information to the data storage space of the central deployment node of the target virtual machine; wherein, the perception agent component of the distributed perception architecture is deployed on the host machine of the target virtual machine, the compute priority network router of the distributed perception architecture is deployed on the communication network of the target virtual machine, and the distributed perception architecture integrates at least one of PXE service, DHCP service, TFTP service and MySQL service.
[0135] Optionally, in this embodiment of the invention, the multidimensional migration association information includes at least virtual machine running status information, host machine resource information, and network-storage link information; wherein, the virtual machine running status information includes at least one of the following: memory dirty page rate, CPU load, and disk I / O operations per second; the host machine resource information includes at least one of the following: number of remaining CPU cores, available memory capacity, and storage throughput; the network-storage link information includes at least one of the following: link bandwidth, transmission latency, packet loss rate, and storage image synchronization rate.
[0136] Optionally, the migration execution module 340 is specifically used to call the computation-priority network router to collect network resource registration information or periodically collect information; filter the network resource registration information or periodically collected information according to the optimal migration parameters of the target virtual machine to determine candidate host hosts; determine the network-storage link cost of each candidate host host according to the service routing table of the computation-effective network router, and select the candidate host host with the lowest network-storage link cost as the target host host; and hot migrate the target virtual machine to the target host host.
[0137] The virtual machine hot migration performance optimization device provided in this embodiment of the invention can execute the virtual machine hot migration performance optimization method provided in any embodiment of the invention, and has the corresponding beneficial effects of executing the method.
[0138] Example 5
[0139] Embodiment 5 of the present invention provides an electronic device for performing a virtual machine hot migration performance optimization method, a computer-readable storage medium, and a computer program product.
[0140] Figure 4 A schematic diagram of an electronic device is shown that can be used to implement the virtual machine hot migration performance optimization method of any embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown in the embodiments of the present invention, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the embodiments of the present invention described and / or claimed herein.
[0141] like Figure 4As shown, the electronic device includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for device operation. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0142] Multiple components in the electronic device are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, optical disk, etc.; and a communication unit 19, such as a network card, modem, wireless transceiver, etc. The communication unit 19 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0143] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as virtual machine hot migration performance optimization methods.
[0144] In some embodiments, the virtual machine hot migration performance optimization method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the virtual machine hot migration performance optimization method may be performed. Alternatively, in other embodiments, processor 11 may be configured as the virtual machine hot migration performance optimization method by any other suitable means (e.g., by means of firmware).
[0145] Various embodiments of the systems and technologies described above in these embodiments of the present invention can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), payload programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input system, and at least one output system, and transmitting data and instructions to the storage system, the at least one input system, and the at least one output system.
[0146] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing system, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0147] In the context of embodiments of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, system, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0148] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display system (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing system (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of systems can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or haptic feedback); and input from the user can be received in any form (including: sound input, voice input, or haptic input).
[0149] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.
[0150] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product within the cloud computing service system. This addresses the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.
[0151] It should be understood that the various forms of procedures shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and no limitation is imposed herein.
[0152] The specific embodiments described above do not constitute a limitation on the scope of protection of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for optimizing the performance of virtual machine hot migration, characterized in that, The method includes: In response to a virtual machine hot migration task, the target virtual machine of the virtual machine hot migration task is determined, and the multi-dimensional migration association information of the target virtual machine is obtained; At least one set of migration parameter combinations is determined based on the virtual machine hot migration task, and the performance prediction results of each set of migration parameter combinations are obtained based on the pre-trained migration performance prediction model and the multi-dimensional migration association information. The optimal migration parameters are determined within each combination of migration parameters based on the preset optimization objective and the performance prediction results. The target virtual machine is hot-migrated according to the optimal migration parameters.
2. The method according to claim 1, characterized in that, The step of determining the target virtual machine for the virtual machine hot migration task and obtaining the multi-dimensional migration association information of the target virtual machine includes: Extract the virtual machine identifier carried by the virtual machine hot migration task, and determine the target virtual machine corresponding to the virtual machine identifier; The multidimensional migration association information of the target virtual machine is collected based on a distributed perception architecture; The multidimensional migration association information is saved to the data storage space of the central deployment node of the target virtual machine; The distributed perception architecture's perception agent component is deployed on the host machine of the target virtual machine, and the distributed perception architecture's compute-priority network router is deployed on the target virtual machine's communication network. The distributed perception architecture integrates at least one of the following services: PXE, DHCP, TFTP, and MySQL.
3. The method according to claim 1 or 2, characterized in that, The multidimensional migration association information includes at least virtual machine running status information, host machine resource information, and network-storage link information; The virtual machine running status information includes at least one of the following: memory dirty page rate, CPU load, and disk I / O operations per second; The host machine resource information includes at least one of the following: number of remaining CPU cores, available memory capacity, and storage throughput; The network-storage link information includes at least one of the following: link bandwidth, transmission latency, packet loss rate, and storage mirror synchronization rate.
4. The method according to claim 1, characterized in that, The training process of the transfer performance prediction model includes: Collect historical migration data from different virtual machines, wherein the historical migration data includes multi-dimensional migration association information samples, migration parameter combination samples, and actual performance indicators of the migration parameter combination samples; Train a long short-term memory-random forest fusion algorithm model based on the historical migration data; When the prediction error of the Long Short-Term Memory-Random Forest fusion algorithm model is determined to be lower than the threshold, the Long Short-Term Memory-Random Forest fusion algorithm model is used as the migration performance prediction model.
5. The method according to claim 1, characterized in that, The hot migration of the target virtual machine according to the optimal migration parameters includes: Call the computation-priority network router to collect network resource registration information or periodically collect information; The candidate host host is determined by filtering the network-wide resource registration information or the periodically collected information according to the optimal migration parameters of the target virtual machine. The network-to-storage link cost of each candidate host is determined according to the service routing table of the effective network router, and the candidate host with the lowest network-to-storage link cost is selected as the target host. The target virtual machine is hot-migrated to the target host machine.
6. The method according to claim 1, characterized in that, Also includes: Monitor the hot migration process metrics of the target virtual machine during the hot migration; The optimal migration parameters are adjusted based on the thermal migration process indicators and preset indicator thresholds.
7. The method according to claim 6, characterized in that, The step of adjusting the optimal migration parameters based on the thermal migration process indicators and preset indicator thresholds includes at least one of the following: The hot migration process indicators include the memory dirty page rate. If the memory dirty page rate is greater than the first preset threshold of the preset indicator threshold, the number of pre-copy iterations in the optimal migration parameters is increased or the memory deduplication algorithm of the optimal migration parameters is restarted. The hot migration process indicators include link bandwidth utilization. If the link bandwidth utilization is greater than the second preset threshold of the preset indicator threshold, the bandwidth allocation ratio of the optimal migration parameters is adjusted. The hot migration process indicators include the target host CPU utilization rate. If the target host CPU utilization rate is greater than the third preset threshold of the preset indicator threshold, the virtual machine CPU scheduling priority of the optimal migration parameters is reduced.
8. A virtual machine hot migration performance optimization device, characterized in that, The device includes: The information acquisition module is used to respond to the virtual machine hot migration task, determine the target virtual machine of the virtual machine hot migration task, and obtain the multi-dimensional migration association information of the target virtual machine; The performance prediction module is used to determine at least one set of migration parameter combinations based on the virtual machine hot migration task, and to obtain the performance prediction results of each set of migration parameter combinations based on the pre-trained migration performance prediction model and the multi-dimensional migration association information. The optimal parameter determination module is used to determine the optimal migration parameter within each combination of migration parameters based on the preset optimization objective and each of the performance prediction results. The migration execution module is used to perform hot migration of the target virtual machine according to the optimal migration parameters.
9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the virtual machine hot migration performance optimization method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the virtual machine hot migration performance optimization method according to any one of claims 1-7.