Cloud virtual machine starting optimization method and system

Abstract

The invention provides a cloud virtual machine starting optimization method and system, and relates to the technical field of cloud computing infrastructure virtualization, and the method comprises the steps: collecting system mirror image data of a cloud virtual machine instance, a service dependency relationship, and an I / O rate and a CPU occupancy rate in a starting process, and constructing a dependency graph based on the system mirror image data; analyzing the dependency graph to obtain an analysis result, and predicting future resource demands according to the system mirror image data and the I / O rate and the CPU occupancy rate in the starting process to obtain a prediction result; executing prospective resource optimization according to the prediction result to obtain a starting optimization result of the cloud virtual machine; compared with the prior art, the method has the advantages that through prospective resource optimization, resource competition can be effectively inhibited, and the stability and the speed during concurrent starting of multiple instances are ensured, so that the stability and the utilization efficiency of host machine resources are improved.

Description

Technical Field

[0001] This invention relates to the field of cloud computing infrastructure virtualization technology, and in particular to a cloud virtual machine startup optimization method and system. Background Technology

[0002] Cloud Virtual Machine (CVM) is a virtualized server based on cloud computing technology. It can help enterprises quickly build efficient, reliable and secure application environments. A CVM instance consists of computing resources, storage resources, network resources, security resources and management resources.

[0003] Computing resources include CPU, memory, and disk space. These resources can be elastically scaled according to the needs of an enterprise to meet business requirements. For example, an enterprise can increase the number of CPU cores when needed to improve application performance.

[0004] Storage resources include hard disk space, CDN, SSDs, etc. These resources help enterprises achieve reliable data storage and backup to ensure data security. For example, enterprises can increase hard disk space as needed to meet their data storage requirements.

[0005] Network resources include bandwidth, peak bandwidth, and network latency. These resources help businesses achieve fast network connections and transmissions, thereby improving network efficiency. For example, businesses can increase peak bandwidth when needed to improve network transmission speed.

[0006] Security resources include firewalls, security groups, and DDoS protection. These resources help enterprises achieve data security and reliability to prevent data breaches and attacks. For example, enterprises can add firewalls as needed to protect data security.

[0007] Management resources include a management console, APIs, and SDKs. These resources help enterprises manage their CVM (Cloud Virtual Machine) instances to meet business needs. For example, enterprises can add a management console as needed for better management of their CVM instances.

[0008] In a cloud computing environment, to improve user experience, optimize resource utilization, and reduce operating costs, it is necessary to quickly launch ARM CVM instances.

[0009] However, the following drawbacks exist in the current technology for starting an ARM CVM instance:

[0010] 1. High startup latency and startup path redundancy: Traditional CVM startup uses a fixed, serial initialization process, which has a large number of implicit dependencies and non-critical services blocking the core startup path, resulting in excessively long first service availability time;

[0011] 2. Resource contention during startup storm: When multiple CVM instances start in parallel, the instantaneous demand for host CPU and I / O resources explodes, intensifying resource contention and extending the startup time of all instances.

[0012] 3. Difficulty in adapting to heterogeneous ARM architectures: Traditional solutions struggle to proactively customize the optimal loading and startup sequence based on the configurations of different ARM platforms. Summary of the Invention

[0013] This invention provides a cloud virtual machine startup optimization method and system, the purpose of which is to improve the stability and utilization efficiency of host machine resources.

[0014] To achieve the above objectives, this invention provides a cloud virtual machine startup optimization method for ARM architecture cloud virtual machines. The optimization method includes:

[0015] Step 1: Collect system image data, service dependencies, and I / O rate and CPU utilization during startup of cloud virtual machine instances, and build a dependency graph based on the system image data;

[0016] Step 2: Analyze the dependency graph to obtain the analysis results, and predict future resource requirements based on system image data and I / O rate and CPU utilization during startup to obtain the prediction results;

[0017] Step 3: Perform forward-looking resource tuning based on the prediction results to obtain the startup optimization results of the cloud virtual machine.

[0018] Furthermore, this involves collecting system image data, service dependencies, and I / O rates and CPU usage during the startup process of cloud virtual machines, including:

[0019] Deploy acquisition probes in ARM host machines and cloud virtual machines;

[0020] The system image data, service dependencies, I / O rate and CPU utilization during startup of the cloud virtual machine are collected in real time at a preset frequency.

[0021] Furthermore, building a dependency graph based on system image data includes:

[0022] Construct a directed graph to describe the complex dependencies between the startup service and the cloud virtual machine;

[0023] Perform cross-correlation analysis or Granger causality verification on system image data to obtain implicit resource dependencies between services;

[0024] The directed graph is modified based on implicit resource dependencies to obtain a dependency graph.

[0025] Furthermore, step 2 includes:

[0026] The dependency graph is analyzed using graph analysis methods to obtain analysis results, including the starting critical path and topological redundancy services.

[0027] The system image data and the I / O rate and CPU utilization during the startup process are input into the time series prediction model for prediction. The prediction results include the peak instantaneous CPU demand and instantaneous I / O throughput demand of the instance on the host machine resources within the future time window.

[0028] Furthermore, when multiple cloud virtual machine instances are started in parallel, I / O soft limits are applied to non-critical instances that are performing topology redundancy services based on the prediction results.

[0029] Furthermore, based on the forecast results, forward-looking resource optimization is performed as follows:

[0030] During the expected execution of the critical path, the CPU weight or I / O quota of cloud virtual machine instances is dynamically and proactively increased based on the peak instantaneous CPU demand of the host machine resources.

[0031] Furthermore, following step 3, the following also includes:

[0032] The initial service availability time and actual resource consumption after the cloud virtual machine instance starts are collected as feedback data.

[0033] The internal parameters of the graph analysis method and the time series forecasting model are adjusted using feedback data.

[0034] This invention also provides a cloud virtual machine startup optimization system for ARM architecture cloud virtual machines. The optimization system includes:

[0035] The data acquisition module is used to collect system image data, service dependencies, and I / O rate and CPU utilization during the startup process of cloud virtual machine instances, and to build a dependency graph based on the system image data.

[0036] The prediction module is used to analyze the dependency graph to obtain analysis results, and predict future resource requirements based on system image data and I / O rate and CPU utilization during startup.

[0037] The optimization module is used to perform forward-looking resource tuning based on the prediction results, and obtain the startup optimization results of the cloud virtual machine.

[0038] Furthermore, system optimization also includes:

[0039] The feedback module is used to collect the initial service availability time and actual resource consumption of the cloud virtual machine instance after startup, and adjust the internal parameters of the graph analysis method and the time series prediction model.

[0040] Furthermore, the feedback module includes:

[0041] The data collection submodule is used to collect the initial service availability time and actual resource consumption of the cloud virtual machine instance after startup as feedback data.

[0042] The adjustment submodule is used to adjust the internal parameters of the graph analysis method and the time series forecasting model using feedback data.

[0043] The above-described solution of the present invention has the following beneficial effects:

[0044] This invention collects system image data, service dependencies, and I / O rates and CPU utilization during the startup process of cloud virtual machine instances, and constructs a dependency graph based on the system image data. The dependency graph is analyzed to obtain analysis results, and future resource requirements are predicted based on the system image data and I / O rates and CPU utilization during startup. Based on the prediction results, forward-looking resource optimization is performed to obtain optimized startup results for the cloud virtual machine. Compared with existing technologies, this invention, through forward-looking resource optimization, can effectively suppress resource contention, ensure stability and speed during concurrent startup of multiple instances, thereby improving the stability and utilization efficiency of host machine resources.

[0045] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation

[0047] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0048] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] This invention addresses existing problems by providing a cloud virtual machine startup optimization method and system.

[0050] like Figure 1 As shown, an embodiment of the present invention provides a cloud virtual machine startup optimization method for ARM architecture cloud virtual machines. The optimization method includes:

[0051] Step 1: Collect system image data, service dependencies, and I / O rate and CPU utilization during startup of cloud virtual machine instances, and build a dependency graph based on the system image data;

[0052] Step 2: Analyze the dependency graph to obtain the analysis results, and predict future resource requirements based on system image data and I / O rate and CPU utilization during startup to obtain the prediction results;

[0053] Step 3: Perform forward-looking resource tuning based on the prediction results to obtain the startup optimization results of the cloud virtual machine.

[0054] Specifically, this involves collecting system image data, service dependencies, and I / O rates and CPU usage during the startup process of cloud virtual machines, including:

[0055] Deploy acquisition probes in ARM host machines and cloud virtual machines;

[0056] The system image data, service dependencies, I / O rate and CPU utilization during startup of the cloud virtual machine are collected in real time at a preset frequency.

[0057] In this embodiment of the invention, a high-frequency real-time acquisition system image data of a cloud virtual machine is performed at a preset frequency of 1000Hz or higher. The system image data includes host load, number of concurrent instances, system configuration data, and ARM hardware configuration data.

[0058] Specifically, the dependency graph is constructed based on system image data, including:

[0059] Construct a directed graph to describe the complex dependencies between the startup service and the cloud virtual machine;

[0060] Perform cross-correlation analysis or Granger causality verification on system image data to obtain implicit resource dependencies between services;

[0061] The directed graph is modified based on implicit resource dependencies to obtain a dependency graph.

[0062] It should be noted that the embodiments of the present invention correct the edge weights of the directed graph based on implicit resource dependencies.

[0063] Specifically, step 2 includes:

[0064] The dependency graph is analyzed using graph analysis methods to obtain analysis results, including the starting critical path and topological redundancy services.

[0065] The system image data and the I / O rate and CPU utilization during the startup process are input into the time series prediction model for prediction. The prediction results include the peak instantaneous CPU demand and instantaneous I / O throughput demand of the instance on the host machine resources within the future time window.

[0066] It should be noted that the graph analysis method used in the embodiments of the present invention is a combination of topological sorting and critical path algorithm, while the time series prediction model is a hybrid model consisting of an autoregressive model and a moving average model, referred to as the ARMA model, or other methods for studying time series.

[0067] In this embodiment of the invention, the cloud virtual machine generates a highly parallel optimized startup sequence based on the critical startup path and topology redundancy services, and generates dynamic resource tuning instructions based on I / O instantaneous throughput requirements.

[0068] This invention can reduce the average availability time of cloud virtual machine instances by 80% by decoupling the startup dependency graph and highly parallelizing the startup sequence.

[0069] The preferred approach is to implement soft I / O limits on non-critical instances that are performing topology redundancy services when multiple cloud virtual machine instances are started in parallel, based on the prediction results.

[0070] Specifically, based on the forecast results, proactive resource optimization is performed as follows:

[0071] During the expected execution of the critical path, the CPU weight or I / O quota of cloud virtual machine instances is dynamically and proactively increased based on the peak instantaneous CPU demand of the host machine resources.

[0072] The embodiments of the present invention ensure stability and speed when multiple instances are started concurrently by predicting resource peaks in advance and implementing dynamic I / O / CPU quotas and storm suppression.

[0073] Specifically, after step 3, the following is also included:

[0074] The initial service availability time and actual resource consumption after the cloud virtual machine instance starts are collected as feedback data.

[0075] The internal parameters of the graph analysis method and the time series forecasting model are adjusted using feedback data.

[0076] This invention relates to a cloud server cluster based on the ARM Neoverse architecture. The available time is defined from the moment the CVM startup command is issued from the virtualization layer until the ports and processes of critical applications (such as Web Server or Cloud Phone services) within the CVM are in the Running state. The following experimental results were obtained:

[0077] Table 1 Experimental Results

[0078] As can be seen from Table 1 above, the optimization method provided by the embodiments of the present invention shortens the average TTM of ARM instances by 80%, and at the same time, through dynamic resource quotas, the maximum instantaneous CPU occupancy rate during the startup phase is reduced by 33%, effectively suppressing startup storms and improving the stability and utilization efficiency of host resources.

[0079] This invention collects system image data, service dependencies, and I / O rates and CPU utilization during the startup process of cloud virtual machine instances, and constructs a dependency graph based on the system image data. The dependency graph is analyzed to obtain analysis results, and future resource requirements are predicted based on the system image data and I / O rates and CPU utilization during startup. Based on the prediction results, forward-looking resource optimization is performed to obtain optimized startup results for the cloud virtual machine. Compared with existing technologies, this invention, through forward-looking resource optimization, can effectively suppress resource contention, ensure stability and speed during concurrent startup of multiple instances, and thus improve the stability and utilization efficiency of host machine resources.

[0080] This invention also provides a cloud virtual machine startup optimization system for ARM architecture cloud virtual machines. The optimization system includes:

[0081] The data acquisition module is used to collect system image data, service dependencies, and I / O rate and CPU utilization during the startup process of cloud virtual machine instances, and to build a dependency graph based on the system image data.

[0082] The prediction module is used to analyze the dependency graph to obtain analysis results, and predict future resource requirements based on system image data and I / O rate and CPU utilization during startup.

[0083] The optimization module is used to perform forward-looking resource tuning based on the prediction results, and obtain the startup optimization results of the cloud virtual machine.

[0084] Specifically, system optimization also includes:

[0085] The feedback module is used to collect the initial service availability time and actual resource consumption of the cloud virtual machine instance after startup, and adjust the internal parameters of the graph analysis method and the time series prediction model.

[0086] Specifically, the feedback module includes:

[0087] The data collection submodule is used to collect the initial service availability time and actual resource consumption of the cloud virtual machine instance after startup as feedback data.

[0088] The adjustment submodule is used to adjust the internal parameters of the graph analysis method and the time series forecasting model using feedback data.

[0089] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A cloud virtual machine startup optimization method, characterized in that, The cloud virtual machine for ARM architecture, the optimization method comprises: Step 1, collecting system image data, service dependency relationship and I / O rate and CPU occupancy rate in the starting process of the cloud virtual machine instance, and constructing a dependency graph based on the system image data; Step 2, analyzing the dependency graph to obtain an analysis result, and predicting future resource demand according to the system image data and the I / O rate and CPU occupancy rate in the starting process to obtain a prediction result; Step 3, performing forward-looking resource optimization according to the prediction result to obtain the starting optimization result of the cloud virtual machine.

2. The cloud virtual machine launch optimization method of claim 1, wherein, Collecting system image data, service dependency relationship and I / O rate and CPU occupancy rate in the starting process of the cloud virtual machine comprises: Deploying a collection probe in the ARM host and the cloud virtual machine; Collecting system image data, service dependency relationship and I / O rate and CPU occupancy rate in the starting process of the cloud virtual machine in real time at a preset frequency.

3. The cloud virtual machine launch optimization method of claim 2, wherein, Constructing a dependency graph based on the system image data comprises: Constructing a directed graph for describing the complex dependency relationship between the starting service and the cloud virtual machine; Performing cross-correlation analysis or Granger causality verification on the system image data to obtain implicit resource dependency between services; Based on the implicit resource dependency, the directed graph is modified to obtain a dependency graph.

4. The cloud virtual machine launch optimization method of claim 3, wherein, The step 2 comprises: Using a graph analysis method to analyze the dependency graph to obtain an analysis result, the analysis result comprising a starting critical path and a topological redundant service; Inputting the system image data and the I / O rate and CPU occupancy rate in the starting process into a time series prediction model for prediction to obtain a prediction result, the prediction result comprising CPU instantaneous demand peak value and I / O instantaneous throughput demand of the instance to the host resource in a future time window.

5. The cloud virtual machine launch optimization method of claim 4, wherein, When a plurality of cloud virtual machine instances are started in parallel, according to the prediction result, non-critical instances executing the topological redundant service are implemented I / O soft limit.

6. The cloud virtual machine launch optimization method of claim 4, wherein, According to the prediction result, forward-looking resource optimization is performed, which is: During the predicted execution of the critical path, the CPU weight or I / O quota of the cloud virtual machine instance is dynamically and forwardly improved according to the CPU instantaneous demand peak value of the host resource.

7. The cloud virtual machine starting optimization method according to claim 4, further comprising, after step 3: Collecting the first service available time and the actual resource consumption after the cloud virtual machine instance is started as feedback data; Using the feedback data to adjust the internal parameters of the graph analysis method and the time series prediction model.

8. A cloud virtual machine startup optimization system, comprising: The cloud virtual machine for ARM architecture, the optimization system comprises: A collection module for collecting system image data, service dependency relationship and I / O rate and CPU occupancy rate in the starting process of the cloud virtual machine instance, and constructing a dependency graph based on the system image data; A prediction module is configured to analyze the dependency graph to obtain an analysis result, and predict future resource demand according to the system image data and I / O rate and CPU occupancy rate in the startup process to obtain a prediction result. An optimization module is configured to perform prospective resource tuning according to the prediction result to obtain a startup optimization result of the cloud virtual machine.

9. The cloud virtual machine launch optimization system of claim 8, wherein, Further comprising: A feedback module is configured to collect a first service available time and actual resource consumption after the cloud virtual machine instance is started, and adjust internal parameters of a graph analysis method and a time series prediction model.

10. The cloud virtual machine launch optimization system of claim 9, wherein, The feedback module comprises: A collection submodule is configured to collect the first service available time and actual resource consumption after the cloud virtual machine instance is started as feedback data. An adjustment submodule is configured to adjust the internal parameters of the graph analysis method and the time series prediction model by using the feedback data.

Images