Public cloud multi-tenant performance monitoring method, device, equipment, storage medium and program product

Abstract

The application relates to a public cloud multi-tenant performance monitoring method, device, equipment, storage medium and program product. The method comprises the following steps: collecting a plurality of low-level indexes of a cloud application runtime on a host of a target virtual machine node and a target virtual machine; the low-level indexes comprise at least one of CPU utilization, a hardware counter, a host resource state, resource usage characteristics, a hardware event vector and interference intensity; inputting the plurality of low-level indexes into a proxy model for performance evaluation to obtain a first evaluation result; the first evaluation result comprises a target performance degradation estimation value, a performance degradation reason of the cloud application and a contribution degree of the degradation reason; the target performance degradation estimation value is used for representing a degradation degree of the cloud application; and performing degradation evaluation according to the first evaluation result to obtain a second evaluation result. The above method improves the resource utilization rate and cost efficiency of the public cloud infrastructure, and further enhances the foresight and accuracy of cloud application performance management.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and in particular to a method, apparatus, device, storage medium, and program product for multi-tenant-aware performance monitoring of public clouds. Background Technology

[0002] With the rapid development of cloud computing technology, multi-tenancy has become the mainstream design for public cloud infrastructure. By sharing hardware resources (such as CPU, memory, cache, network bandwidth, etc.) on the same physical host, public clouds can improve resource utilization and cost efficiency. However, this multi-tenant co-location can lead to interference on shared resources, resulting in performance degradation of cloud applications. Therefore, cloud providers need to monitor the performance of cloud applications, accurately determine the timing and severity of performance degradation, and implement interference-aware migration or other optimization measures to mitigate resource waste caused by performance degradation.

[0003] Traditional technologies primarily rely on threshold judgments, classic performance metrics, or machine learning methods to improve the accuracy of monitoring cloud application performance. However, these traditional technologies, which depend on fixed models or static thresholds, struggle to accurately monitor cloud application performance under dynamic loads. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, device, storage medium, and program product for multi-tenant-aware performance monitoring of public clouds to address the aforementioned technical issues.

[0005] Firstly, this application provides a method for multi-tenant-aware performance monitoring in public clouds, including:

[0006] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of the cloud application; the low-level metrics include at least one of CPU utilization, hardware counters, host resource status, resource usage characteristics, hardware event vectors, and interference intensity.

[0007] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result. The first evaluation result includes the target performance degradation estimate, the causes of cloud application performance degradation, and the contribution of the causes of degradation. The target performance degradation estimate is used to characterize the degree of degradation of the cloud application.

[0008] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0009] In one embodiment, the proxy model includes a feature enhancement unit, a transfer learning unit, and an interpretation fusion unit. Low-level metrics are input into the proxy model for performance evaluation to obtain a first evaluation result, including:

[0010] Multiple low-level indicators are input into the feature enhancement unit to obtain the enhanced perturbation features;

[0011] The enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results; the initial evaluation results include the reasons for the performance degradation caused by multiple current perturbations, as well as the estimated value of the target performance degradation;

[0012] The initial evaluation results are input into the interpretation and fusion unit to obtain the first evaluation result.

[0013] In one embodiment, multiple low-level indicators are input into the feature enhancement unit to obtain enhanced perturbation features, including:

[0014] Initial modeling is performed based on multiple low-level indicators to obtain a feature set;

[0015] The feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

[0016] In one embodiment, a degradation assessment is performed based on the first assessment result to obtain a second assessment result, including:

[0017] The target performance degradation estimate is compared with the preset degradation estimate threshold;

[0018] If the target performance degradation estimate is greater than the preset degradation estimation threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0019] If the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring.

[0020] In one embodiment, a degradation assessment is performed based on the first assessment result to obtain a second assessment result, including:

[0021] The first evaluation result is input into the machine learning monitoring controller for degradation evaluation, resulting in multiple monitoring decisions.

[0022] By using a reward function to filter multiple monitoring decisions, a second evaluation result is obtained, which includes the target monitoring decision.

[0023] In one embodiment, the above-mentioned public cloud multi-tenant-aware performance monitoring method further includes:

[0024] Various types of interference are applied to the benchmark application using multiple interference generators to obtain sample data.

[0025] The initial agent model is trained based on the sample data to obtain the trained agent model.

[0026] Secondly, this application also provides a multi-tenant-aware performance monitoring device for public clouds, comprising:

[0027] The data collection module is used to collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications.

[0028] The first evaluation module is used to input multiple low-level metrics into the proxy model for performance evaluation and obtain the first evaluation result.

[0029] The second assessment module is used to perform a degradation assessment based on the first assessment result to obtain the second assessment result.

[0030] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0031] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0032] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0033] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0034] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0035] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0036] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0037] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0038] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0039] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0040] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0041] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0042] The aforementioned public cloud multi-tenant-aware performance monitoring method, device, equipment, storage medium, and program product collects multiple low-level indicators during the runtime of cloud applications on the target virtual machine node and the host machine of the target virtual machine. These low-level indicators include at least one of CPU utilization, hardware counters, host resource status, resource usage characteristics, hardware event vectors, and interference intensity. Multiple low-level indicators are input into a proxy model for performance evaluation to obtain a first evaluation result. The first evaluation result includes a target performance degradation estimate, the cause of cloud application performance degradation, and the contribution of the degradation cause. The target performance degradation estimate is used to characterize the degree of cloud application degradation. A degradation evaluation is performed based on the first evaluation result to obtain a second evaluation result. This method, by collecting low-level indicators covering multiple dimensions such as hardware operating status, resource usage characteristics, and interference intensity, provides... The performance evaluation of cloud applications establishes a comprehensive and accurate data foundation. Simultaneously, leveraging the dual-dimensional first evaluation results—the target performance degradation estimate and the causes of degradation—output by the proxy model, it not only achieves a quantitative characterization of the degree of cloud application performance degradation but also accurately pinpoints the causes, overcoming the limitations of traditional technologies in terms of monitoring accuracy under dynamic loads. Furthermore, the second evaluation results, derived from the first evaluation, deepen the understanding of performance degradation trends and their impact scope, thereby helping cloud providers to promptly implement targeted optimization measures such as interference-aware migration. This effectively alleviates the resource waste caused by performance degradation, while also improving the resource utilization and cost efficiency of public cloud infrastructure. Ultimately, this enhances the foresight and accuracy of cloud application performance management, ensuring the overall stability and efficiency of cloud application operation in a multi-tenant mode. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is an application environment diagram of a multi-tenant-aware performance monitoring method for public clouds in one embodiment;

[0045] Figure 2 This is one of the flowcharts illustrating a multi-tenant-aware performance monitoring method for public clouds in one embodiment;

[0046] Figure 3 This is the second flowchart of a public cloud multi-tenant-aware performance monitoring method in one embodiment;

[0047] Figure 4This is the third flowchart of a public cloud multi-tenant-aware performance monitoring method in one embodiment;

[0048] Figure 5 This is the fourth flowchart of a multi-tenant-aware performance monitoring method for public clouds in one embodiment;

[0049] Figure 6 This is the fifth flowchart illustrating a multi-tenant-aware performance monitoring method for public clouds in one embodiment.

[0050] Figure 7 This is the sixth flowchart of a public cloud multi-tenant-aware performance monitoring method in one embodiment;

[0051] Figure 8 Here is a flowchart of the interference generation and offline dataset construction process in one embodiment;

[0052] Figure 9 This is the seventh flowchart of a multi-tenant-aware performance monitoring method for public clouds in one embodiment;

[0053] Figure 10 This is a technical flowchart of the overall solution in one embodiment;

[0054] Figure 11 This is a structural block diagram of a multi-tenant perception performance monitoring device for a public cloud in one embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0057] With the rapid development of cloud computing technology, multi-tenancy has become the mainstream design for public cloud infrastructure. By sharing hardware resources (such as CPU, memory, cache, network bandwidth, etc.) on the same physical host, public clouds can improve resource utilization and cost efficiency. However, this multi-tenant co-location can lead to interference on shared resources, resulting in performance degradation of cloud applications. Therefore, cloud providers need to monitor the performance of cloud applications, accurately determine the timing and severity of performance degradation, and implement interference-aware migration or other optimization measures to mitigate resource waste caused by performance degradation.

[0058] Traditional technologies primarily rely on threshold judgments, classic performance metrics, or machine learning methods to improve the accuracy of monitoring cloud application performance. However, these traditional technologies, which depend on fixed models or static thresholds, struggle to accurately monitor cloud application performance under dynamic loads.

[0059] In view of the above-mentioned technical problems, this application provides a method for monitoring the multi-tenant-aware performance of a public cloud. The following embodiments will specifically illustrate the method for monitoring the multi-tenant-aware performance of a public cloud.

[0060] The public cloud multi-tenant-aware performance monitoring method provided in this application embodiment can be applied to, for example... Figure 1 The computer device shown includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a multi-tenant-aware performance monitoring method for public clouds. The display unit of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0061] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0062] In one exemplary embodiment, such as Figure 2 As shown, a multi-tenant-aware performance monitoring method for public clouds is provided, which is then applied to... Figure 1 Taking computer devices as an example, the explanation includes:

[0063] S201 collects multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications.

[0064] The low-level metrics include at least one of the following: CPU utilization, hardware counters, host resource status, resource usage characteristics, hardware event vectors, interference intensity, and hardware event characteristics.

[0065] In this embodiment, during the monitoring period, the computer device collects low-level performance metrics data in real time from all target virtual machine nodes and their corresponding physical hosts in the public cloud multi-tenant environment. These low-level metrics include, but are not limited to, CPU utilization (e.g., single-core / multi-core utilization, idle / loaded state), hardware counters (e.g., instruction cycle, cache hit rate, memory access latency), host resource status (e.g., memory utilization, disk I / O throughput, network bandwidth usage), hardware event vectors (e.g., interrupt request frequency, device error logs, power state changes), interference intensity (e.g., performance fluctuations caused by resource contention between tenants, the impact of virtual machine migration on the host), resource usage characteristics, and hardware event characteristics. Simultaneously, during the collection process, the computer device must ensure the timeliness (e.g., second-level sampling frequency), completeness (e.g., all active nodes), and accuracy of the low-level metric data.

[0066] S202 inputs multiple low-level metrics into the agent model for performance evaluation and obtains the first evaluation result.

[0067] The first evaluation result includes the target performance degradation estimate, the causes of cloud application performance degradation, and the contribution of the causes of degradation; the target performance degradation estimate is used to characterize the degree of degradation of cloud application; the surrogate model includes a feature enhancement unit, a transfer learning unit, and an explanation fusion unit.

[0068] In this embodiment, the computer device estimates the degradation value of cloud applications based on multiple low-level indicators, inputting these indicators into a proxy model. First, the low-level indicators are input into a feature enhancement unit, which uses deep learning and other artificial intelligence algorithms to capture the nonlinear perturbations of complex interferences (such as network jitter and resource contention) on the indicators, outputting an enhanced feature vector with the same dimension. Then, the enhanced feature vector is input into a transfer learning unit, which is built based on a pre-trained cloud application performance prediction model. The transfer learning unit maps the enhanced feature vector to the target performance space, obtaining the estimated target performance degradation value caused by the current interference and multiple degradation causes. Next, the computer device calls an interpretation and fusion unit to integrate the preliminary results from the transfer learning unit and uses a weighted fusion algorithm to quantify the contribution ratio of multiple potential causes to cloud application performance degradation, selecting the most significant Top-k (e.g., k=3) key features and their specific contributions. Finally, the computer device integrates the estimated target performance degradation value, the causes of cloud application performance degradation, and the contribution of the degradation causes through the proxy model to obtain a first evaluation result.

[0069] S203, based on the first assessment result, a degradation assessment is performed to obtain the second assessment result.

[0070] The second evaluation result is obtained by the machine learning monitoring controller; its evaluation results include the causes of cloud application degradation and target monitoring decisions (such as migration or rate limiting measures).

[0071] In this embodiment, after the computer device generates a first evaluation result based on multiple low-level indicators through a proxy model, it inputs the first evaluation result into a machine learning monitoring controller for a second evaluation of the degradation performance to obtain a second evaluation result. Specifically, obtaining the second evaluation result can be achieved in two ways: One way is that the computer device can use the threshold comparison unit in the machine learning monitoring controller to perform a threshold comparison of the degradation performance, and generate a corresponding target monitoring decision based on the threshold comparison result. If the degradation degree does not exceed the threshold, the decision to maintain the current monitoring configuration is output; if it exceeds the threshold, the computer device triggers the corresponding level of adjustment strategy by combining the degree and duration of the current degradation degree exceeding the threshold. The other way is that after the computer device inputs the first evaluation result into the machine learning monitoring controller, the machine learning monitoring controller first maps the causes of performance degradation of multiple cloud applications contained in the first evaluation result into actions in the action space. Each action represents a monitoring decision, and the action benefits among the multiple monitoring decisions are filtered according to a preset filtering function. The monitoring decision with the highest action benefit is identified as the target monitoring decision.

[0072] The aforementioned multi-tenant-aware performance monitoring method for public clouds collects multiple low-level metrics during the runtime of cloud applications on the target virtual machine node and its host machine. These low-level metrics include at least one of CPU utilization, hardware counters, host resource status, resource usage characteristics, hardware event vectors, and interference intensity. These low-level metrics are input into a proxy model for performance evaluation, yielding a first evaluation result. This first evaluation result includes an estimated target performance degradation value, the cause of the cloud application's performance degradation, and the contribution of the degradation cause. The estimated target performance degradation value characterizes the degree of degradation in the cloud application. A second evaluation result is obtained based on the first evaluation result. This method, by collecting low-level metrics covering multiple dimensions such as hardware operating status, resource usage characteristics, and interference intensity, provides a basis for performance evaluation of cloud applications. A comprehensive and accurate data foundation was established. Simultaneously, leveraging the dual-dimensional first assessment results—the target performance degradation estimate and the causes of degradation—output by the proxy model, not only was the degree of cloud application performance degradation quantitatively characterized, but the causes of performance degradation were also accurately located, overcoming the limitations of traditional technologies in terms of insufficient monitoring accuracy under dynamic loads. Furthermore, the second assessment results, derived from the first assessment, deepened the understanding of performance degradation trends and their impact scope, thereby helping cloud providers to promptly implement targeted optimization measures such as interference-aware migration. This effectively alleviated the resource waste caused by performance degradation, while also improving the resource utilization and cost efficiency of public cloud infrastructure. Ultimately, this enhanced the foresight and accuracy of cloud application performance management, ensuring the overall stability and efficiency of cloud application operation in a multi-tenant mode.

[0073] In one exemplary embodiment, such as Figure 3 As shown, the phrase "inputting low-level metrics into the proxy model for performance evaluation to obtain the first evaluation result" in S202 above includes:

[0074] S301 inputs multiple low-level indicators into the feature enhancement unit to obtain the enhanced perturbation features.

[0075] In this embodiment, the computer device inputs multiple low-level metrics into a feature enhancement unit. The feature enhancement unit is based on a pre-trained offline denoising autoencoder (DAE). During the offline phase, the DAE uses runtime data from a large number of virtual machines running in an interference-free environment to learn the inherent patterns or typical distribution characteristics of multiple low-level metrics under normal operating conditions, as well as the potential impact of other different interference factors. Subsequently, the DAE uses the interference-free operating knowledge learned in offline mode to calibrate the multiple low-level metrics input by the computer device (such as smoothing or completing missing data), and outputs the theoretical baseline metrics that the target virtual machine should achieve under ideal interference-free conditions. Then, the computer device calculates the deviation between the multiple low-level metrics and the theoretical baseline metrics. Finally, the computer device obtains the enhanced perturbation features with the same degree and dimension of perturbation of the low-level metrics of the virtual machine by various interferences during the operation of the cloud application.

[0076] S302, the enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results.

[0077] The initial assessment results include the causes of performance degradation due to multiple current disturbances, as well as the estimated target performance degradation.

[0078] In this embodiment, the computer device inputs enhanced perturbation features of the same dimension into the transfer learning unit in the surrogate model. The transfer learning unit is constructed based on a pre-trained cloud application performance prediction model, and maps the enhanced feature vectors to the target performance space to obtain an estimate of the target performance degradation caused by the current perturbation and multiple degradation causes. For example, the computer device can use a Domain Adaptation Neural Network (DANN) to quantitatively evaluate the initial degradation performance of the cloud application and generate an initial evaluation result that includes an estimate of the target performance degradation (such as the percentage increase in response time and the decrease in throughput) and multiple causes of performance degradation caused by the current perturbation (such as insufficient CPU resources and network bandwidth bottlenecks). This DANN can transfer offline-trained knowledge to unknown application scenarios. First, the DANN processes the enhanced perturbation features input from the computer device through a feature extractor. At the same time, it uses the performance degradation patterns (such as the quantitative relationship of throughput decline under memory bottleneck) obtained from offline training in known applications (such as e-commerce trading platforms and traditional network services) during the label prediction period to predict the degradation indicators of cloud applications (such as a throughput decrease of 18%). Then, the domain discriminator continuously learns the differences between known applications and target cloud applications (such as the statistical distribution of perturbation response features) through adversarial training, and feeds these differences back to the feature extractor to adjust the parameters in the feature extractor, thereby improving the DANN's generalization ability in dynamic scenarios.

[0079] S303, input the initial evaluation results into the interpretation and fusion unit to obtain the first evaluation result.

[0080] In this embodiment, the computer device inputs the initial evaluation results into the third unit of the proxy model, namely the interpretation and fusion unit, to integrate the initial evaluation results and obtain a first evaluation result. For example, the computer device uses Shapley Additive exPlanations (SHAP) to integrate and calculate the initial evaluation results, quantifies the contribution ratio of multiple potential causes in the initial evaluation results to the performance degradation of cloud applications, and selects the most significant Top-k (e.g., k=2) key features and their specific contributions (e.g., abnormal CPU utilization contributes 65%, network latency fluctuation contributes 25%). Finally, the proxy model integrates the target performance degradation estimate, the causes of cloud application performance degradation, and the contribution of the causes to obtain the first evaluation result.

[0081] In one exemplary embodiment, such as Figure 4 As shown, the "inputting multiple low-level indicators into the feature enhancement unit to obtain enhanced perturbation features" in S301 above includes:

[0082] S401, initial modeling is performed based on multiple low-level indicators to obtain a feature set.

[0083] In this embodiment, during the monitoring period, the computer device performs initial modeling on the low-level performance metrics data of all target virtual machine nodes and their corresponding physical hosts in the public cloud multi-tenant environment. The target virtual machine nodes are treated as monitoring objects, and each target virtual machine node includes resource usage and hardware time characteristics to represent the impact of interference and performance degradation. The computer device models multiple low-level metrics into a feature set, represented as: ,in, This indicates that the resources are used for the target virtual machine node. Indicates hardware time. This represents the overall resources of the host machine. The initial characteristics of each target virtual machine node are represented as follows: ,in, Indicates the central processing unit (CPU) utilization. Indicates memory usage. This indicates the utilization rate of the last-level cache.

[0084] S402, the feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

[0085] In this embodiment, after modeling multiple low-level indicators into a feature set, the computer device inputs the feature set into a feature enhancement unit to process the perturbation information contained in the feature set, thereby obtaining enhanced perturbation features. This embodiment relates to a specific method for inputting a feature set into a feature enhancement unit, processing the perturbation information contained in the feature set, and obtaining enhanced perturbation features. This method is similar to the aforementioned... Figures 2-4 The methods described in any implementation are basically the same, and for details, please refer to the foregoing explanation, which will not be repeated here.

[0086] In one exemplary embodiment, such as Figure 5 As shown, "performing a degradation assessment based on the first assessment result to obtain a second assessment result" in S203 above includes:

[0087] S501, compare the target performance degradation estimate with the preset degradation estimate threshold.

[0088] In this embodiment, after the computer device generates a first evaluation result based on multiple low-level indicators through a proxy model, it inputs the first evaluation result into a machine learning monitoring controller for a second evaluation of the degraded performance to obtain a second evaluation result. Specifically, there are two ways to obtain the second evaluation result: one way is that the computer device can use the threshold comparison unit in the machine learning monitoring controller to compare the target performance degradation estimate in the first evaluation result with preset degradation estimate thresholds (including response latency threshold, CPU / memory resource saturation threshold, service availability threshold, etc.) one by one, and generate a corresponding target monitoring decision based on the comparison result.

[0089] S502: If the target performance degradation estimate is greater than the preset degradation estimate threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0090] In this embodiment, the computer device compares the target performance degradation estimate in the first evaluation result with a preset degradation estimation threshold. If the target performance degradation estimate is greater than the preset degradation estimation threshold (e.g., a service quality breach, such as a response time 25% longer than the baseline under no-interference conditions or exceeding the upper limit agreed upon in the protocol), the computer device immediately integrates the causes of performance degradation of the cloud application and outputs an alarm report. At the same time, it automatically triggers migration or rate limiting measures according to the severity of the degradation cause. For example, if there is a shortage of local resources and there are idle availability zones, the computer device automatically migrates the affected cloud application to a node with sufficient resources to achieve rapid expansion. If the network congestion is caused by a sudden surge in traffic across the entire network at a certain time point, the computer device triggers a rate limiting mechanism to prioritize core requests and limit non-core requests to prevent further performance degradation of the cloud application.

[0091] S503 If the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring.

[0092] In this embodiment, the computer device compares the target performance degradation estimate in the first evaluation result with a preset degradation estimation threshold. If the target performance degradation estimate is not greater than the preset degradation estimation threshold, the next monitoring cycle begins. Multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during cloud application runtime are collected to evaluate the degradation performance of the cloud application and obtain the evaluation result. This embodiment involves a specific method for collecting multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during cloud application runtime, evaluating the degradation performance of the cloud application, and obtaining the evaluation result. This method is similar to the aforementioned... Figures 2-4 The methods described in any implementation are basically the same, and for details, please refer to the foregoing explanation, which will not be repeated here.

[0093] In another exemplary embodiment, such as Figure 6 As shown, "performing a degradation assessment based on the first assessment result to obtain a second assessment result" in S203 above includes:

[0094] S601 inputs the first evaluation result into the machine learning monitoring controller for degradation evaluation, and obtains multiple monitoring decisions.

[0095] In this embodiment, after the computer device generates a first evaluation result based on multiple low-level indicators through a proxy model, it inputs the first evaluation result into a machine learning monitoring controller for a second evaluation of the degradation performance to obtain a second evaluation result. Specifically, obtaining the second evaluation result can be achieved in two ways: one is through threshold comparison, which is similar to the aforementioned method. Figure 5 The methods described in the embodiments are basically the same, and the details are described above. They will not be repeated here. Another method is that after the computer device inputs the first evaluation result to the machine learning monitoring controller, the machine learning monitoring controller first extracts and classifies the performance degradation reasons of multiple cloud applications (such as service failure, sudden fluctuations in traffic, etc.) contained in the first evaluation result; then it maps the classified degradation reasons into actions in the preset action space (such as dynamic resource allocation, service instance expansion or rate limiting, etc., and other types of monitoring actions), where each action corresponds to a specific monitoring decision.

[0096] S602 uses a reward function to filter multiple monitoring decisions and obtains a second evaluation result that includes the target monitoring decision.

[0097] The reward function is a composite reward function constructed based on the target performance degradation estimate, the actual feedback degradation value, and the actual service quality indicators (such as latency and throughput).

[0098] In this embodiment, the computer device uses a preset reward function to filter the action benefits among multiple candidate monitoring decisions (such as resource expansion, traffic limiting, or service degradation). It calculates the action benefit value (i.e., the ratio of performance improvement to execution cost) for each mapped monitoring decision and selects the monitoring decision with the highest benefit value as the target monitoring decision. Finally, the computer device, through the aforementioned... Figures 5-6Either of the two methods described in any implementation yields a second evaluation result that includes the target monitoring decision and the reasons for cloud application performance degradation. Upon receiving the target monitoring decision, the computer immediately executes it. After execution, the computer collects multiple feedback pieces of information for subsequent optimization of the agent model, such as performance degradation indicators of each virtual machine node, overall system service quality satisfaction, and actual interference contribution versus prediction deviation. Subsequently, the computer uses learning algorithms (such as online fine-tuning strategies) and updates the parameters in the agent model (DANN) based on reward signals and multiple feedback pieces of information to ensure that the agent model can adapt to new load changes or unseen application types, achieving adaptive evolution.

[0099] In one exemplary embodiment, such as Figure 7 As shown, the above-mentioned multi-tenant-aware performance monitoring method for public clouds also includes:

[0100] S701 uses various interference generators to apply different types of interference to a benchmark application to obtain sample data.

[0101] The benchmark applications include 12 typical cloud applications, such as network applications, database applications, caching, stream computing, big data, and artificial intelligence inference. The interference includes four types of interference, including LLC pressure, memory bandwidth pressure, disk I / O pressure, and network bandwidth pressure.

[0102] In this embodiment, the computer device pre-integrates various typical interference devices, including LLC pressure, memory bandwidth pressure, disk I / O pressure, and network bandwidth pressure. For each interference device, multiple levels of pressure parameters are set, such as mild (e.g., 30% resource usage), moderate (e.g., 60% resource usage), and severe (e.g., 90% resource usage), with durations ranging from 10 seconds to 10 minutes. This forms a candidate interference pool covering various interference scenarios, including hardware resources, network transmission limitations, storage I / O blocking, and application layer request surges. Subsequently, the computer device sets up 12 typical cloud applications and performs interference tests on benchmark applications using single-type interference modes (e.g., calling a pressure program targeting disk bandwidth alone), multi-type combined interference modes (e.g., running two pressure programs simultaneously, one for memory bandwidth and one for network bandwidth, to simulate the operation of a composite resource bottleneck), or dynamically switching interference modes (e.g., first calling disk bandwidth interference for 10 minutes, then switching to network bandwidth interference to simulate the operation of a scenario switching). Finally, the computer device obtains sample data containing the correspondence between interference type, interference intensity, system state, and application performance.

[0103] S702, the initial agent model is trained based on the sample data to obtain the trained agent model.

[0104] In this embodiment, after obtaining sample data containing the correspondence between interference type, interference intensity, system state, and application performance, the computer device trains an initial agent model in a laboratory environment using the sample data. This allows the initial agent model to learn to recognize different resource contention patterns, ultimately resulting in a trained agent model, such as... Figure 8 As shown.

[0105] In summary, based on all the above embodiments, a method for monitoring multi-tenant-aware performance in public clouds is also provided, such as... Figure 9 As shown, the method includes:

[0106] S801 collects multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications.

[0107] S802, initial modeling is performed based on multiple low-level indicators to obtain a feature set;

[0108] S803, input the feature set into the surrogate model, execute S804-S806; execute S814-S815, train the initial surrogate model, and obtain the trained surrogate model;

[0109] S804, the feature set is input into the feature enhancement unit to obtain the enhanced perturbation features;

[0110] S805, the enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results;

[0111] S806, input the initial evaluation results into the interpretation and fusion unit to obtain the first evaluation result;

[0112] S807, combine the two methods to conduct a degradation assessment based on the first assessment result, obtain the second assessment result, execute S808-S811 to implement the first method; execute S812-S813 to implement the second method;

[0113] S808, compare the target performance degradation estimate with the preset degradation estimation threshold;

[0114] S809, If the target performance degradation estimate is greater than the preset degradation estimation threshold, execute S810; if the target performance degradation estimate is not greater than the preset degradation estimation threshold, execute S811.

[0115] S810: If the target performance degradation estimate is greater than the preset degradation estimate threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0116] S811, if the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring;

[0117] S812 inputs the first evaluation result into the machine learning monitoring controller for degradation evaluation, and obtains multiple monitoring decisions;

[0118] S813 uses a reward function to filter multiple monitoring decisions and obtains a second evaluation result that includes the target monitoring decision;

[0119] S814 uses various types of interference to apply various types of interference to the benchmark application based on multiple interference generators to obtain sample data;

[0120] S815, train the initial agent model based on the sample data to obtain the trained agent model.

[0121] The methods described in each of the above steps have been described in the foregoing embodiments. For details, please refer to the foregoing descriptions. They will not be repeated here.

[0122] In summary, based on all the above embodiments, a method for monitoring multi-tenant-aware performance in public clouds is also provided. The overall flowchart of this method is as follows: Figure 10 As shown, the method includes:

[0123] The computer device collects multiple low-level metrics from the target virtual machine node and its physical host, models these metrics, and generates a feature set. This feature set is then input into a denoising encoder to restore the theoretical baseline of the low-level metrics for each virtual machine node under interference-free conditions, and to enhance the features. The computer device then inputs the enhanced features into a domain-adaptive neural network, outputting the performance degradation ratio caused by the current interference. The SHAP interpreter is then used to calculate the prediction result, outputting the importance values ​​of key features and the main sources of interference. Finally, the computer performs a threshold judgment on the degradation ratio. If the degradation ratio exceeds a preset threshold, an alarm report is generated; otherwise, normal monitoring continues into the next cycle.

[0124] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0125] Based on the same inventive concept, this application also provides a public cloud multi-tenant perceived performance monitoring device for implementing the above-mentioned public cloud multi-tenant perceived performance monitoring method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more public cloud multi-tenant perceived performance monitoring device embodiments provided below can be found in the limitations of the public cloud multi-tenant perceived performance monitoring method above, and will not be repeated here.

[0126] In one exemplary embodiment, such as Figure 11 As shown, a multi-tenant-aware performance monitoring device for public clouds is provided, comprising:

[0127] The data collection module 11 is used to collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications.

[0128] The first evaluation module 12 is used to input multiple low-level indicators into the agent model for performance evaluation and obtain the first evaluation result.

[0129] The second assessment module 13 is used to perform a degradation assessment based on the first assessment result to obtain the second assessment result.

[0130] In one embodiment, the first evaluation module 12 includes:

[0131] The feature enhancement unit is used to input multiple low-level indicators to obtain enhanced perturbation features;

[0132] The transfer learning unit is used to input the enhanced perturbation features to obtain the initial evaluation results; the initial evaluation results include the reasons for the performance degradation caused by multiple current perturbations, as well as the estimated value of the target performance degradation;

[0133] The interpretation fusion unit is used to input the initial evaluation results to obtain the first evaluation result.

[0134] In one embodiment, the feature enhancement unit includes:

[0135] The first feature enhancement subunit is used to perform initial modeling based on multiple low-level metrics to obtain a feature set;

[0136] The second feature enhancement subunit is used to input the feature set into the feature enhancement unit to obtain the enhanced perturbation features.

[0137] In one embodiment, the second evaluation module 13 includes:

[0138] The comparison unit is used to compare the target performance degradation estimate with a preset degradation estimation threshold.

[0139] The first output unit is used to generate and output an alarm report based on the cause of the cloud application's performance degradation if the target performance degradation estimate is greater than the preset degradation estimate threshold.

[0140] The second output unit is used to enter the next monitoring cycle to continue monitoring if the target performance degradation estimate is not greater than the preset degradation estimate threshold.

[0141] In one embodiment, the second evaluation module 13 further includes:

[0142] The evaluation unit is used to input the first evaluation result into the machine learning monitoring controller for degradation evaluation, and obtain multiple monitoring decisions.

[0143] The filtering unit is used to filter multiple monitoring decisions using a reward function to obtain a second evaluation result that includes the target monitoring decision.

[0144] In one embodiment, the first evaluation module 12 further includes:

[0145] The first training unit is used to apply various types of interference to the benchmark application based on multiple interference generators to obtain sample data.

[0146] The second training unit is used to train the initial agent model based on the sample data to obtain the trained agent model.

[0147] Each module in the aforementioned public cloud multi-tenant awareness performance monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can invoke and execute the corresponding operations of each module.

[0148] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0149] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0150] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0151] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0152] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0153] Multiple low-level indicators are input into the feature enhancement unit to obtain the enhanced perturbation features;

[0154] The enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results;

[0155] The initial evaluation results are input into the interpretation and fusion unit to obtain the first evaluation result.

[0156] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0157] Initial modeling is performed based on multiple low-level indicators to obtain a feature set;

[0158] The feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

[0159] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0160] The target performance degradation estimate is compared with the preset degradation estimate threshold;

[0161] If the target performance degradation estimate is greater than the preset degradation estimation threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0162] If the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring.

[0163] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0164] The first evaluation result is input into the machine learning monitoring controller for degradation evaluation, resulting in multiple monitoring decisions.

[0165] By using a reward function to filter multiple monitoring decisions, a second evaluation result is obtained, which includes the target monitoring decision.

[0166] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0167] Various types of interference are applied to the benchmark application using multiple interference generators to obtain sample data.

[0168] The initial agent model is trained based on the sample data to obtain the trained agent model.

[0169] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0170] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0171] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0172] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0173] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0174] Multiple low-level indicators are input into the feature enhancement unit to obtain the enhanced perturbation features;

[0175] The enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results;

[0176] The initial evaluation results are input into the interpretation and fusion unit to obtain the first evaluation result.

[0177] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0178] Initial modeling is performed based on multiple low-level indicators to obtain a feature set;

[0179] The feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

[0180] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0181] The target performance degradation estimate is compared with the preset degradation estimate threshold;

[0182] If the target performance degradation estimate is greater than the preset degradation estimation threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0183] If the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring.

[0184] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0185] The first evaluation result is input into the machine learning monitoring controller for degradation evaluation, resulting in multiple monitoring decisions.

[0186] By using a reward function to filter multiple monitoring decisions, a second evaluation result is obtained, which includes the target monitoring decision.

[0187] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0188] Various types of interference are applied to the benchmark application using multiple interference generators to obtain sample data.

[0189] The initial agent model is trained based on the sample data to obtain the trained agent model.

[0190] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0191] Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications;

[0192] Multiple low-level metrics are input into the proxy model for performance evaluation to obtain the first evaluation result;

[0193] A degradation assessment is conducted based on the first assessment result to obtain the second assessment result.

[0194] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0195] Multiple low-level indicators are input into the feature enhancement unit to obtain the enhanced perturbation features;

[0196] The enhanced perturbation features are input into the transfer learning unit to obtain the initial evaluation results;

[0197] The initial evaluation results are input into the interpretation and fusion unit to obtain the first evaluation result.

[0198] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0199] Initial modeling is performed based on multiple low-level indicators to obtain a feature set;

[0200] The feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

[0201] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0202] The target performance degradation estimate is compared with the preset degradation estimate threshold;

[0203] If the target performance degradation estimate is greater than the preset degradation estimation threshold, an alarm report will be generated and output based on the cause of the cloud application's performance degradation.

[0204] If the target performance degradation estimate is not greater than the preset degradation estimate threshold, then proceed to the next monitoring cycle to continue monitoring.

[0205] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0206] The first evaluation result is input into the machine learning monitoring controller for degradation evaluation, resulting in multiple monitoring decisions.

[0207] By using a reward function to filter multiple monitoring decisions, a second evaluation result is obtained, which includes the target monitoring decision.

[0208] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0209] Various types of interference are applied to the benchmark application using multiple interference generators to obtain sample data.

[0210] The initial agent model is trained based on the sample data to obtain the trained agent model.

[0211] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0212] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0213] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for monitoring the multi-tenant-aware performance of a public cloud, characterized in that, The method includes: Collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications; the low-level metrics include at least one of CPU utilization, hardware counter, host resource status, resource usage characteristics, hardware event vector, interference intensity, and hardware event characteristics; The multiple low-level metrics are input into the proxy model for performance evaluation to obtain a first evaluation result; the first evaluation result includes a target performance degradation estimate, the cause of the cloud application's performance degradation, and the contribution of the cause of degradation; the target performance degradation estimate is used to characterize the degree of degradation of the cloud application; A degradation assessment is performed based on the first assessment result to obtain a second assessment result.

2. The method according to claim 1, characterized in that, The proxy model includes a feature enhancement unit, a transfer learning unit, and an interpretation fusion unit. The step of inputting the low-level metrics into the proxy model for performance evaluation to obtain a first evaluation result includes: The low-level indicators are input into the feature enhancement unit to obtain the enhanced perturbation features; The enhanced perturbation features are input into the transfer learning unit to obtain an initial evaluation result; the initial evaluation result includes the reasons for the performance degradation caused by multiple current perturbations, as well as the estimated value of the target performance degradation; The initial evaluation result is input into the interpretation and fusion unit to obtain the first evaluation result.

3. The method according to claim 2, characterized in that, The step of inputting the plurality of low-level indicators into the feature enhancement unit to obtain the enhanced perturbation features includes: An initial model is performed based on the aforementioned multiple low-level indicators to obtain a feature set; The feature set is input into the feature enhancement unit to obtain the enhanced perturbation features.

4. The method according to any one of claims 1-3, characterized in that, The step of performing a degradation assessment based on the first assessment result to obtain a second assessment result includes: The target performance degradation estimate is compared with a preset degradation estimation threshold; If the target performance degradation estimate is greater than the preset degradation estimation threshold, an alarm report is generated and output based on the cause of the cloud application's performance degradation. If the estimated target performance degradation value is not greater than the preset degradation estimation threshold, then the next monitoring cycle will begin to continue monitoring.

5. The method according to any one of claims 1-3, characterized in that, The step of performing a degradation assessment based on the first assessment result to obtain a second assessment result includes: The first evaluation result is input into the machine learning monitoring controller for degradation evaluation, resulting in multiple monitoring decisions. The multiple monitoring decisions are filtered using a reward function to obtain a second evaluation result that includes the target monitoring decision.

6. The method according to any one of claims 1-4, characterized in that, The method further includes: Various types of interference are applied to the benchmark application using multiple interference generators to obtain sample data. The initial proxy model is trained based on the sample data to obtain the trained proxy model.

7. A multi-tenant-aware performance monitoring device for public clouds, characterized in that, The device includes: The data collection module is used to collect multiple low-level metrics of the target virtual machine node and the host machine of the target virtual machine during the runtime of cloud applications. The first evaluation module is used to input the multiple low-level indicators into the proxy model for performance evaluation and obtain the first evaluation result. The second evaluation module is used to perform a degradation evaluation based on the first evaluation result to obtain a second evaluation result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images