The tenant may select an available area via the available area selection control 306. The available area indicates a deployment location of the instance purchased by the tenant. An instance close to a task location is selected, to reduce a network latency and improve an access speed. Alternatively, the tenant may not select the available area, but select an available area selected by the cloud computing management system 200.

The tenant may specify an instance specification via the instance specification selection module 308. Generally, different instances include different quantities of vCPUs, computing capabilities of vCPUs, and memory space sizes. Some instances may further include other types of processor resources, for example, a GPU and a field programmable gate array (FPGA). The instance specifications may be classified into general-purpose, memory-intensive, and computing-intensive types. The tenant can select the instance specification based on a task characteristic, to better meet the task of the tenant.

It should be noted that each type may include instances of a plurality of specifications. For example, general-purpose instances include instances 1 to n, where n is greater than and equal to 2. The tenant can purchase the instance of an appropriate specification as required. For example, the general-purpose instances include two types of instances: 16 vCPUs and 24 vCPUs. As shown in the general-purpose instances in FIG. 3, an instance 1 includes 16 vCPUs, and an instance 2 includes 24 vCPUs.

In some possible implementations, the tenant may choose, by tapping an accept adjustment control 308A, to accept adjustment performed by the cloud computing management system 200 on the instance specification provided for the tenant. In other words, even if the tenant selects the instance of the specific specification in the instance specification selection module 310, after the tenant chooses to accept the adjustment by the cloud computing management system 200, when the cloud computing management system 200 cannot formulate the instance allocation solution that can meet the resource requirement, the cloud computing management system 200 may adjust the instance specification to make response to the instance allocation request of the tenant. A specific response method is described below.

In some possible implementations, the tenant may select instances of different specifications at the same time, and set a proportion for a quantity of instances. For example, an instance specification proportion setting control (not shown in FIG. 3) may be added to the instance specification selection module 308. The tenant may set an instance specification that is expected to be allocated to the tenant to 50% as the general-purpose instances and 50% as the memory-intensive instances. Further, the tenant may select a specific instance specification under each type of instance specification and set a proportion.

The tenant may set a minimum quantity of synchronous running instances via the minimum quantity of synchronous running instances setting control 310. For some tasks for which at least two instances need to be started to run at the same time, if the cloud computing management system 200 determines that a quantity of instances that can be currently allocated is less than the minimum quantity of synchronous running instances, it is considered that the tasks cannot run, and therefore no instance is allocated to the tenant.

The tenant may set shortest valid running time of the instance via the shortest valid running time setting control 312. Specifically, any continuous running time (from start of the instance to suspension or termination) of the instance allocated by the cloud computing management system 200 to the tenant should be greater than the shortest valid running time. Generally, the cloud service provider may set different priorities for instances of different instance types. For example, a priority of a spot instance is lower than that of an on demand instance. If instance resources are limited, running of the spot instance can be stopped in advance and a computing resource originally supporting running of the spot instance is used to run the on demand instance, to make response to the corresponding instance allocation request.

As mentioned above, when the spot instance is stopped in advance, the task is terminated in advance. For some tasks, if a task is stopped halfway, the entire task fails, for example, training of a neural network model. For some tasks that can be performed segment by segment, if a task is stopped halfway, the task may fail partially. For example, for a rendering task, a minimum unit of rendering is a pixel, and rendering results of pixels do not affect each other. Therefore, when the rendering task is stopped halfway, a rendering result obtained by a pixel that has been rendered can be stored and available.

Therefore, the tenant can set the shortest valid running time to ensure normal running of the task and prevent a failure of the entire task caused by task interruption or early termination. In addition, this avoids a waste of the computing resources.

For example, when the shortest valid running time set by the tenant is 10 minutes, any continuous running duration of a computing instance allocated to the tenant cannot be less than 10 minutes.

In some possible implementations, the cloud computing management system 200 may predict the shortest valid time by using a prediction model based on historical purchase information of the tenant. Specifically, the tenant may use, by tapping a control 312A, a predicted value provided by the system 200.

In some possible implementations, the tenant may set, by tapping a control 312B, whether to allow instance running interruption.

Specifically, when the tenant taps the control 312B, it indicates that the tenant does not allow interruption of the instance allocated to the tenant in a running process. In this case, the instance is more appropriate for running a task that has a high requirement on continuity, such as an online task.

When the tenant does not tap the control 312B, it indicates that the tenant allows interruption of the instance allocated to the tenant in a running process. In this case, the instance is more appropriate for running a task that has a low requirement on continuity, such as an offline task.

The tenant may set, via the deadline setting control 314, a deadline corresponding to the instance allocation request. The deadline indicates latest time at which the cloud computing management system 200 meets the resource requirement uploaded by the tenant. In other words, the resource requirement that is set by the tenant in the control 302 needs to be met and provided for the tenant before the deadline. In addition, the instance allocated to the tenant to meet the resource requirement further needs to meet various parameters set by the tenant in the controls 304 to 312.

The tenant may obtain an estimated expense of this purchase via the estimated expense control 316. The estimated expense is determined by parameters set by the tenant in the controls 302 to 314.

The tenant may complete, by tapping the OK control 318, uploading of the instance allocation request and confirmation of the estimated expense.

The tenant may further configure, on the client, one or more of the following parameters as required: an image used by the instance, a snapshot policy, a security group used by the instance, and the like (not shown in FIG. 3).

It should be noted that, the controls 304 to 314 on the client shown in FIG. 3 are all optional.

After receiving the instance allocation request of the tenant, the cloud computing management system 200 may formulate the instance allocation solution based on a prediction result of an idle state of a computing resource in a future period of time. A predicted value of an idle state of a computing resource indicates idle time and an idle quantity of various types of resources (for example, the processor resource, the network resource, and/or the storage resource) in a future time unit. Specifically, a computing resource allocation part includes S102 to S108.

It should be noted that, in some possible implementations, the cloud computing management system 200 may further receive, through an application programming interface (API), the instance allocation request uploaded by the tenant.

S102: The cloud computing management system 200 predicts an idle state of a computing resource.

After the tenant uploads the instance allocation request in S101, the cloud computing management system 200 may be triggered to predict the idle state of the computing resource in a future period of time. The prediction may be triggered when the tenant taps the OK control 318 or when the tenant completes filling the parameters in the controls 302 to 314.

Optionally, the prediction of the idle state of the computing resource may not be directly triggered by the tenant, but may be performed at a specific period. For example, the prediction is performed every 10 seconds.

Descriptions are provided below by using an example in which a tenant triggers prediction of an idle state of a computing resource.

A predicted idle state of the computing resource may be obtained by using a computing resource prediction model. The computing resource prediction model may be established based on a neural network model. For example, the computing resource prediction model is established based on a back propagation (BP) algorithm or a long short-term memory (LSTM) algorithm. It should be noted that a method for establishing the computing resource prediction model is not limited in this embodiment of this application.

Specifically, an input of the prediction model may be one or more of the following parameters: a computing resource that can be provided by a cloud service provider, an idle computing resource at a current moment, a quantity of idle computing devices at the current moment, a communication speed of an idle computing device, a communication distance of an idle computing device, a quantity of tenants who submit an instance allocation request, a request response rate of each tenant, and the like. Specifically, the parameters used as the input may be selected from parameters related to an output by using a correlation analysis method.

It should be noted that the parameters may be collected and obtained by the cloud computing management system 200.

Before performing the prediction, the computing resource prediction model further needs to be trained. Likewise, an input of model training may be one or more of the following parameters: a computing resource that can be provided by a cloud service provider at a historical moment, an idle computing resource at a historical moment, a quantity of idle computing devices at a historical moment, a communication speed of an idle computing device at a historical moment, a communication distance of an idle computing device at a historical moment, a quantity of tenants who submit an instance allocation request at a historical moment, a request response rate of each tenant at a historical moment, and the like. An output of model training is a computing resource actually used by the tenant that sends the instance allocation request at the historical moment. The computing resource actually used by the tenant may be obtained by the cloud computing management system 200 by collecting historical data.

For example, a tenant once requested to use an on demand instance at 13:00. At least one of a computing resource that can be provided by a server provider, an idle computing resource at 13:00, a quantity of idle computing devices at 13:00, a communication speed of an idle computing device at 13:00, a communication distance of an idle computing device at 13:00, a quantity of tenants who submit an instance allocation request at 13:00, and a request response rate of each tenant at 13:00 may be used as an input of training of the computing resource prediction model. In addition, it is assumed that the tenant terminates running of the on demand instance at 14:00, a computing resource used by the tenant between 13:00 and 14:00 may be used as an output of training of the computing resource prediction model.

FIG. 4A shows a prediction result of an idle state of a computing resource. As shown in FIG. 4A, an idle status of computing resources in a future period of time may be represented by using a matrix. A server is used as an example below.

Specifically, each column of the matrix represents a moment, and each row represents a specification of a server. n indicates n moments included in the future period of time, and m indicates a quantity of server types in prediction. An interval between prediction points and a quantity n of prediction points may be set as required, and the time interval multiplied by n is equal to total prediction duration.

A server has two states: running and not running. Correspondingly, when the server is idle, that is, can be used, a running status of the server is set to 1; otherwise, a running status of the server is set to 0. Considering that there is at least one server of a same specification, each element in the matrix obtained through the prediction represents a quantity of servers of the specification that exist at a moment and that can be used. For example, as shown in FIG. 4B, Sin being equal to 7 indicates that at a moment n, it is predicted that there are seven available servers 1. Likewise, S_mnbeing equal to 4 indicates that at a moment n, it is predicted that there are four available servers m.

It should be noted that the total quantity m of the servers in the prediction may be a quantity of all server specifications that can be provided by the cloud service provider, or may be a quantity of servers of this type determined based on the instance specifications selected by the tenant in S101.

In some possible implementations, a floating price of the spot instance may be predicted. For a specific prediction method, refer to the foregoing prediction method. A difference lies in that one or more of the following parameters are used as an input: a spot instance price at the current moment, a spot instance price at a previous moment, a quantity of tenants who submit an instance allocation request, a request response rate of each tenant, and the like. A type of the input of the prediction model is not limited in this application. Likewise, the parameters used as the input may be selected from parameters related to an output by using a correlation analysis method. The output is a matrix formed by prices of spot instances at moments within a period of time.

Training data of a spot instance price prediction model may also be obtained by the cloud computing management system 200 through collecting historical data.

S103: The cloud computing management system 200 formulates the instance allocation solution based on the instance allocation request and the predicted idle state of the computing resource.

As described above, after triggering the prediction of the idle state of the computing resource, the tenant may obtain the predicted idle state of the computing resource by using the computing resource prediction model. The instance allocation solution may be formulated based on the instance allocation request uploaded by the tenant in S101 and the predicted idle state.

In some possible implementations, the instance allocation solution may be obtained by using a single-objective or multi-objective optimization method based on the instance allocation request uploaded by the tenant in S101 and the prediction result. An optimization objective may be set as required.

For example, the optimization objective may be one or more of the following indicators: a high instance allocation request response rate, low computing resource purchase costs, short average time consumption, and the like. The instance allocation request response rate indicates a probability of a response after the tenant uploads the instance allocation request; the computing resource rental costs indicate costs of purchasing a corresponding computing resource; and the average time consumption indicates average of time spent in completing instance allocation requests of one or more tenants. It should be noted that a type of the optimization objective is not limited in this embodiment of this application.

Optimization methods specifically used in single-objective/multi-objective optimization include optimization methods such as linear programming, a genetic algorithm, and a particle swarm algorithm.

After the optimization objective is determined, an objective function may also be obtained. Further, a constraint condition further needs to be determined. The constraint condition may be a type of condition that ensures normal running of the cloud computing management system. For example, at a moment, the computing resources provided for the tenant should not exceed available computing resources at that moment. In addition, the constraint condition may further include the parameter set by the tenant. For example, running of the instance should meet the uploaded instance allocation request in FIG. 3.

After the objective function and the constraint condition are determined, a single-objective/multi-objective optimization problem can be solved by using an optimization algorithm, so as to obtain the instance allocation solution. Specifically, after the optimization problem is solved, at least the following three cases may occur: no solution, a unique solution, and a non-unique solution. The three solutions respectively correspond to an allocation failure, an instance allocation solution, and a plurality of instance allocation solutions.

S104: The cloud computing management system 200 determines whether there is the instance allocation solution that meets the instance allocation request.

Whether there is the instance allocation solution that meets the instance allocation request of the tenant may be determined based on the solution of the optimization problem obtained in S103. When the solution corresponds to the allocation failure, go to S105. When the instance allocation solution is one or more instance allocation resources, go to S107.

S105: The cloud computing management system 200 generates an adjustment solution.

Specifically, if determining that response fails to be made to the instance allocation request of the tenant, the cloud computing management system 200 provides one or more adjustment solutions for the tenant.

In some possible implementations, the cloud computing management system 200 may notify, on the client, the tenant that the instance allocation request cannot be met. In addition, the tenant may be requested to modify the parameters in the instance allocation request and upload the instance allocation request obtained after modification. That is, S101 is performed again.

In some possible implementations, the cloud computing management system 200 adjusts the solution and may provide one or more adjustment solutions for the tenant, and each adjustment solution includes an adjustment solution for at least one parameter in the instance allocation request. For example, it can be recommendations such as reducing the resource requirement, changing or expanding the instance type, changing or expanding the available area, changing or expanding the instance specification, and postponing the deadline.

In some possible implementations, the parameters in the instance allocation request have a priority sequence. That is, when an adjustment solution is generated, some parameters that are easy to accept adjustment by the tenant, for example, the deadline, are preferably adjusted. After the foregoing type of parameters is adjusted and the adjustment solution is formed, some parameters that are difficult to accept adjustment by the tenant, for example, the available area and the instance specification, may be adjusted, to form another adjustment solution. The priority sequence between the parameters may be set as required.

In some possible implementations, there is also a priority sequence between the one or more adjustment solutions provided by the cloud computing management system 200 for the tenant. As described above, there is the priority sequence between the parameters in the instance allocation request. Therefore, there is also the priority sequence between adjustment solutions formed based on parameter adjustment. Further, a present sequence of the adjustment solutions provided for the tenant to select may be performed according to the priority sequence. For example, an adjustment solution with a higher priority is preferably provided for the tenant. If the tenant is not satisfied with the current adjustment solution, the tenant can be provided with a solution whose priority is lower than the current priority.

Optionally, the adjustment solutions provided for the tenant to select may also be presented in the interaction interface. A quantity of adjustment solutions may be set as required. In other words, a quantity of adjustment solutions presented in the interaction interface may be less than a quantity of adjustment solutions that actually meet the resource requirement of the tenant. The presented adjustment solutions may be selected randomly, or may be selected according to the priority sequence. For example, a priority of a solution obtained based on postponing the deadline may be higher than a priority of an adjustment solution formed based on instance specification adjustment, and the solution obtained based on postponing the deadline is displayed above the interaction interface. A specific priority sequence may be set as required.

The tenant can be notified of the adjustment solution in a pop-up window or a link. FIG. 5 shows a manner of presenting an adjustment solution. As shown in FIG. 5, an adjustment solution interface 400 is displayed on a client page in a pop-up window. The adjustment solution interface 400 includes at least one adjustment solution, an OK control 404, and a Back control 406.

It is assumed that a deadline in an initial instance allocation request of a tenant is set to 11:00. An adjustment solution 1 shown in FIG. 5 is “Postpone the deadline to 11:30”. Further, the adjustment solution may be recommending adjusting two or more parameters at the same time. For example, a solution is “Change the instance specification (from the instance 1 to the instance 2) and postpone the deadline to 11:15”. In addition, the tenant may tap an adjustment solution selection control 402 according to an actual situation to select a feasible adjustment solution, and complete determining of the adjustment solution by tapping the OK control 404. If the tenant cannot accept the adjustment solution provided in the adjustment solution, the tenant may close the adjustment solution interface 400 by tapping the Back control 406.

S106: Whether the cloud computing management system 200 receives acknowledgment information of the adjustment solution is determined.

The tenant may determine whether the adjustment solution provided in S105 can meet the task of the tenant. If the tenant determines that there is at least one adjustment solution that can meet the task of the tenant in the adjustment solution, the tenant may select the feasible adjustment solution by tapping the adjustment selection control 402. In addition, the adjustment solution is determined by tapping the OK control 404, so as to upload the adjustment solution. Correspondingly, the cloud computing management system 200 receives the instance allocation solution acknowledged by the tenant.

Alternatively, the tenant may choose not to accept the adjustment solution, to give up initiating the instance allocation request. Specifically, if the tenant determines that the adjustment solution provided by the cloud computing management system 200 cannot meet the task of the tenant, the tenant may choose not to accept the adjustment solution. For example, the tenant may choose not to accept the adjustment solution by tapping the Back control 406 in FIG. 5.

In some possible implementations, the tenant may complete selection and acknowledgment of the adjustment solution through an API. Correspondingly, the cloud computing management system 200 may receive the acknowledgment information of the tenant for the adjustment solution in an API response manner.

S107: The offline cloud computing management system 200 sends acknowledgment information of the instance allocation solution.

The cloud computing management system 200 may determine the final instance allocation solution based on the instance allocation solution that meets the instance allocation request in S104 or the instance allocation solution that the tenant acknowledges to accept in S106.

Specifically, information about the instance allocation solution that needs to be acknowledged by the tenant may be displayed in the interaction interface 300, or may be displayed on another interaction interface or client. Descriptions are provided below by using another interaction interface 500 as an example.

As shown in FIG. 6, the interaction interface 500 includes at least one or more of the following parameters: an instance type 501, an available area 502, an instance specification 503, a minimum quantity of synchronous running instances 504, shortest valid running time 505, a deadline 506, an estimated expense 507, an OK control 508, and a Back control 509.

The foregoing parameter information presents information included in the instance allocation request uploaded by the tenant in S101, or parameter information obtained through adjustment based on the adjustment solution in S105. For example, when the tenant selects the spot instance and the on demand instance as the instance types in S101, and the instance allocation solution generated in S103 can meet the instance allocation request, or the instance types are not adjusted based on the adjustment solution in S105, the parameter 501 presents the instance types that can be purchased by the tenant: the spot instance and the on demand instance.

In some possible implementations, the instance types may be configured to be in a specific proportional relationship. For example, in the instance allocation solution formulated for the tenant, there are 80% spot instances, and remaining 20% on demand instances. This part of information may also be presented at the parameter 501 (not shown in FIG. 6).

In some possible implementations, the instance specification 503 also includes a plurality of instance specifications. For example, the instance allocation solution formulated for the tenant includes a specific proportion of general-purpose instances 1 and a specific proportion of memory-intensive instances 5. This part of information may also be presented at the instance specification 503 (not shown in FIG. 6).

In addition, the estimated expense of the current instance allocation solution is also presented at the parameter 507. The estimated expense is determined based on one or more of the parameter 501 to the parameter 506.

It should be noted that, as described above, the tenant may configure, on the client based on an actual requirement, one or more of the following parameters in the interaction interface 300 in FIG. 3: the image used by the instance, the snapshot policy, the security group used by the instance, and the like (not shown in FIG. 3). Therefore, the foregoing parameters (not shown in FIG. 6) set by the tenant in FIG. 3 should also be included in FIG. 6.

After acknowledging the current instance allocation solution, the tenant may complete acknowledgment of the instance allocation solution by tapping the control 508. Optionally, the control 509 may also be tapped to return to a previous interface, so as to adjust each parameter again, or give up uploading the instance allocation request.

After the tenant acknowledges the instance allocation solution, the cloud computing management system 200 may allocate the instance based on the solution, so that the tenant can run the task of the tenant on the instance obtained through allocation.

In some possible implementations, the acknowledgment information of the instance allocation solution may also be sent to the tenant through an API, so that the tenant acknowledges the instance allocation solution.

S108: The cloud computing management system 200 determines whether instance start time arrives.

In some possible implementations, after the tenant acknowledges the instance allocation solution, the cloud computing management system 200 may control running of each instance based on start time and stop time of the instance in the instance allocation solution in S107.

Based on the instance allocation solution acknowledged by the tenant in S107, when determining that the instance start time does not arrive, the cloud computing management system 200 does not start the instance. When determining that the instance start time arrives, the cloud computing management system 200 starts the corresponding instance based on the instance allocation solution.

It should be noted that the instance allocation solution provided in this application indicates when an instance of a specific specification is allocated to each tenant and a quantity of instances of each specification. Therefore, the instance is not necessarily started when the tenant acknowledges the instance allocation solution in S107. In addition, the instance allocation solution includes when to start the instance, and also includes when to interrupt or stop the instance.

In some possible implementations, the tenant may monitor when the cloud computing management system allocates the instance to the tenant. For example, the tenant may listen to an instance allocation status via a listening device, an API interface, or the like.

In some possible implementations, the tenant may upload the task of the tenant to the cloud computing management system 200.

The operation of uploading the task of the tenant by the tenant may be performed after the instance allocation request is uploaded (S101), or may be performed before the instance allocation request is uploaded (S101). Optionally, the instance allocation request may further include the task.

S109: The cloud computing management system 200 starts the instance.

In some possible implementations, after detecting, through the API, that the allocated instance has been obtained, the tenant may deploy and run the task of the tenant on the instance.

Optionally, when the tenant has uploaded the task to the cloud computing management system 200, the cloud computing management system 200 deploys and runs the task on the instance allocated to the tenant.

In some possible implementations, after the instance runs, as other tenants continuously upload instance allocation requests to the cloud computing management system 200, a quantity of available instances dynamically changes.

In some possible implementations, the instance may be allocated to and run for the tenant based on the instance allocation solution in S107 and a real-time idle state of the computing resource. The real-time idle state of the computing resource indicates a quantity of idle resources of each type at a current moment. Specifically, in a running process of the instance, an instance resource allocated to the tenant may be temporarily in shortage. As a result, time for completing the task of the tenant may be later than the deadline.

Optionally, when the instance resource allocated to the tenant is temporarily in shortage, the cloud computing management system 200 may allocate an available instance of a higher specification to the tenant without additional charging. This can ensure running of the task of the tenant.

In some possible implementations, the instance allocation solution may be adjusted according to a specific rule. After a new instance allocation request is received, the instance allocation solution of the tenant connected before this time may be modified, to implement optimal instance allocation. The prerequisite may be that a task completion status of the previously connected tenant is not affected.

For example, for a tenant 1, a deadline set in an instance allocation request is 22:00, and the cloud computing management system 200 allocates, to the tenant 1, one instance that can run between 11:00 and 12:00. If the system 200 receives a large quantity of instance allocation requests with an earlier deadline at 22:00, the system 200 may allocate the instance allocated to the tenant 1 to another user between 11:00 and 12:00, and allocate one instance between 14:00 and 15:00 to the tenant, to ensure that a task of the tenant 1 can be completed before the deadline.

In this possible implementation, the system 200 may determine, by calculating a violation probability of the tenant, whether the instance allocation solution of the tenant needs to be updated. The violation probability indicates a probability that the instance provided for the tenant in an updated instance allocation solution cannot meet a computing resource that is still required by the tenant before the deadline. Optionally, the probability may be obtained through prediction by using a violation probability prediction model. The violation probability model may be obtained through training based on historical data.

It should be noted that, for a method for updating the instance allocation solution, refer to the method for obtaining the instance allocation solution in S103.

S110: The cloud computing management system 200 determines whether to trigger an alarm condition.

After starting the instance, the cloud computing management system 200 may generate an alarm for the tenant based on the provided computing resource. Specifically, a remaining resource requirement may be obtained based on the resource requirement set by the tenant in S101 and a used computing resource. When the used computing resource or the remaining resource requirement reaches a threshold, the cloud computing management system 200 may send alarm information to the tenant. The threshold may be set as required.

For example, when the computing resources used by the tenant have reached 80%, that is, when the remaining resource requirement of the tenant is reduced to 20%, the cloud computing management system 200 may send the alarm information to the tenant, to notify the tenant of a service process. Further, the tenant can determine, based on the alarm information, whether to purchase more resources or stop running of the instance in advance.

It should be noted that one or more thresholds may be set for triggering the alarm information. That is, during the running of the task, the alarm information may be triggered once or more times.

When it is detected that the alarm condition is met, go to S112.

S111: The cloud computing management system 200 notifies the tenant of alarm information.

The alarm information sent by the cloud computing management system 200 to the tenant includes at least one of the following parameters: the used computing resource and the remaining computing resource. Optionally, the alarm information may further include at least one of the following parameters: time spent, duration to the deadline, and the like.

Further, the tenant may determine, based on the alarm information, whether the remaining resource requirement can be met before the deadline.

In some possible implementations, when the tenant determines that the remaining resource requirement of the tenant may not be met before the deadline, the tenant may adjust the purchased computing resource. For example, more computing resources may be purchased.

In some possible implementations, on a premise that an agreement is signed with the tenant in advance, the instance allocation request may be actively adjusted for the tenant based on the alarm information. The active adjustment operation may be performed by the cloud computing management system 200.

Optionally, the operation may alternatively be performed by a task management system at a big data layer. The task management system at the big data layer is a system that manages and runs a computing resource.

S112: The cloud computing management system 200 determines whether a running termination condition is met.

During the running of the instance, the running status of the instance is monitored to determine whether the instance meets the running termination condition. The running termination condition may be that execution of the instance allocation solution is completed, the deadline arrives, or the like.

The cloud computing management system 200 determines whether the instance meets the running termination condition, and may monitor a running status of the computing resource at a specific period. Duration of the specific period may be set as required.

It should be noted that there is no specific sequence between S112 and alarm-related steps in S110 and S111. That is, the operation of determining whether the instance meets the running termination condition may occur before or after the alarm operation; or optionally, the operation may be performed synchronously with the alarm operation.

S113: The cloud computing management system releases the instance.

When determining that the running termination condition is met in S112, the cloud computing management system 200 releases the instance. For example, when it is determined in S112 that execution of an instance allocation solution of a tenant has been completed or a deadline has arrived, an instance allocated to the tenant is released.

This application further provides a cloud computing management system 200. As shown in FIG. 7, the cloud computing management system 200 includes an interaction module 202, a storage module 204, and a processing module 206. In addition, the cloud computing management system 200 is deployed on a server. The server may be one or more local servers, or may be one or more cloud servers. Optionally, the server may alternatively be a combination of one or more local servers and one or more cloud servers.

Modules in the cloud computing management system 200 are described below.

The interaction module 202 is configured to receive, in S101, the instance allocation request uploaded by the tenant. The interaction module 202 is further configured to present the adjustment solution generated in S105 to the tenant. The interaction module 202 also receives an acknowledgment status of the tenant for the adjustment solution in S106. The interaction module 202 is further configured to present, in S107, the instance allocation solution in S104 or the instance allocation solution in S105 to the tenant. The interaction module 202 is further configured to receive the task uploaded by the tenant before S112.

The storage module 204 is configured to store the parameters of the instance allocation request received in S101. In the instance allocation method 100, the storage module 204 also stores the training input and the training output that are collected by the cloud computing management system and that are used to train the computing resource prediction model. The storage module 204 is further configured to store the instance allocation solution acknowledged by the tenant in S107. The storage module 204 also stores the alarm condition set by the tenant in S110. The storage module 204 is further configured to store the task uploaded by the tenant to the cloud computing management system before S112.

The processing module 206 is configured to: in the instance allocation method 100, train the computing resource prediction model in S102, and obtain the predicted idle state of the computing resource by using the computing resource prediction model. The processing module 206 is further configured to formulate the instance allocation solution in S103 based on the instance allocation request of the tenant and the predicted idle state of the computing resource. The processing module 206 also generates the adjustment solution in S105. The processing module 206 also determines the instance allocation solution based on a status that the tenant receives the adjustment solution in S106. The processing module 206 is further configured to generate, in S107, the information about the instance allocation solution to be acknowledged by the tenant. The processing module 206 determines, based on the instance allocation solution acknowledged in S107, whether the instance start time arrives in S108, and whether the alarm condition is met in S110. The processing module 206 is further configured to run the instance in S109 and notify the tenant of the alarm information in S111 based on the instance allocation solution. The processing module 206 is further configured to run, on the instance, the stored task uploaded by the user before S112. The processing module 206 determines whether the instance meets the running termination condition in S112. The processing module 206 is further configured to release the instance in S113.

Specifically, the processing module 206 may include a prediction module 208, a scheduling module 210, and an execution module 212.

The prediction module 208 is configured to: in the instance allocation method 100, train the computing resource prediction model in S102, and obtain the predicted idle state of the computing resource by using the computing resource prediction model.

The scheduling module 210 is configured to formulate the instance allocation solution in S103 based on the instance allocation request of the tenant and the predicted idle state of the computing resource. The scheduling module 210 also generates the adjustment solution in S105. The scheduling module 210 also determines, based on whether the tenant receives the adjustment solution in S106, whether to make response to the instance allocation request. The scheduling module 210 is further configured to generate, in S107, the instance allocation solution to be acknowledged by the tenant.

The execution module 212 is configured to: determine, in S107, whether the acknowledged instance allocation solution arrives at the computing resource start time in S108, and determine, in S110, whether the alarm condition is met. The execution module 212 is further configured to run the instance in S109 and notify the tenant of the alarm information in S111 based on the instance allocation solution. The execution module 212 is further configured to run, on the instance, the stored task uploaded by the user before S112. The execution module 212 determines whether the instance meets the running termination condition in S112. The execution module 212 is further configured to release the instance in S113.

This application further provides a computing device 600. As shown in FIG. 8, the computing device includes a bus 602, a processor 604, a memory 606, and a communications interface 608. The processor 604, the memory 606, and the communications interface 608 communicate with each other through the bus 602. It should be understood that quantities of processors and memories in the computing device 600 are not limited in this application. The bus 602 may be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, or the like. Buses may be classified into an address bus, a data bus, a control bus, and the like. To facilitate representation, the bus is represented by using only one line in