CN117215883A - Method and computing device for predicting service quality - Google Patents

Abstract

The method for predicting the service quality provided by the application comprises the following steps: after the first detection data are obtained, the first detection data are output to a target black box index set of the target virtual machine through a classifier, then a target prediction model corresponding to the target black box index set of the target virtual machine is determined, after the second detection data are obtained, the service quality degradation degree of the target virtual machine on the second physical machine is predicted through the target prediction model. The training sample set of the target prediction model is related to the load of the target virtual machine, and the target virtual machine is related to interference sources of the first physical machine and the second physical machine, so that the white box index degradation degree of the target virtual machine transferred to the second physical machine can be predicted more accurately according to the prediction model corresponding to the target black box index. The application also provides a computing device capable of realizing the method.

Description

Method and computing device for predicting service quality

Technical Field

The present application relates to the field of cloud technologies, and in particular, to a method and a computing device for predicting quality of service.

Background

The cloud platform may provide virtual machines to the user. For a user to run an application, task or job on a virtual machine, the cloud platform should guarantee the quality of service of the application, task or job according to service level agreement (service level agreement, SLA) indicators.

There is currently a method for predicting quality of service, which is generally as follows: and performing cache pressure test on the target application program to obtain the corresponding relation between the cache pressure and the service quality of the target application program, wherein the corresponding relation can be represented by a sensitivity curve. And determining the buffer pressure corresponding to the lower limit of the service quality of the target application program according to the sensitivity curve. The quality of service of the target application on the candidate physical machine can be predicted according to the cache pressure of the candidate physical machine and the sensitivity curve of the target application.

The reasons for the degradation of the service quality in the actual environment are complex and various, and the sensitivity curve cannot well represent the interference of the actual environmental factors on the service quality.

Disclosure of Invention

In view of this, the present application provides a method for predicting service quality, which can predict service quality of a virtual machine after migration, and perform virtual machine migration according to the predicted service quality, thereby improving reliability of migrating virtual machines.

The first aspect provides a method for predicting service quality, in the method, after first detection data is acquired, the first detection data is input into a classifier, and a target black box index set of a target virtual machine is output through the classifier; determining a target prediction model corresponding to a target black box index set of the target virtual machine; after the second detection data are acquired, the first detection data and the second detection data are input into a target prediction model; and predicting the degradation degree of the white box index of the target virtual machine on the second physical machine through the target prediction model.

The first detection data comprise target virtual machine data and physical machine data of the first physical machine, the second detection data comprise physical machine data of the second physical machine in a target period, the target virtual machine data comprise resource use data of the target virtual machine and bottom index data of the target virtual machine, the target virtual machine is any virtual machine running on the first physical machine, and the physical machine data comprise resource use data of the physical machine and bottom index data of the physical machine. The target black box index set comprises one or more target black box indexes, wherein the target black box indexes are black box indexes with the correlation degree with the white box indexes of the target virtual machine being larger than or equal to the preset correlation degree. The white box index degradation degree refers to the ratio of degraded white box index data to white box index data without interference.

According to the implementation, the target virtual machine data is related to the load of the target virtual machine, and the physical machine data of the first physical machine is related to the interference source of the target virtual machine in the first physical machine, so that the target black box index set obtained by classification through the classifier is related to both the load and the interference source of the target virtual machine. And the target black box index set has good correlation with the white box index of the target virtual machine, and the service quality of the target virtual machine on the first physical machine can be reflected more accurately. Moreover, the training sample set of the target prediction model is related to the load of the target virtual machine, and the target virtual machine is related to the interference sources of the first physical machine and the second physical machine, so that the white box index degradation degree of the target virtual machine transferred to the second physical machine can be predicted more accurately according to the prediction model corresponding to the target black box index.

In a first possible implementation manner of the first aspect, inputting the first detection data into the classifier includes: and when the first detection data meets the preset interference condition, inputting the first detection data into a classifier. When the virtual machine in the current period has interference, the degradation degree of the white box index of the virtual machine can be automatically predicted.

With reference to the first aspect or the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect, predicting, by the target prediction model, a white-box indicator degradation degree of the target virtual machine on the second physical machine includes: inputting the first detection data into a noise reduction self-encoder, and outputting virtual machine characteristic data through the noise reduction self-encoder; inputting the second detection data into a self-encoder, and outputting second physical machine characteristic data through the self-encoder; combining the virtual machine characteristic data and the second physical machine characteristic data into input data of a target prediction model; and outputting the white box index degradation degree of the target virtual machine on the second physical machine through the target prediction model. The noise reduction self-encoder can reduce the noise of the target virtual machine data, obtain noiseless or low-noise virtual machine characteristic data, and simulate the detection data of the target virtual machine on the second physical machine by combining the noiseless or low-noise virtual machine characteristic data with the second physical machine characteristic data, so that the white box index degradation degree of the target virtual machine on the second physical machine can be predicted. And the self-encoder can reduce the dimension of the physical machine data of the second physical machine and improve the calculation speed.

With reference to the first aspect or the foregoing possible implementation manner of the first aspect, in a third possible implementation manner of the first aspect, when the white-box indicator degradation degree is smaller than a preset amplitude value, migrating the target virtual machine to the second physical machine. Therefore, the degradation degree of the white box index after migration can be automatically predicted, so that a physical machine meeting the service quality requirement is selected for migration, and the migration efficiency and reliability are improved.

With reference to the first aspect or the foregoing possible implementation manner of the first aspect, in a fourth possible implementation manner of the first aspect, a training sample set of an ith virtual machine on the first physical machine is obtained, where the training sample set includes physical machine data of the first physical machine and virtual machine data of the ith virtual machine in a plurality of interference periods, then a plurality of fusion index vectors are determined according to the virtual machine data of the ith virtual machine, at least one target fusion index vector is selected from the plurality of fusion index vectors, and a target black box index set of the ith virtual machine is determined according to the at least one target fusion index vector; and training the classifier according to the training sample sets of the plurality of virtual machines and the target black box index sets of the plurality of virtual machines. The correlation degree of the target fusion index vector and the preset white box index vector is larger than or equal to the preset correlation degree, so that black box indexes related to the white box indexes can be screened out. And taking the target black box index sets of the plurality of virtual machines as training labels, so that the class of the classifier is related to the actual service quality of the virtual machines.

With reference to the first aspect or the foregoing possible implementation manner of the first aspect, in a fifth possible implementation manner of the first aspect, determining a plurality of fusion index vectors according to virtual machine data of an ith virtual machine includes: selecting a plurality of vector pairs from the virtual machine data of the ith virtual machine, calculating the elements of the fusion index vector according to a preset operation rule by all single index data of each vector pair, and forming the elements of the fusion index vector into the fusion index vector. Each vector includes a single index data for a plurality of interference periods. The preset operation rule is division operation, multiplication operation or weighting operation, so that a mode of fusing various indexes is provided.

With reference to the fifth possible implementation manner of the first aspect, in a sixth possible implementation manner of the first aspect, a first training data set and a second training data set are obtained, the first training data set is input into a noise reduction self-encoder, and a virtual machine feature data set is output through the noise reduction self-encoder; inputting the second training data set into the self-encoder, and generating a third training data set according to the virtual machine characteristic data set and the second physical machine characteristic data set through the second physical machine characteristic data set output by the self-encoder; and training a prediction model corresponding to the ith target black box index set according to the third training data set and the preset degradation degree of the white box index. Wherein, white box index degradation degree is training label. The first training data set comprises training data sets of at least one virtual machine and the training data sets of the at least one virtual machine correspond to the ith target black box indicator, and the second training data set comprises physical machine data of a second physical machine of a plurality of interference periods. Each training sample in the third training data set includes one virtual machine feature data and one second physical machine feature data. Thus, a specific method for training the prediction model is provided, and one prediction model can be trained for each target black box index set.

With reference to the first aspect or the foregoing possible implementation manner of the first aspect, in a seventh possible implementation manner of the first aspect, the resource usage data of the target virtual machine includes one or more of an operation duration of a processor in the target virtual machine, a remaining memory of the target virtual machine, a network read byte count, a network write byte count, a block read byte count, and a block write byte count; the bottom index data of the target virtual machine comprises a back-off number and/or a last-stage cache miss number in the target virtual machine; the resource usage data of the physical machine comprises one or more of the number of processing messages per second, the processor utilization rate of the physical machine, the memory utilization rate, the number of packets received per second and the number of packets sent per second; the bottom index data of the physical machine comprises one or more of instruction times per second of the physical machine, memory bandwidth counted in unit time and occupied shared buffer memory size in unit time.

A second aspect provides a computing device that may have functionality to implement the method of predicting quality of service in any one of the embodiments of the first aspect. The functions can be realized by hardware, and can also be realized by executing corresponding software by hardware. The hardware or software includes one or more modules corresponding to the functions described above.

A third aspect provides a computing device comprising a processor and a memory for storing program code; the processor is configured to implement the method of the first aspect by executing program code.

A fourth aspect provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the method of the first aspect.

A fifth aspect provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of the first aspect.

A sixth aspect provides a system on a chip comprising at least one processor coupled to a memory for storing a computer program or instructions for executing the computer program or instructions to implement the method of the first aspect.

Drawings

FIG. 1 is a schematic diagram of an application scenario in an embodiment of the present application;

FIG. 2 is a schematic diagram of a virtual machine according to an embodiment of the present application;

FIG. 3 is a schematic diagram of processor virtualization according to an embodiment of the present application;

FIG. 4 is a schematic diagram of processor virtualization according to an embodiment of the present application;

FIG. 5 is a schematic diagram of memory virtualization according to an embodiment of the present application;

FIG. 6 is a flow chart of training a classifier in an embodiment of the present application;

FIG. 7 is a flow chart of training a predictive model in accordance with an embodiment of the application;

FIG. 8 is a flow chart of a method for predicting quality of service in an embodiment of the present application;

FIG. 9 is a schematic diagram of a classification in an embodiment of the application;

FIG. 10 is a schematic diagram of predicting quality of service in an embodiment of the present application;

FIG. 11 is a block diagram of a computing device in accordance with an embodiment of the present application;

fig. 12 is another structural diagram of a computing device in an embodiment of the present application.

Detailed Description

The method of predicting quality of service of the present application may be applied to infrastructure as a service (infrastructure as a service, iaaS) scenarios. Referring now to fig. 1, an IaaS scenario is described that includes, in one example, a client 11, the internet 12, and a data center 13. In the data center 13, the cloud management platform 131 connects a plurality of servers, such as a server 133, a server 134, and a server 135, through a data center internal network 132. Each server may be considered a physical machine, and multiple virtual machines may be created on each server.

Cloud management platform function: providing an access interface (such as an interface or an API), enabling a tenant to operate a client to remotely Cheng Jieru access the interface to register a cloud account number and a password in a cloud management platform, logging in the cloud management platform, enabling the tenant to pay for selecting and purchasing a virtual machine with a specific specification (a processor, a memory and a disk) in the cloud management platform after the cloud account number and the password are successfully authenticated by the cloud management platform, enabling the cloud management platform to provide a remote login account number password of the purchased virtual machine after the payment purchase is successful, enabling the client to remotely log in the virtual machine, and installing and running an application of the tenant in the virtual machine.

Cloud management platform logic function division: user console, computing management service, network management service, storage management service, authentication service, and image management service. The user console provides interfaces or APIs to interact with tenants, the computing management service is used for managing servers running virtual machines and containers and bare metal servers, the network management service is used for managing network services (such as gateways, firewalls and the like), the storage management service is used for managing storage services (such as data bucket services), the authentication service is used for managing account passwords of tenants, and the mirror image management service is used for managing virtual machine mirrors.

Cloud management platform client function: and receiving a control plane command sent by the cloud management platform, creating the control plane command on a server, and carrying out full life cycle management on the virtual machine.

A tenant may create, manage, log in, and operate virtual machines in the data center 13 through the cloud management platform 131. Virtual machines may also be referred to as cloud servers (elastic compute service, ECS) or elastic instances.

Virtualization techniques include computing virtualization, I/O virtualization, and the like. Referring to fig. 2, the virtualization technology is used as a core technology of a cloud scenario, and a virtual machine is used as granularity to share one physical server for a plurality of tenants, so that the tenants can conveniently and flexibly use physical resources on the premise of safety isolation, and the utilization rate of the physical resources can be greatly improved.

Computing virtualization is the provision of computing resources such as processors and memory of servers to virtual instances, which may be virtual machines, in some scenarios, containers, bare metal servers, etc.

(1) Processor virtualization

The server running the virtual machine is typically provided with a plurality of physical CPUs, wherein the physical CPUs are processors in a hardware layer and the vCPU is a virtual processor in a software layer, also called hyper-threading. The virtual machine cannot perceive the physical CPU, but only the vCPU that the virtual machine manager presents to the virtual machine.

Fig. 3 is an architecture example in which the server is provided with 4 physical CPUs, CPU0, CPU1, CPU2, and CPU3, respectively, each physical CPU including 4 CPU cores (cores), the number of hyper-threads of each CPU core being 2. Thus, the virtual machine manager may provide the virtual machine with the total number of vcpus=the total number of hyper-threads=the number of physical CPUs (also referred to as socket number) x the number of CPU cores per physical CPU band x the number of hyper-threads supported per CPU core=32. It is to be noted that this is only an example, and the present embodiment does not limit the number of physical CPUs set by the server, the number of CPU cores contained in the physical CPUs, and the number of hyper-threads on each CPU core.

As shown in fig. 4, 1 hyper-thread may be provided to the virtual machine as 1 VCPU, in other implementations, 1 hyper-thread may also be provided to the virtual machine as a plurality of VCPUs in a time-division multiplexing manner, depending on the CPU service quality (quality of service, qoS) of the virtual machine, in some practical applications, a tenant may pay more for the cloud management platform to purchase a virtual machine with a higher CPU QoS, where 1 hyper-thread may be provided to the virtual machine as 1 VCPU.

(2) Memory virtualization

The purpose of memory virtualization technology is to provide virtual machines with a contiguous physical memory space starting at 0 addresses, effectively isolating and scheduling memory resources between virtual machines.

Memory virtualization techniques mainly involve translation of guest virtual addresses (Guest Virtual Address, GVA) — guest physical addresses (Guest Physical Address, GPA) — host virtual addresses (Guest Virtual Address, GVA) — host physical addresses (Host Physical Address, HPA).

In the virtualization technology, a plurality of virtual machines are often run on a physical host, where each virtual machine is configured to monopolize a memory space of the physical host, so that the virtual machine uses a guest physical address (Guest Physical Address, GPA) to represent a memory space owned by the virtual machine, where the memory space is considered to be contiguous by the virtual machine (i.e., it is understood that the virtual machine considers itself to possess a completed physical memory bank).

The GVA is an address formed by mapping GPA by an operating system of the virtual machine, the operating system of the virtual machine provides the GVA for a process or application software arranged on the operating system of the virtual machine to use, the operating system of the virtual machine records the mapping relation between the GVA and the GPA, and the conversion from the GVA to the GPA is realized by a page table of the operating system of the virtual machine.

HPA is the actual physical memory address, HVA is the address formed by mapping HPA by the operating system of host machine, the operating system of host machine provides HVA to the process (e.g. virtual machine) on the operating system for use, the operating system of host machine records the mapping relation of HVA to HPA, the conversion of HVA to HPA is realized by the page table of the operating system of host machine.

In fig. 5, virtual machine 1 and virtual machine 2 are disposed in the same server (hereinafter referred to as host machine), and the virtual machine manager of the host machine sets the GPA address range of virtual machine 1 to 0-5GB, which corresponds to the HPA address ranges of 1.5GB-4.5GB and 6.5GB-8.5GB on the physical memory. And, the host's virtual machine manager sets the GPA address range of virtual machine 2 to 0-4GB, corresponding to HPA address ranges 9GB-11GB and 13GB-15GB on physical memory. Therefore, virtual machine 1 exclusively occupies a GPA address range of 0-5GB and virtual machine 2 exclusively occupies a GPA address range of 0-4 GB. The GPA address range of 0-5GB and the GPA address range of 0-4GB can both correspond to different HPA address ranges on the physical memory, thereby realizing the isolation of the virtual machine memory.

The GPA address range is related to the above virtual machine specification, for example, a tenant may set a virtual machine 1 with a memory size of 5G in the virtual machine specification in the cloud management platform, and at this time, the virtual machine manager is notified by the cloud management platform that a virtual machine 1 with GPA of 0-5G needs to be created.

(3) I/O virtualization

I/O virtualization refers to simulating a corresponding device for each virtual machine by means of software or hardware-assisted virtualization. For example, each virtual machine considers that it owns a complete disk device. In effect, the VMM creates a file on the physical hard disk for each virtual machine or divides a region as the "physical hard disk" of the virtual machine; when the client operating system accesses the "physical hard disk," the Hypervisor will translate the hard disk number to an offset from the file or region and return the result to the client operating system. I/O virtualization faces three basic tasks of device discovery, access interception, and device simulation.

A) Device discovery

Hypervisor needs to provide a way for device discovery so that the client operating system can discover virtual devices and load the corresponding drivers. The manner in which devices are discovered depends on the type of device being virtualized, including both bus enumeration and non-enumeration. Enumeration refers to the reading of information about device types, communication modes, etc. by the operating system accessing various configuration spaces.

B) Access interception

After the client operating system finds the corresponding virtual device through the device discovery, the client operating system sends out an I/O request according to the interface resource of the device. The VMM intercepts I/O requests of the guest operating system in different ways according to the different ways of device access to provide the device emulation phase with the corresponding device functions.

C) Device simulation

After receiving the I/O request from the virtual machine, the Hypervisor needs to simulate the function of the physical device, so that the virtual machine accesses the virtual device to achieve the effect of accessing the real physical device. The function of the virtual device is emulated by the Hypervisor, but when to emulate which device is determined by the physical device, the policy of the Hypervisor, and the needs of the client operating system.

The following terms of the application are presented:

the quality of service index includes white box index and black box index. The white box index refers to a performance index when a user uses an application program, a processing task, or a job of the virtual machine. White box indicators include, but are not limited to, the number of messages processed per second (transaction persecond, TPS) and latency (latency).

The black box index is an index acquired in a black box scene. The black box index includes a resource usage index and a bottom layer index. The resource usage index includes, but is not limited to, CPU running time, remaining memory size, network read-write byte count, block read-write byte count. The underlying metrics include, but are not limited to, the number of single cycle instructions (instruction per cycle, IPC), the number of cycles of a single instruction (cycle per instruction, CPI), the number of cache misses per thousand instructions (misses per kiloinstructions, MPKI), and the number of instruction rollbacks.

Virtual machine phase change (VM phase change) refers to a change in resource usage behavior of a virtual machine. For example, virtual machines are disk intensive when starting to run, and then become computationally intensive. The load change of the virtual machine may be considered a virtual machine phase change.

The interferer (source of interference, SOI) refers to a shared resource where interference is present. Such as Last Level Cache (LLC), memory bandwidth, network bandwidth, etc. When the processor chip includes L1, L2, and L3, L3 is LLC. When the processor chip includes L1 and L2, L2 is LLC.

Because of the limited shared resources, virtual machines contend for the shared resources and may generate interference. In a scenario where multiple virtual machines are co-located, interference-aware quality of service prediction (interference aware QoS prediction) is a technical problem to be solved by the present application. In the existing prediction method, the sensitivity curve only can represent the corresponding relation between the cache pressure and the service quality of the target application program, and is inaccurate in predicting the service quality. In this regard, the application provides a method for predicting service quality, which can obtain a black box index set related to white box indexes through a classifier, and then predict the service quality degradation degree of a virtual machine deployed on a new physical machine according to a prediction model corresponding to the black box index set.

First, a process of training a classifier in the method for predicting quality of service of the present application will be described, referring to fig. 6, in one embodiment, the method for predicting quality of service of the present application includes:

step 601, a training sample set of an ith virtual machine on a first physical machine is obtained.

The first training sample set includes physical machine data of a first physical machine in a plurality of interference periods and virtual machine data of an ith virtual machine. The physical machine data of the first physical machine and the virtual machine data of the ith virtual machine in each interference period are taken as a training sample. The physical machine data of the first physical machine includes resource usage data and underlying index data of the first physical machine. The virtual machine data includes resource usage data and underlying index data for the virtual machine.

For example, the physical machine data of the first physical machine is denoted as P, and the resource usage data of the first physical machine is denoted as R _PM The bottom index data of the first physical machine is marked as H _PM 。R _PM Resource usage data including a plurality of interference periods, e.g. the resource usage data of the jth period is noted asH _PM The bottom layer index data including a plurality of interference periods, e.g., the bottom layer index data of the jth period is denoted +.>i and j are both positive integers.

The resource usage data of the physical machine includes one or more of a number of messages processed per second, a processor utilization of the physical machine, a memory utilization, a number of packets received per second, and a number of packets sent per second. The underlying index data of the physical machine includes one or more of a number of instructions per second of the physical machine, a memory bandwidth per unit time, and a shared cache size occupied per unit time. It should be understood that the resource usage data of the physical machine and the underlying index data of the physical machine may also select other indexes according to actual situations, which is not limited by the present application.

Resource usage data of an ith virtual machine on a first physical machine is recorded asThe bottom index data of the ith virtual machine is marked as +.> Resource usage data including a plurality of periods, e.g., resource usage data of the jth period is recorded as +.> The underlying index data including a plurality of time periods, e.g. the underlying index data of the jth time period is noted as

The resource usage data of the virtual machine includes one or more of a processor running time of the virtual machine, a remaining memory of the virtual machine, a network read byte count, a network write byte count, a block read byte count, and a block write byte count. The underlying index data of the virtual machine includes an instruction rollback number and/or a last level cache miss number of the virtual machine.

Step 602, determining a plurality of fusion index vectors according to virtual machine data of the ith virtual machine.

Optionally, step 602 includes: selecting a plurality of vector pairs from the virtual machine data of the ith virtual machine, calculating the elements of the fusion index vector according to a preset operation rule by all single index data of each vector pair, and forming the elements of the fusion index vector into the fusion index vector. The preset operation rule is division operation, multiplication operation or weighting operation.

In one example, virtual machine data of an ith virtual machine in a plurality of interference periods includes 3 vectors, each vector including processor run length, remaining memory size, and MPKI for the plurality of interference periods.

The vector corresponding to the processor run length is denoted as [ Ct ] ₁ ,Ct ₂ ,...,Ct _n ]The vector corresponding to the remaining memory size is denoted as [ m ] ₁ ,m ₂ ,...,m _n ]The vector corresponding to MPKI is denoted as [ MPKI ] ₁ ,MPKI ₂ ,...,MPKI _n ]. The fusion index vector obtained by dividing all the single index data in the three vectors can be:

step 603, selecting at least one target fusion index vector from the multiple fusion index vectors, where the correlation between the target fusion index vector and the preset white box index vector is greater than or equal to the preset correlation.

The preset white box index vector comprises white box index data measured in a plurality of interference time periods.

And respectively carrying out pearson correlation operation on the plurality of fusion index vectors and the white box index vector, and determining the ith fusion index vector as a target fusion index vector when the correlation degree between the ith fusion index vector and the white box index vector is greater than or equal to the preset correlation degree. And when the correlation degree between the ith fusion index vector and the white box index vector is smaller than the preset correlation degree, determining that the ith fusion index vector is not the target fusion index vector. The number of target fusion index vectors may be one or more.

Step 604, determining a target black box index set of the ith virtual machine according to the at least one target fusion index vector.

The target black box index set comprises two indexes corresponding to the target fusion index vector. For example, the target fusion index vector isThe target black box index set includes the remaining memory size and MPKI.

Step 605, training a classifier according to the training sample sets of the plurality of virtual machines and the target black box index sets of the plurality of virtual machines, wherein the target black box indexes of the plurality of virtual machines are training labels. In some cases, the training sample sets of several virtual machines have the same set of target black box indicators. Methods of training the classifier include, but are not limited to, xgboost.

The embodiment can screen black box indexes with high correlation degree with white box indexes, and can reflect the influence of the actual environment on the white box indexes more accurately than the existing cache pressure indexes.

In an alternative embodiment, before determining the plurality of fusion index vectors according to virtual machine data of the ith virtual machine, the method further includes: selecting two vectors from the virtual machine data of the ith virtual machine, performing pearson correlation operation on the two vectors to obtain the similarity of the two vectors, and removing any one of the two vectors when the similarity of the two vectors is greater than or equal to the preset similarity. This reduces the similarity of vectors in the training sample set.

In another alternative embodiment, before determining a plurality of fusion index vectors according to virtual machine data of an ith virtual machine, selecting two vectors from the virtual machine data of the ith virtual machine, performing pearson correlation operation on the two vectors to obtain similarity of the two vectors, when the similarity of the two vectors is greater than or equal to a preset similarity, obtaining the similarity of the first vector and other vectors in the virtual machine data of the ith virtual machine as a first group of similarity, obtaining the similarity of the second vector and other vectors in the virtual machine data of the ith virtual machine as a second group of similarity, and when the sum of the similarity of the first group is greater than the sum of the similarity of the second group, removing the second vector. And when the sum of the second group of similarity is larger than the sum of the first group of similarity, removing the first vector. This provides another way to reduce the similarity of vectors in a training sample set.

Referring to fig. 7, in another embodiment, the method for predicting quality of service according to the present application includes:

step 701, acquiring a first training data set and a second training data set.

The first training data set corresponds to an ith target black box index set, which includes training data sets of one or more virtual machines corresponding to the ith target black box index set. The second training data set includes physical machine data of a second physical machine in a plurality of interference periods.

Step 702, inputting the first training data set into a noise reduction self-encoder, and outputting the virtual machine characteristic data set through the noise reduction self-encoder.

The noise reduction self-encoder is used for reducing noise of training samples of the first training data set to obtain virtual machine characteristic data. The virtual machine feature data set includes a plurality of virtual machine feature data.

Step 703, inputting the second training data set into the self-encoder, and outputting the second physical machine feature data set through the self-encoder. The second physical machine characteristic data set includes a plurality of second physical machine characteristic data. The self-encoder may reduce the dimension such that the second physical machine characteristic data occupies less memory space than the physical machine data of the second physical machine.

Step 704, generating a third training data set according to the virtual machine feature data set and the second physical machine feature data set, wherein each training sample in the third training data set comprises one virtual machine feature data and one second physical machine feature data.

And step 705, training a prediction model corresponding to the ith target black box index set according to the third training data set and the preset white box index degradation degree.

The white box indicator degradation level may be a TPS degradation level or a latency degradation level. The degree of deterioration of the white box index is measured in advance in a test environment of a plurality of interference periods.

In this embodiment, the plurality of virtual machine training sample sets, the target black box index set and the prediction model have a corresponding relationship, so that the input of the prediction model is related to the load and the interference of the virtual machine, and thus the interference of the actual environment can be reflected more accurately than the existing cache pressure index, and a more accurate prediction result can be obtained.

Second, since the virtual machine feature data may represent noise reduced virtual machine data, the second physical machine feature data may represent physical machine data of a second physical machine having interference, and thus the training sample including the virtual machine feature data and the second physical machine feature data may represent virtual machine data when the second physical machine has an interference source. The target prediction model trained according to the third training data set and the white-box index degradation degree can be used for predicting the white-box index degradation degree of the virtual machine on the second physical machine, and the white-box index degradation degree can reflect the service quality of the virtual machine.

Based on the classifier and the prediction model, the method and the device can predict the service quality of any virtual machine. Referring now to fig. 8, in one embodiment, a method for predicting quality of service according to the present application includes:

step 801, acquiring first detection data, where the first detection data includes target virtual machine data and physical machine data of a first physical machine in a current period.

The target virtual machine data includes resource usage data of the target virtual machine and underlying index data of the target virtual machine. The physical machine data of the first physical machine includes resource usage data of the first physical machine and underlying index data of the first physical machine. The current period may be any period. The target virtual machine is any virtual machine running on the first physical machine.

Step 802, inputting the first detection data into a classifier, and outputting a target black box index set of the target virtual machine through the classifier.

The class to which the classifier corresponds includes a plurality of target black box index sets, each of which represents a class. And when the first detection data meets the preset interference condition, inputting the first detection data into a classifier.

Optionally, when the first detection data meets a preset interference condition, the first detection data is input into the classifier. Specifically, whether interference exists or not can be judged according to physical machine data of the first physical machine or target virtual machine data in the first detection data. For example, when the processor utilization rate of the first physical machine is detected to be greater than or equal to the preset utilization rate in the current period, it is determined that the first detection data satisfies the preset interference condition. For example, when the current period detects that the last level of cache miss number is greater than or equal to the preset miss number, it is determined that the first detection data satisfies the preset interference condition.

It should be noted that, the present application may also determine whether the target virtual machine is interfered in the current period according to other data in the target virtual machine data. And/or, the application can also judge whether the target virtual machine is interfered in the current period according to other data in the physical machine data of the first physical machine. When a virtual machine is disturbed on a first physical machine, the quality of service of the virtual machine migrating to a second physical machine can be automatically predicted.

Step 803, determining a target prediction model corresponding to the target black box index set of the target virtual machine.

Step 804, obtaining second detection data, where the second detection data includes physical machine data of the second physical machine.

The target period refers to a period in which the virtual machine migrates to the second physical machine. The physical machine data of the second physical machine includes resource usage data of the second physical machine and underlying index data of the second physical machine.

Step 805, inputting the first detection data and the second detection data into the target prediction model.

And step 806, predicting the white box index degradation degree of the target virtual machine on the second physical machine through the target prediction model.

Optionally, step 806 includes: inputting the first detection data into a noise reduction self-encoder; inputting the second detection data from the encoder; combining the virtual machine feature data output from the encoder with the second physical machine feature data output from the encoder to form input data of a target prediction model; and outputting the white box index degradation degree of the target virtual machine on the second physical machine through the target prediction model.

In this embodiment, the target virtual machine data is related to the load of the target virtual machine, the physical machine data of the first physical machine is related to the interference source of the target virtual machine, and after the target virtual machine data and the physical machine data of the first physical machine are classified by the classifier, the obtained target black box index is related to both the load of the target virtual machine and the interference source, and the target black box index has a good correlation with the white box index of the target virtual machine, so that the service quality of the target virtual machine on the first physical machine can be reflected more accurately.

Because the training sample set of the target prediction model is related to the load of the target virtual machine, the target virtual machine is related to the interference sources of the first physical machine and the second physical machine, and therefore the service quality of the target virtual machine migrated to the second physical machine can be predicted more accurately according to the prediction model (namely the target prediction model) corresponding to the target black box index.

Referring to fig. 9 and 10, the following description describes a classification process and a process for predicting quality of service, and referring to fig. 9, 8 classes corresponding to the classifier are: target black box index set 1-target black box index set 8. Taking a training sample set including the CPU utilization and IPC of the first physical machine in a plurality of interference periods, the running time of a processor of the target virtual machine, the residual memory size and the MPKI as an example, a target black box index set corresponding to the training sample set takes a target black box index set 4 as an example, and the target black box index set 4 includes the residual memory size and the MPKI of the virtual machine.

When the target virtual machine in the current period has interference, classifying the CPU utilization rate and IPC of the first physical machine in the current period, the running time of a processor of the target virtual machine in the current period, the residual memory size and MPKI by an xgboost classifier to obtain a target black box index set of the target virtual machine as a target black box index set 4.

Referring to fig. 10, the target prediction model corresponds to the target black box index 4. And inputting the running time of the processor, the residual memory size and the MPKI of the target virtual machine, the CPU utilization rate and IPC of the first physical machine into a noise reduction self-encoder, and outputting the virtual machine characteristic data by the noise reduction self-encoder. The CPU utilization and IPC of the second physical machine are input to the self-encoder, and the self-encoder outputs the characteristic data of the second physical machine. And synthesizing the virtual machine characteristic data and the second physical machine characteristic data into input data of a target prediction model, and outputting the degradation degree of the white box index by the target prediction model.

The present application provides a computing device capable of implementing the method of the embodiments shown in fig. 6 through 8. Referring to fig. 11, a computing device 1100 of the present application includes:

an obtaining unit 1101, configured to obtain first detection data, where the first detection data includes target virtual machine data and physical machine data of a first physical machine in a current period, the target virtual machine data includes resource usage data of a target virtual machine and bottom indicator data of the target virtual machine, the target virtual machine is any virtual machine running on the first physical machine, and the physical machine data of the physical machine includes resource usage data of the physical machine and bottom indicator data of the physical machine;

The classification unit 1102 is configured to input the first detection data into a classifier, and output a target black box index set of the target virtual machine through the classifier;

a processing unit 1103, configured to determine a target prediction model corresponding to a target black box index set of the target virtual machine;

the obtaining unit 1101 is further configured to obtain second detection data, where the second detection data includes physical machine data of the second physical machine in the target period;

a prediction unit 1104 for inputting the first detection data and the second detection data into a target prediction model, and predicting the white-box index degradation degree of the target virtual machine on the second physical machine through the target prediction model.

In an alternative embodiment, the classification unit 1102 is specifically configured to input the first detection data into the classifier when the first detection data meets a preset interference condition.

In another alternative embodiment, the prediction unit 1104 is specifically configured to input the first detection data into a noise reduction self-encoder, and output the virtual machine feature data through the noise reduction self-encoder; inputting the second detection data into a self-encoder, and outputting second physical machine characteristic data through the self-encoder; combining the virtual machine characteristic data and the second physical machine characteristic data into input data of a target prediction model; and outputting the white box index degradation degree of the target virtual machine on the second physical machine through the target prediction model.

In another alternative embodiment, the computing apparatus 1100 further includes a migration unit configured to migrate the target virtual machine to the second physical machine when the white-box indicator degradation level is less than a preset magnitude.

In another optional embodiment, the obtaining unit 1101 is further configured to obtain a training sample set of an ith virtual machine on the first physical machine, where the training sample set includes physical machine data of the first physical machine and virtual machine data of the ith virtual machine in a plurality of interference periods;

the computing device 1100 further includes:

the index fusion unit is used for determining a plurality of fusion index vectors according to the virtual machine data of the ith virtual machine;

the index screening unit is used for selecting at least one target fusion index vector from a plurality of fusion index vectors, and the correlation degree between the target fusion index vector and a preset white box index vector is greater than or equal to the preset correlation degree; determining a target black box index set of the ith virtual machine according to the at least one target fusion index vector;

the training classifier unit is used for training the classifier according to training sample sets of the plurality of virtual machines and target black box index sets of the plurality of virtual machines, wherein the target black box index sets of the plurality of virtual machines are training labels.

Optionally, the index fusion unit is specifically configured to select a plurality of vector pairs from virtual machine data of the ith virtual machine, where each vector includes single index data of a plurality of interference periods; and calculating the elements of the fusion index vector according to all the single index data of each vector pair according to a preset operation rule, and forming the fusion index vector by the elements of the fusion index vector, wherein the preset operation rule is division operation, multiplication operation or weighting operation.

In an alternative embodiment of the present invention in the form of a further alternative embodiment,

the obtaining unit 1101 is further configured to obtain a first training data set and a second training data set, where the first training data set includes a training data set of at least one virtual machine, and the training data set of the at least one virtual machine corresponds to an i-th target black box index set, and the second training data set includes physical machine data of a second physical machine in a plurality of interference periods;

the computing device 1100 further includes:

the noise reduction self-encoding unit is used for inputting the first training data set into a noise reduction self-encoder and outputting a virtual machine characteristic data set through the noise reduction self-encoder;

the self-encoding unit is used for inputting the second training data set into the self-encoder and outputting a second physical machine characteristic data set through the self-encoder;

The processing unit 1103 is further configured to generate a third training data set according to the virtual machine feature data set and the second physical machine feature data set, where each training sample in the third training data set includes one virtual machine feature data and one second physical machine feature data; and training a prediction model corresponding to the ith target black box index set according to the third training data set and the preset degradation degree of the white box index.

In a further alternative embodiment of the present invention,

the resource use data of the target virtual machine comprises one or more of the operation time of a processor in the target virtual machine, the residual memory of the target virtual machine, the network read byte number, the network write byte number, the block read byte number and the block write byte number;

the bottom index data of the target virtual machine comprises a back-off number and/or a last-stage cache miss number in the target virtual machine;

the resource usage data of the first physical machine comprises one or more of a processing message number per second, a processor utilization rate of the physical machine, a memory utilization rate, a packet receiving number per second and a packet sending number per second;

the bottom layer index data of the first physical machine comprises one or more of instruction times per second of the physical machine, memory bandwidth counted in unit time and occupied shared buffer size in unit time.

Referring now to FIG. 12, another embodiment of a computing device 1200 of the present application includes: a processor 1201, a memory 1202 and a network interface 1203 connected via a bus 1204.

In this embodiment, the memory 1202 is used for storing program codes. The processor 1201 is configured to perform the method of predicting quality of service in the embodiments illustrated in fig. 6-8 by invoking program code stored in the memory 1202.

It should be appreciated that the processor 1201 mentioned in this embodiment may be a central processing unit (central processing unit, CPU), but may also be other general purpose processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), off-the-shelf programmable gate arrays (field programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

It should also be appreciated that the memory 1202 referred to in embodiments of the present application may be either volatile memory or nonvolatile memory, or include both volatile and nonvolatile memory. The nonvolatile memory may be a read-only memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an electrically Erasable EPROM (EEPROM), or a flash memory. The volatile memory may be random access memory (random access memory, RAM) which acts as an external cache. By way of example, and not limitation, many forms of RAM are available, such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous DRAM (SLDRAM), and direct memory bus RAM (DR RAM).

The network interface 1203 may be used to receive information or to send information. The information may be, but is not limited to, a message, a data message, or an instruction.

It should be noted that, because the content of information interaction and execution process between the modules/units of the above-mentioned device is based on the same concept as the method embodiment of the present application, the technical effects brought by the content are the same as the method embodiment of the present application, and the specific content can be referred to the description in the foregoing illustrated method embodiment of the present application, which is not repeated herein.

The present application provides a computer readable storage medium having a computer program stored therein, which when run on a computer causes the computer to perform the method of predicting quality of service in the above-described or alternative embodiments.

The application also provides a method comprising a computer program product which, when run on a computer, causes the computer to perform the method of predicting quality of service in the embodiments or alternative embodiments as described above.

The application also provides a chip system, wherein the chip system comprises a processor and a memory which are mutually coupled. The memory is used for storing a computer program or instructions, and the processing unit is used for executing the computer program or instructions stored in the memory, so that the computer executes the steps executed by the computing device in the above embodiments. Alternatively, the memory is an on-chip memory, such as a register, a cache, etc., and the memory may be an off-chip memory located in a site, such as a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (random access memory, RAM), etc. The processor referred to in any of the foregoing may be a general purpose central processing unit, a microprocessor, an application specific integrated circuit (application specific integrated circuit, ASIC) or one or more integrated circuits for implementing the method of predicting quality of service described above.

It should be noted that the above-described embodiment of the apparatus is only illustrative, and the units described as separate units may or may not be physically separated, and the units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the embodiment of the device provided by the application, the connection relation between the modules represents that the modules have communication connection, and can be specifically implemented as one or more communication buses or signal lines.

From the above description of the embodiments, it will be apparent to those skilled in the art that the present application may be implemented by means of software plus necessary general purpose hardware, or of course by means of special purpose hardware including application specific integrated circuits, special purpose CPUs, special purpose memories, special purpose components, etc. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions can be varied, such as analog circuits, digital circuits, or dedicated circuits. However, a software program implementation is a preferred embodiment for many more of the cases of the present application. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a readable storage medium, such as a floppy disk, a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk or an optical disk of a computer, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method of the embodiments of the present application.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions in accordance with embodiments of the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL), or wireless (e.g., infrared, wireless, microwave, etc.) means, the computer-readable storage medium may be any available medium that can be stored by the computer or a data storage device such as a server, data center, etc., that contains an integration of one or more available media.

Claims (18)

1. A method of predicting quality of service, comprising:

acquiring first detection data, wherein the first detection data comprises target virtual machine data and physical machine data of a first physical machine in a current period, the target virtual machine data comprises resource use data of a target virtual machine and bottom index data of the target virtual machine, the target virtual machine is any virtual machine running on the first physical machine, and the physical machine data comprises the resource use data of the physical machine and the bottom index data of the physical machine;

inputting the first detection data into a classifier, and outputting a target black box index set of the target virtual machine through the classifier;

determining a target prediction model corresponding to a target black box index set of the target virtual machine;

acquiring second detection data, wherein the second detection data comprises physical machine data of a second physical machine in a target period;

inputting the first detection data and the second detection data into a target prediction model;

and predicting the degradation degree of the white box index of the target virtual machine on the second physical machine through the target prediction model.

2. The method of claim 1, wherein the inputting the first detection data into a classifier comprises:

And inputting the first detection data into a classifier when the first detection data meets a preset interference condition.

3. The method of claim 1, wherein predicting, by the target prediction model, a white-box indicator degradation level of a target virtual machine on the second physical machine comprises:

inputting the first detection data into a noise reduction self-encoder, and outputting virtual machine characteristic data through the noise reduction self-encoder;

inputting the second detection data into a self-encoder, and outputting second physical machine characteristic data through the self-encoder;

combining the virtual machine feature data and the second physical machine feature data into input data of a target prediction model;

and outputting the degradation degree of the white box index of the target virtual machine on the second physical machine through the target prediction model.

4. The method according to claim 1, wherein the method further comprises:

and when the degradation degree of the white box index is smaller than a preset amplitude value, migrating the target virtual machine to the second physical machine.

5. The method according to any one of claims 1 to 4, further comprising:

acquiring a training sample set of an ith virtual machine on a first physical machine, wherein the training sample set comprises physical machine data of the first physical machine and virtual machine data of the ith virtual machine in a plurality of interference periods;

Determining a plurality of fusion index vectors according to the virtual machine data of the ith virtual machine;

selecting at least one target fusion index vector from the fusion index vectors, wherein the correlation degree between the target fusion index vector and a preset white box index vector is greater than or equal to a preset correlation degree;

determining a target black box index set of the ith virtual machine according to the at least one target fusion index vector;

training a classifier according to training sample sets of a plurality of virtual machines and target black box index sets of the plurality of virtual machines, wherein the target black box index sets of the plurality of virtual machines are training labels.

6. The method of claim 5, wherein determining a plurality of fusion metrics vectors from virtual machine data of an ith virtual machine comprises:

selecting a plurality of vector pairs from the virtual machine data of the ith virtual machine, wherein each vector comprises single index data of a plurality of interference periods;

calculating the elements of the fusion index vector according to a preset operation rule, wherein the preset operation rule is division operation, multiplication operation or weighting operation;

and forming elements of the fusion index vector into the fusion index vector.

7. The method of claim 5, wherein the method further comprises:

acquiring a first training data set and a second training data set, wherein the first training data set comprises training data sets of at least one virtual machine, the training data sets of the at least one virtual machine correspond to an ith target black box index set, and the second training data set comprises physical machine data of a second physical machine in a plurality of interference time periods;

inputting a first training data set into a noise reduction self-encoder, and outputting a virtual machine characteristic data set through the noise reduction self-encoder;

inputting a second training data set into a self-encoder, and outputting a second physical machine characteristic data set through the self-encoder;

generating a third training data set according to the virtual machine characteristic data set and the second physical machine characteristic data set, wherein each training sample in the third training data set comprises virtual machine characteristic data and second physical machine characteristic data;

training a prediction model corresponding to the ith target black box index set according to the third training data set and the preset white box index degradation degree.

8. The method according to any one of claim 1 to 4, wherein,

The resource use data of the target virtual machine comprises one or more of the operation time of a processor in the target virtual machine, the residual memory of the target virtual machine, the network read byte number, the network write byte number, the block read byte number and the block write byte number;

the bottom index data of the target virtual machine comprises an instruction backspacing number and/or a last-stage cache miss number in the target virtual machine;

the resource use data of the physical machine comprises one or more of the number of processing messages per second, the processor utilization rate of the physical machine, the memory utilization rate, the number of packets received per second and the number of packets sent per second;

the bottom index data of the physical machine comprises one or more of instruction times per second of the physical machine, memory bandwidth counted in unit time and occupied shared buffer memory size in unit time.

9. A computing device, comprising:

the device comprises an acquisition unit, a storage unit and a storage unit, wherein the acquisition unit is used for acquiring first detection data, the first detection data comprises target virtual machine data and physical machine data of a first physical machine in a current period, the target virtual machine data comprises resource use data of a target virtual machine and bottom index data of the target virtual machine, the target virtual machine is any virtual machine running on the first physical machine, and the physical machine data comprises the resource use data of the physical machine and the bottom index data of the physical machine;

The classification unit is used for inputting the first detection data into a classifier, and outputting a target black box index set of the target virtual machine through the classifier;

the processing unit is used for determining a target prediction model corresponding to a target black box index set of the target virtual machine;

the acquisition unit is further used for acquiring second detection data, wherein the second detection data comprise physical machine data of a second physical machine in a target period;

and the prediction unit is used for inputting the first detection data and the second detection data into a target prediction model, and predicting the white box index degradation degree of the target virtual machine on the second physical machine through the target prediction model.

10. The computing device of claim 9, wherein the computing device is configured to,

the classification unit is specifically configured to input the first detection data into a classifier when the first detection data meets a preset interference condition.

11. The computing device of claim 9, wherein the prediction unit is specifically configured to input the first detection data into a noise-reducing self-encoder, and output virtual machine feature data through the noise-reducing self-encoder; inputting the second detection data into a self-encoder, and outputting second physical machine characteristic data through the self-encoder; combining the virtual machine feature data and the second physical machine feature data into input data of a target prediction model; and outputting the degradation degree of the white box index of the target virtual machine on the second physical machine through the target prediction model.

12. The computing device of claim 9, further comprising a migration unit to migrate the target virtual machine to the second physical machine when the white-box indicator degradation level is less than a preset magnitude.

13. The computing device of any one of claim 9 to 12, wherein,

the acquisition unit is further used for acquiring a training sample set of an ith virtual machine on the first physical machine, wherein the training sample set comprises physical machine data of the first physical machine and virtual machine data of the ith virtual machine in a plurality of interference periods;

the computing device further includes:

an index fusion unit, configured to determine a plurality of fusion index vectors according to virtual machine data of the ith virtual machine;

the index screening unit is used for selecting at least one target fusion index vector from the fusion index vectors, and the correlation between the target fusion index vector and a preset white box index vector is greater than or equal to the preset correlation; determining a target black box index set of the ith virtual machine according to the at least one target fusion index vector;

the training classifier unit is used for training the classifier according to training sample sets of the plurality of virtual machines and target black box index sets of the plurality of virtual machines, wherein the target black box index sets of the plurality of virtual machines are training labels.

14. The computing device of claim 13, wherein the computing device is configured to,

the index fusion unit is specifically configured to select a plurality of vector pairs from virtual machine data of an ith virtual machine, where each vector includes single index data of a plurality of interference periods; and calculating the elements of the fusion index vector according to all the single index data of each vector pair according to a preset operation rule, and forming the fusion index vector by the elements of the fusion index vector, wherein the preset operation rule is division operation, multiplication operation or weighting operation.

15. The computing device of claim 13, wherein the computing device is configured to,

the acquiring unit is further configured to acquire a first training data set and a second training data set, where the first training data set includes a training data set of at least one virtual machine, the training data set of at least one virtual machine corresponds to an i-th target black box index set, and the second training data set includes physical machine data of a second physical machine in a plurality of interference periods;

the noise reduction self-encoding unit is used for inputting the first training data set into a noise reduction self-encoder and outputting a virtual machine characteristic data set through the noise reduction self-encoder;

A self-encoding unit for inputting the second training data set into a self-encoder, and outputting a second physical machine characteristic data set through the self-encoder;

the processing unit is further configured to generate a third training data set according to the virtual machine feature data set and the second physical machine feature data set, where each training sample in the third training data set includes one virtual machine feature data and one second physical machine feature data; and training a prediction model corresponding to the ith target black box index set according to the third training data set and the preset white box index degradation degree.

16. The computing device of any one of claim 9 to 12, wherein,

the resource use data of the target virtual machine comprises one or more of the operation time of a processor in the target virtual machine, the residual memory of the target virtual machine, the network read byte number, the network write byte number, the block read byte number and the block write byte number;

the bottom index data of the target virtual machine comprises an instruction backspacing number and/or a last-stage cache miss number in the target virtual machine;

the resource use data of the physical machine comprises one or more of the number of processing messages per second, the processor utilization rate of the physical machine, the memory utilization rate, the number of packets received per second and the number of packets sent per second;

The bottom index data of the physical machine comprises one or more of instruction times per second of the physical machine, memory bandwidth counted in unit time and occupied shared buffer memory size in unit time.

17. A computing device comprising a processor and a memory;

the memory is used for storing program codes;

the processor is configured to invoke program code in the memory to cause the processor to perform the method of any of claims 1 to 8.

18. A computer readable storage medium storing instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1 to 8.