CN116204314A - Resource scheduling method, device and storage medium - Google Patents

Abstract

The application provides a resource scheduling method, equipment and a storage medium, wherein the method comprises the following steps: determining actual computing power performance values corresponding to the heterogeneous physical processors respectively, and determining reference computing power performance values according to the actual computing power performance values corresponding to the physical processors respectively; and determining the occupation amount of the computing power resources of the target virtual equipment instance running on the target physical processor to the target physical processor according to the reference computing power performance value. The computing power performance value reached by the computing power resource occupation amount corresponding to the target virtual equipment instance consumed by the target physical processor does not exceed the reference computing power performance value, so that the consistency of the upper limit computing power of each virtual equipment instance is realized. And each heterogeneous physical processor is brought into a uniform resource pool to schedule, so that the virtual equipment instance of the user can run on any physical processor in the resource pool, and the flexible supply of resources is realized.

Description

Resource scheduling method, device and storage medium

Technical Field

The present invention relates to the field of cloud computing technologies, and in particular, to a resource scheduling method, device, and storage medium.

Background

With the continuous update of central processing units (Central Processing Unit, abbreviated as CPUs) produced by different hardware manufacturers, physical servers introduced by cloud service providers are also continuously changing, and one of the main performances of the continuous change is that the physical CPUs are continuously changing. In this way, the cloud service provider side forms multiple server cluster islands. The server cluster island refers to a product having a plurality of physical server clusters, wherein physical CPUs adopted in the same cluster are isomorphic, and physical CPUs adopted in different clusters are heterogeneous, such as products produced by different hardware manufacturers or different generations produced by the same hardware manufacturer.

Based on the existence of the physical server cluster island, in the conventional scheme, corresponding virtual device instance clusters are constructed corresponding to each physical server cluster, and the virtual device instance is, for example, a cloud server (i.e., a virtual server), a container or the like. That is, the virtual device instance is built using the CPU resources in a certain physical server cluster. The physical server clusters are isolated, so that the coupling between the virtual equipment instance and the physical CPU on the cloud is too high, and the elastic calling capability of the resource is poor. Moreover, the performances of different physical CPUs are obviously different, and the physical server cluster island ensures that the computing power of the virtual equipment instance of the user cannot be unified, so that the user experience is poor.

Disclosure of Invention

The embodiment of the invention provides a resource scheduling method, equipment and a storage medium, which can improve the flexibility of resource calling and ensure the computational consistency of virtual equipment instances of users.

In a first aspect, an embodiment of the present invention provides a resource scheduling method, where the method includes:

determining actual computing power performance values corresponding to each of the heterogeneous physical processors;

determining a reference computing power performance value according to the actual computing power performance values corresponding to the physical processors respectively;

And determining the computing power resource occupation amount of a target virtual device instance running on a target physical processor to the target physical processor according to the reference computing power performance value, wherein the computing power performance value reached by the computing power resource occupation amount corresponding to the target physical processor consumed by the target virtual device instance does not exceed the reference computing power performance value, and the target physical processor is any one of the physical processors.

In a second aspect, an embodiment of the present invention provides a resource scheduling apparatus, where the apparatus includes:

the determining module is used for determining actual computing power performance values corresponding to the heterogeneous physical processors respectively, and determining reference computing power performance values according to the actual computing power performance values corresponding to the physical processors respectively;

and the scheduling module is used for determining the occupation amount of the computing power resources of the target physical processor by the target virtual device instance running on the target physical processor according to the reference computing power performance value, wherein the computing power performance value reached by the occupation amount of the computing power resources corresponding to the target physical device instance consumed by the target physical processor does not exceed the reference computing power performance value, and the target physical processor is any one of the physical processors.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, a communication interface; wherein the memory has executable code stored thereon which, when executed by the processor, causes the processor to perform the resource scheduling method of the first aspect.

In a fourth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to at least implement a resource scheduling method as described in the first aspect.

In the resource scheduling scheme provided by the embodiment of the invention, heterogeneous physical processors in each physical server maintained by the cloud service provider are brought into a unified resource pool to be scheduled, so that a virtual equipment instance of a user can run on any physical processor in the resource pool, and the flexible supply of resources is realized. Moreover, in order to provide the user with a virtual device instance that masks the hardware differences of the underlying physical processors, it is necessary to ensure that each virtual device instance has a uniform upper limit of computing power, that is, no matter what physical processor is used to provide the user with the virtual device instance, the virtual device instance has a uniform upper limit of computing power, so as to ensure the user experience, and not to obtain a virtual device instance with poor performance due to poor computing power performance of the adopted physical processor.

Therefore, in the embodiment of the invention, the actual computing performance of each physical processor is tested in advance to obtain the actual computing performance value of each physical processor, and then a reference computing performance value is determined according to the actual computing performance value of each physical processor. And determining the computing power resource occupation amount of the target virtual equipment instance to the target physical processor according to the reference computing power performance value aiming at any target virtual equipment instance running on any target physical processor. The computing power performance value obtained by the target virtual device instance does not exceed the reference computing power performance value, that is, the computing power performance value obtained by the target physical device instance consuming the computing power resource occupation amount corresponding to the target virtual device instance does not exceed the reference computing power performance, so that the consistency of the upper limit computing power of each virtual device instance is realized.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic diagram of a resource scheduling system according to an embodiment of the present invention;

FIG. 2 is a flowchart of a resource scheduling method according to an embodiment of the present invention;

FIG. 3 is a flowchart of a method for scheduling resources according to an embodiment of the present invention;

FIG. 4 is a flowchart of a resource scheduling method according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a virtual device instance migration scenario provided in an embodiment of the present invention;

FIG. 6 is a flowchart of a method for scheduling resources according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of a resource scheduling device according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram of an electronic device according to the present embodiment.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention. In addition, the sequence of steps in the method embodiments described below is only an example and is not strictly limited.

The resource scheduling method provided by the embodiment of the invention can be applied to the resource scheduling system shown in fig. 1, and the resource scheduling system can comprise a resource scheduling service cluster and a plurality of heterogeneous physical processors.

The resource scheduling service cluster may include a plurality of cloud servers that provide resource scheduling services, which are referred to as resource scheduling servers in fig. 1.

Among the heterogeneous plurality of physical processors are, for example, physical processor 1, physical processor 2, … and physical processor N illustrated in fig. 1. Heterogeneous physical processors may be different models of physical processors produced by different hardware vendors. These heterogeneous physical processors act as scheduled hardware resources. In practice, these physical processors may be deployed in different physical servers (or physical hosts), one or more of which may be deployed in each physical server.

In the embodiment of the invention, the physical processor may be a CPU, or may be other types of processors. Taking a CPU as an example, in practice, a physical CPU chip may include a plurality of CPU cores, and in this embodiment of the present invention, the physical processor may be, for example, each physical processor core, for example, a CPU core.

In the embodiment of the invention, a large number of heterogeneous physical processors are incorporated into a unified resource pool, and unified management and scheduling are carried out by a resource scheduling service cluster. The method specifically includes the steps of creating a required virtual device instance for each user by using physical processors contained in a resource pool, and dynamically adjusting the physical processors occupied in the running process of the virtual device instance and the computing power resource quantity of the occupied physical processors according to application characteristic information of the virtual device instance. Therefore, the method is favorable for fully realizing the elasticity of resource calling, realizing the elastic supply and scheduling of the user virtual equipment instance, and also utilizing the optimized resource utilization rate. Examples of virtual devices may be, for example, virtual machines, containers, elastic computing servers (i.e., cloud servers), and so forth. Virtual device instances are created by virtualization techniques, and when virtual device instances are created, they are allocated relevant physical resources, including but not limited to physical processors, so that one or more virtual device instances may be run on the physical processors. And the application program of the user can run in the virtual device instance, so the application characteristic information refers to the characteristic information of the application program running in the virtual device instance, such as load, delay sensitivity and other characteristics. The load may be considered as the size of the data to be processed generated by the application program, and the delay sensitivity refers to the processing response delay of the data generated by the application program, and may be simply considered as the delay sensitivity for distinguishing whether the application program provides an online service or an offline service, where the delay sensitivity of the online service is higher, and the offline service generally has no obvious sensitivity to the delay.

The following describes an execution process of the resource scheduling method provided by the embodiment of the present invention.

Fig. 2 is a flowchart of a method for scheduling resources according to an embodiment of the present invention, where the method may be performed by the above-mentioned resource scheduling service cluster, and as shown in fig. 2, the method includes the following steps:

201. actual computational power performance values corresponding to each of the heterogeneous plurality of physical processors are determined.

202. And determining a reference computing power performance value according to the actual computing power performance value corresponding to each of the plurality of physical processors.

203. And determining the occupation amount of the computing power resources of the target physical processor by the target virtual device instance running on the target physical processor according to the reference computing power performance value, wherein the computing power performance value reached by the occupation amount of the computing power resources corresponding to the target virtual device instance consumed by the target physical processor does not exceed the reference computing power performance value.

Wherein the target physical processor is any one of the plurality of physical processors.

In the embodiment of the invention, in the case that the cloud service provider brings a plurality of heterogeneous physical processors into a unified resource pool for management and scheduling, each virtual device instance can use the computational power resource provided by any physical processor. A problem arises: if a virtual device instance obtains better performance by using a physical processor with better computing power performance, and the virtual device instance obtains poor performance by using a physical processor with poorer computing power performance, then the user corresponding to the virtual device instance has poor user experience.

Therefore, in the case of incorporating several heterogeneous physical processors into a unified resource pool for management and scheduling, it is necessary to ensure that each virtual device instance has a unified upper limit of computing power, that is, to ensure the consistency of the upper limit of computing power of each virtual device instance. In short, the upper limit of computational power performance that can be achieved by a virtual device instance is consistent regardless of the physical processor used.

The upper limit of the computing power is uniform, which means that the computing power performance which can be obtained by only restricting the virtual device instance does not exceed a certain threshold, that is, if the actual application load of a certain virtual device instance is low, the actually required computing power resource is also less, and when the load is high, the required computing power resource is increased, but the upper limit to which the required computing power resource can be increased is limited.

In order to achieve the purpose, first, a calculation performance test is performed on each physical processor to determine an actual calculation performance value of each physical processor, then, a reference calculation performance value is determined according to the actual calculation performance value of each physical processor, and the reference calculation performance value is used as a reference to restrict the occupation amount of the calculation resources of each physical processor by a virtual device instance running on each physical processor. In this embodiment, determining the computing power resource occupation amount of the target virtual device instance running on the target physical processor to the target physical processor is actually determining the upper limit or the value range of the computing power resource occupation amount of the target virtual device instance to the target physical processor.

In the embodiment of the invention, a physical processor is taken as a CPU as an example. In practice, the computational resources provided by the CPU are actually time slices.

When each physical processor is subjected to the calculation performance test, certain test software can be adopted to enable each physical processor to process the same task so as to obtain the performance score of each physical processor as a calculation performance value.

For ease of understanding, the following simplified illustration is made. Taking physical processor 1 and physical processor 2 as an example, assuming that the time slice length is set to 2 milliseconds, assuming that the same task processing is completed, physical processor 1 uses K1 time slices in total, and physical processor 2 uses K2 time slices, K2> K1, and then physical processor 1 performs better than physical processor 2. Assuming that the performance score of the physical processor 2 is determined to be 80 by the set score calculation rule, the performance score of the physical processor 1 is 100, and the performance scores of the other physical processors are all higher than 80, this performance score 80 of the physical processor 2 will be used as a reference calculation force performance value.

Based on the above example, it can be simply considered that the same task is completed with a shorter time for a physical processor with high computational performance than a physical processor with poor computational performance.

It is assumed here that virtual device instances have been deployed on heterogeneous physical processors, and in order to unify the upper limits of computing power of the virtual device instances, for any physical processor (referred to as a target physical processor for convenience of description) it is necessary to determine, according to the above-mentioned reference computing power performance value, the computing power resource occupation amount of the target physical processor by a target virtual device instance running on the target physical processor.

Specifically, the computing power resource occupation amount of the target virtual device instance on the target physical processor needs to enable the computing power performance value obtained by the target virtual device instance not to exceed the reference computing power performance value. That is, the computing power performance value that can be achieved by the target physical processor consuming the computing power resource occupation amount corresponding to the target virtual device instance does not exceed the reference computing power performance value.

Assuming that the actual computing power performance value of the target physical processor is 100 minutes and is 80 minutes higher than the reference computing power performance value, a certain target virtual device instance running on the target physical processor can occupy at most computing power resources of the target physical processor, which can reach the computing power performance value of 80 minutes. For example, assuming that one time slice is 2 ms and one scheduling period is set to be 1 second, 500 time slices are included in one scheduling period, assuming that the target physical processor works for 500 time slices to achieve a calculated performance value of 100 minutes in each scheduling period, and assuming that the target physical processor works for 400 (assuming 80% of time slices) time slices to achieve a calculated performance value of 80 minutes in each scheduling period, the calculated resource occupation amount of the target virtual device instance for the target physical processor is set to be the upper limit of 80% of the total calculated resource that the target physical processor can provide, that is, 80% of the time slices of the target physical processor are occupied by each scheduling period. That is, when the actual occupied resource amount of the target virtual device instance to the target physical processor reaches the 80% computing power resource amount, the computing power performance value obtained by the target virtual device instance is equal to 80 minutes of the reference computing power performance value; that is, the calculated performance value that can be achieved when the target physical processor consumes 80% of the calculated resource occupation amount corresponding to the target virtual device instance is 80 minutes of the reference calculated performance value. It will be appreciated that the range of the amount of resources that the target virtual device instance may actually use on the target physical processor at this time is: [0 ] the amount of computational resources that the target physical processor can provide 80% ], the specific computational resource occupancy dynamically changing according to the application load of the target virtual device instance.

By the scheme, the upper limit of the computing power performance which can be obtained by each virtual device instance is unified by taking the reference computing power performance value of the physical processor as a reference. Therefore, under the condition that a plurality of heterogeneous physical processors maintained by a cloud service provider are subjected to unified pooling management, the unified upper limit of the computing power of the virtual equipment instance can be realized, the user can not perceive the inconsistent performance of the virtual equipment instance due to the hardware difference of the underlying physical processors, and the user experience is ensured.

In practical applications, the use of physical processors by virtual device instances is divided into two modes of use, one being exclusive and one being shared. The term "exclusive" refers to a physical processor that is used by only one virtual device instance, and the term "shared" refers to a physical processor that is commonly used by at least two virtual device instances.

For the two usage modes, the ways of ensuring the consistency of the upper limit of the computing power of the virtual equipment instance are different.

First case: the target virtual device instance monopolizes the target physical processor. At this time, the target computing power resource amount required to be consumed by the target physical processor to reach the reference computing power performance value may be determined, so that the computing power resource occupation amount of the target virtual device instance on the target physical processor is determined to take the target computing power resource amount as an upper limit. It may be understood that if the actual computing power performance value of the target physical processor is greater than the reference computing power performance value, it means that the target computing power resource amount required by the target physical processor to reach the reference computing power performance value is smaller than the computing power resource amount that the target physical processor can provide, and then the remaining computing power resource amount corresponds to a time slice, so that the physical processor can be put into a sleep state.

For example, one scheduling period is 1 second, and one time slice is 2 milliseconds, then 500 time slices are included in one scheduling period, and the upper limit of the computing power resource occupation of the target virtual device instance to the target physical processor is assumed to be: each scheduling cycle occupies 80% of the time slice of the target physical processor (assuming that the above-mentioned reference computational power performance value of 80 minutes can be reached at this time), then the amount of remaining computational power resources of each scheduling cycle of the target physical processor is at least 20% of the time slice, since the target physical processor is now exclusively owned by the target virtual device instance, which is dormant for at least 20% of the time slice length.

In the first case, it is actually the computational compression of the high performance physical processor. By setting the operating frequency of the physical processors (e.g., only 80% of the time slices in one scheduling period in the above example), the power calculation performance actually exerted by each physical processor is kept within an acceptable interval: below the reference calculated performance value.

Second case: at least two virtual device instances share a target physical processor. At this time, the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor respectively may be determined according to the current application feature information of the at least two virtual device instances. And the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource amount which can be provided by the target physical processor. The computing power resource occupation amount of any virtual equipment instance in the at least two virtual equipment instances to the target physical processor enables the computing power performance value which can be obtained by any virtual equipment instance not to exceed the reference computing power performance value.

The at least two virtual device instances respectively correspond to the sum of the occupation amounts of the computing power resources of the target physical processor and the total amount of the computing power resources which can be provided by the target physical processor, which means that in the second case, the computing power performance of the target physical processor does not need to be suppressed, the target physical processor can exert all the computing power performance, but the upper limit of the computing power performance which can be obtained by any virtual device instance sharing the target physical processor is limited by slicing time slices. The target physical processor is shared by a plurality of virtual device instances, so that the computing power resource of the target physical processor can be fully utilized, and the resource utilization rate is improved.

For example, assuming that virtual device instance a and virtual device instance b share a target physical processor, one scheduling period is 1 second and one time slice is 2 milliseconds long, 500 time slices are included in one scheduling period. And assuming that the upper limit of the computing power resource occupation amount of the virtual equipment instance a and the virtual equipment instance b to the target physical processor, which are determined according to the reference computing power performance value of 80 minutes, is: each scheduling cycle occupies 80% of the time slices of the target physical processor. Then assuming that the amount of computing power resources of the virtual device instance a to the target physical processor is 60% at this time, the remaining 40% of the amount of computing power resources may be allocated to the virtual device instance b. I.e., the total amount of computing power resources of the target physical processor is shared by virtual device instance a and virtual device instance b, wherein no upper limit that any virtual device instance can occupy must exceed the amount of computing power resources corresponding to the reference computing power performance value.

Specifically, the computing power resource occupation amounts of the virtual device instance a and the virtual device instance b, which are actually allocated, may be determined according to the application feature information corresponding to each of the virtual device instance a and the virtual device instance b. If, for example, the demand of the virtual device instance with high load is preferentially guaranteed, and the load of the virtual device instance a is assumed to be higher than that of the virtual device instance b, the occupation amount of the computing power resources of the virtual device instance a can be determined according to the load demand of the virtual device instance a, and the remaining computing power resources are occupied by the virtual device instance b. For another example, if the requirement of the virtual device instance running the online application is preferentially guaranteed, and if the online application is running in the virtual device instance a and the offline application is running in the virtual device instance b, the computing power resource occupation amount of the virtual device instance a can be determined according to the load requirement of the virtual device instance a, and the remaining computing power resources are occupied by the virtual device instance b.

Fig. 3 is a flowchart of a method for scheduling resources according to an embodiment of the present invention, where the method may be performed by the above-mentioned resource scheduling service cluster, and as shown in fig. 3, the method includes the following steps:

301. and acquiring a first resource pool and a second resource pool which contain heterogeneous physical processors, wherein each physical processor in the first resource pool supports the sharing use of the virtual equipment instance, and each physical processor in the second resource pool supports the sharing use of the virtual equipment instance.

302. And responding to the creation request of the target virtual equipment instance, and determining a target resource pool used for creating the target virtual equipment instance according to the application characteristic information of the target virtual equipment instance, wherein the target resource pool is a first resource pool or a second resource pool.

303. A target physical processor is determined in the target resource pool to create a target virtual device instance based on the target physical processor.

304. And determining a reference computing power performance value according to the corresponding actual computing power performance value of each physical processor.

305. And determining the occupation amount of the computing power resources of the target physical processor by the target virtual device instance according to the reference computing power performance value, wherein the computing power performance value reached by the consumption of the computing power resources corresponding to the target virtual device instance by the target physical processor does not exceed the reference computing power performance value.

In practical applications, in order to improve the resource utilization rate of the physical processor, at least some physical processors may be used by setting a super-selling ratio to generate more virtual processors, for example, setting the super-selling ratio to be 2 for a certain physical processor, then 2 virtual processors may be generated based on the physical processor, and one virtual processor is allocated to one virtual device instance for use, so as to implement shared use of the physical processor by two virtual device instances.

In view of this, in practice, the physical processors may be divided into exclusive use and shared use according to the usage pattern, the physical processors being exclusive use, the super-selling ratio being 1, the physical processors being shared use, the super-selling ratio being greater than 1. Thus, several heterogeneous physical processors of a cloud service provider may be divided into two resource pools according to the usage pattern of the physical processors: and the first resource pool supports the exclusive use of the virtual equipment instance, and the second resource pool supports the shared use of the virtual equipment instance, namely the overstock ratio of the physical processor in the first resource pool is 1, and the overstock ratio of the physical processor in the second resource pool is greater than 1.

When any target virtual device instance needs to be created, determining whether a target resource pool used for creating the target virtual device instance is a first resource pool or a second resource pool according to application characteristic information of the target virtual device instance. Wherein, the application characteristic information includes: delay sensitivity and/or loading.

In practical applications, the staff on the cloud service provider side may determine application feature information of the target virtual device instance, that is, feature information of an application program to be run, such as whether the application program is an online application or an offline application, a load level, and so on, based on communication with a user corresponding to the target virtual device instance. Online class applications include, for example, online games, live broadcasts, etc., and offline class applications include, for example, websites, backup databases, etc.

The above-described on-line/off-line class application is only one example of the delay sensitivity of an application program, and in practice, an acceptable time range of the application program for response delay may be input.

The resource scheduling service cluster side can be provided with respective corresponding thresholds for delay sensitivity and load in advance, and the application programs can be divided into a plurality of application characteristic types of delay sensitivity and delay insensitivity, high load and low load based on the thresholds. And setting the corresponding relation between different application characteristic types and the resource pool. Such as: a delay sensitive class application and a high load class application, using a first resource pool; delay insensitive class applications and low load class applications use a second resource pool.

Based on the comparison result of the application feature information of the target virtual device instance and the corresponding threshold value, the application feature type corresponding to the virtual device instance can be determined, further a target resource pool used for creating the target virtual device instance is determined, and a target physical processor used for creating the target virtual device instance is determined in the target resource pool.

It may be appreciated that if the target physical processor is located in the first resource pool, the target virtual device instance monopolizes the target physical processor; if the target physical processor is located in the second resource pool, the target virtual device instance may share the target physical processor with other virtual device instances.

In practical applications, assuming that it is determined to create a target virtual device instance using the first resource pool, an idle physical processor may be determined from the first resource pool as a target physical processor, and the target virtual device instance may be created based on the target physical processor. If there is no free physical processor in the first resource pool, optionally, application characteristic information of the target virtual device instance and each other virtual device instance occupying the physical processor in the first resource pool may be compared to determine whether there is a virtual device instance X capable of being migrated to the second resource pool, and if so, migrating the virtual device instance X to the second resource pool, and determining the physical processor occupied by the virtual device instance X in the first resource pool as the target physical processor used to create the target virtual device instance. That is, the target virtual device instance preempts the physical processor resources of virtual device instance X.

Assuming that the second resource pool is used for creating the target virtual device instance, one physical processor can be determined from the second resource pool as the target physical processor, one physical processor can be determined randomly as the target physical processor, and one physical processor can be determined as the target physical processor according to the application characteristic information of the virtual device instance currently running on each physical processor in the second resource pool. For example, a physical processor with a low application load of the currently running virtual device instance is determined as the target physical processor.

In summary, for both latency sensitive and load application feature information, a virtual device instance running a latency sensitive application may preempt the physical processor resources occupied by a virtual device instance running a latency insensitive application, and a virtual device instance running a high load application may preempt the physical processor resources occupied by a virtual device instance running a low load application. Thus, if the application load level of the target virtual device instance is higher than that of the virtual device instance X, or if the application of the target virtual device instance is an online class application and the application on the virtual device instance X is an offline class application, it is determined that the target virtual device instance may occupy a physical processor used by the virtual device instance X, and the virtual device instance X is migrated from the first resource pool to the second resource pool, i.e., one physical processor is determined in the second resource pool, and the virtual device instance X is migrated to the processing processor for running.

In this embodiment, the physical processor resources are divided into two resource pools according to the exclusive use and the shared use modes, and when the virtual device instance is created, the resource pools conforming to the application characteristics of the virtual device instance are selected for creation according to the application characteristic information of the virtual device instance, which is helpful for ensuring the stable operation of the virtual device instance and improving the overall resource utilization rate. Because by overstocking some physical processors, the processor resources equivalent to flexible scheduling are more abundant, the use requirements of more users can be met, and moreover, a plurality of virtual device instances share one physical processor, so that the resource utilization rate of the physical processor can be improved. Aiming at virtual equipment instances running delay-sensitive application programs with high loads, the virtual equipment instances use a physical processor in an exclusive mode, so that the physical processor resources occupied by the virtual equipment instances are prevented from being frequently scheduled, the running stability of the application programs can be ensured, and the user experience is ensured.

Fig. 4 is a flowchart of a method for scheduling resources according to an embodiment of the present invention, where the method may be performed by the above-mentioned resource scheduling service cluster, and as shown in fig. 4, the method includes the following steps:

401. and acquiring a first resource pool and a second resource pool which contain heterogeneous physical processors, wherein each physical processor in the first resource pool supports the sharing use of the virtual equipment instance, and each physical processor in the second resource pool supports the sharing use of the virtual equipment instance.

402. And responding to the creation request of the target virtual equipment instance, and determining a target resource pool used for creating the target virtual equipment instance according to the application characteristic information of the target virtual equipment instance, wherein the target resource pool is a first resource pool or a second resource pool.

403. A target physical processor is determined in the target resource pool to create a target virtual device instance based on the target physical processor.

404. And determining a reference computing power performance value according to the corresponding actual computing power performance value of each physical processor.

405. And determining the occupation amount of the computing power resources of the target physical processor by the target virtual device instance according to the reference computing power performance value, wherein the computing power performance value reached by the consumption of the computing power resources corresponding to the target virtual device instance by the target physical processor does not exceed the reference computing power performance value.

406. And in response to the update of the application characteristic information of the target virtual device instance, determining to migrate the target virtual device instance to another physical processor according to the updated application characteristic information, wherein the other physical processor and the target physical processor are positioned in different resource pools or the same resource pool.

In practical application, the application feature information of any target virtual device instance is dynamically changed, for example, the load of an application program running on the target virtual device instance is dynamically changed, even the application program running on the target virtual device instance is changed, and the original offline class application is changed into the online class application. Therefore, according to the dynamic update of the application characteristic information of the target virtual device instance, the occupied physical processor resources also dynamically change, and the change is reflected by the replacement of the physical processor and/or the change of the occupied computing power resources of the physical processor.

Described in step 406 above is the case of updating the occupied physical processor.

Optionally, if the application characteristic information of the target virtual device instance changes from delay sensitive to delay insensitive and/or the application characteristic information of the target virtual device instance changes from high load to low load, determining to migrate the target virtual device instance from the target physical processor in the first resource pool to another physical processor in the second resource pool.

Optionally, if the application characteristic information of the target virtual device instance changes from delay insensitive to delay sensitive and/or the application characteristic information of the target virtual device instance changes from low load to high load, determining to migrate the target virtual device instance from the target physical processor in the second resource pool to another physical processor in the first resource pool.

The two migration scenarios described above are illustrated in fig. 5.

Assuming that the target virtual device instance originally runs on the target physical processor in the first resource pool, after a period of time, for example, the application load of the target virtual device instance changes from high to low (for example, from higher than a set threshold value to lower than the set threshold value), for example, the application program on the target virtual device instance changes from being sensitive to being insensitive to delay (for example, from an online class application to an offline class application, or updates the requirement that the original requirement is lower than the set response delay to a requirement that the response delay can be higher), the target virtual device instance can be migrated to the second resource pool to share another physical processor with other virtual device instances.

Assuming that the target virtual device instance originally runs on the target physical processor in the second resource pool, after a period of running, for example, the application load of the target virtual device instance changes from low to high, and for example, the application program on the target virtual device instance changes from being sensitive to delay ratio to being sensitive to delay, the target virtual device instance may be migrated to another physical processor in the first resource pool to share the other physical processor alone.

In practice, migrating virtual device instances in different resource pools often results in significant changes in application characteristic information of the virtual device instances. When the application characteristic information of the virtual device instance is not obviously changed, migration of the virtual device instance on different physical processors can be omitted so as to reduce scheduling overhead.

For the case that a plurality of virtual device instances in the second resource pool share one physical processor, for example, the virtual device instance a and the virtual device instance b share a certain physical processor R, if the load of the virtual device instance a becomes high, more computing resources occupying the physical processor R may be allocated to the virtual device instance a (of course, the computing resource occupation amount corresponding to the computing performance value cannot be exceeded), at this time, the computing resources actually able to be occupied by the virtual device instance b may be reduced, that is, the virtual device instance b may have the same consumption. It will be appreciated that if the load of the virtual device instance a continues to increase above the set threshold, the virtual device instance a may be migrated to the first resource pool, or if there is no physical processor available in the first resource pool, the virtual device instance b with a lower load may be migrated to another physical processor in the second resource pool, so that the virtual device instance can monopolize the physical processor R.

In the above embodiments, the scheduling process may be considered as a process of scheduling the total amount of computing power resources that can be provided by the physical processor based on the reference computing power performance value and the usage pattern of the virtual device instance on each physical processor.

In another alternative embodiment, the computing power performance of the physical processors may be limited in a unified physical layer in the initial stage, so that the total computing power resources that can be provided by each physical processor are kept consistent, which is equivalent to changing the original heterogeneous physical processors into isomorphic physical processors, and then, the limited physical processors are used to construct and run virtual device instances.

Fig. 6 is a flowchart of a method for scheduling resources according to an embodiment of the present invention, where the method may be performed by the above-mentioned resource scheduling service cluster, and as shown in fig. 6, the method includes the following steps:

601. actual computational power performance values corresponding to each of the heterogeneous plurality of physical processors are determined.

602. And determining a reference computing power performance value according to the actual computing power performance value corresponding to each of the plurality of physical processors.

603. The computing force performance values that the plurality of physical processors can provide are limited based on the reference computing force performance values.

604. And determining the occupation amount of the computing power resources of the target virtual device instance running on the target physical processor to the target physical processor after the limitation.

In step 603, according to the reference performance value, the performance values that can be provided by the multiple physical processors are limited, and the performance that can be provided by the physical processors is pre-suppressed from the physical layer, so that the physical processors whose performance values are higher than the reference performance values cannot exert all the performance, and finally, the heterogeneous performance values that can be provided by the multiple physical processors are kept consistent, so as to form multiple physical processors with the same structure.

In practice, there are various indexes for evaluating the computing power performance of the physical processor, such as a CPU, for example, a CPU frequency, a memory bandwidth occupied by the CPU (the number of memory banks), a capacity of a used cache (such as an L3 level cache capacity), and the like.

Based on this, for a physical processor whose actual calculated performance value is higher than the reference calculated performance value, the calculated performance value of the physical processor may be limited to the level of the reference calculated performance value by one or more of reducing its CPU frequency, reducing the used memory bandwidth, and reducing the used cache capacity.

After the computing power performance of each physical processor is uniformly limited in advance, each physical processor after the limitation can be used for creating and scheduling the virtual device instance for the user. Similar to the above, for any target physical processor, the usage patterns of the virtual device instance may have both exclusive and shared usage patterns.

Assuming that a target virtual device instance running on a target physical processor uses the target physical processor in an exclusive mode, determining that the computing power resource occupation amount of the target virtual device instance on the target physical processor takes the total computing power resource amount which can be provided by the limited target physical processor as an upper limit. That is, the amount of computational resources actually used by the target virtual device instance on the limited target physical processor is [0 ] the total amount of computational resources that the limited target physical processor is capable of providing ].

Assuming that at least two virtual device instances run on the target physical processor, that is, the at least two virtual device instances share the target physical processor, the computing power resource occupation amounts of the at least two virtual device instances on the target physical processor respectively can be determined according to the current application characteristic information of the at least two virtual device instances. And the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource amount which can be provided by the target physical processor after the limitation. It will be appreciated that any target virtual device instance may actually use an amount of computing power resources on the limited target physical processor that does not exceed the total amount of computing power resources that the limited target physical processor is capable of providing. Moreover, if the amount of computational resources actually required by a certain virtual device instance reaches the total amount of computational resources that can be provided by the limited target physical processor, then other virtual device instances sharing the target physical processor will no longer be able to use the computational resources provided by the target physical processor, and at this time, the other virtual device instances can be migrated to other physical processors.

A resource scheduling apparatus of one or more embodiments of the present invention will be described in detail below. Those skilled in the art will appreciate that these means may be configured by the steps taught by the present solution using commercially available hardware components.

Fig. 7 is a schematic structural diagram of a resource scheduling device according to an embodiment of the present invention, as shown in fig. 7, where the device includes: a determining module 11 and a scheduling module 12.

The determining module 11 is configured to determine actual computing power performance values corresponding to each of the heterogeneous physical processors, and determine a reference computing power performance value according to the actual computing power performance values corresponding to each of the physical processors.

And the scheduling module 12 is configured to determine, according to the reference computing power performance value, a computing power resource occupation amount of a target virtual device instance running on a target physical processor to the target physical processor, where the computing power performance value reached by the computing power resource occupation amount corresponding to consumption of the target virtual device instance by the target physical processor does not exceed the reference computing power performance value, and the target physical processor is any one of the multiple physical processors.

Optionally, when the target virtual device instance monopolizes the target physical processor, the scheduling module 12 is specifically configured to: determining a target amount of computing power resources that the target physical processor consumes to reach the reference computing power performance value; and determining that the computing power resource occupation amount of the target virtual equipment instance on the target physical processor takes the target computing power resource amount as an upper limit.

Optionally, when the target virtual device instance is at least two virtual device instances, the scheduling module 12 is specifically configured to: determining the occupation amount of computing power resources of the at least two virtual device instances to the target physical processor respectively according to the current application characteristic information of the at least two virtual device instances; the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource which can be provided by the target physical processor, and the computing power performance value which can be achieved by the target physical processor consuming the computing power resource occupation amount corresponding to any one of the at least two virtual device instances does not exceed the reference computing power performance value.

Optionally, the determining module 11 is further configured to: and limiting the computing force performance values which can be provided by the plurality of physical processors according to the reference computing force performance values. Based on this, the scheduling module 12 is configured to: and determining the occupation amount of the computing power resources of the target virtual device instance running on the target physical processor to the target physical processor after the limitation.

Wherein, optionally, when the target virtual device instance monopolizes the target physical processor, the scheduling module 12 is specifically configured to: and determining the occupation amount of the computing power resources of the target virtual equipment instance to the target physical processor, wherein the total amount of the computing power resources which can be provided by the target physical processor after the limitation is taken as an upper limit. Wherein, optionally, when the target virtual device instance is at least two virtual device instances, the scheduling module 12 is specifically configured to: determining the occupation amount of computing power resources of the at least two virtual device instances to the target physical processor respectively according to the current application characteristic information of the at least two virtual device instances; and the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource which can be provided by the target physical processor after the limitation.

Optionally, the apparatus further comprises: the acquisition module is used for forming a first resource pool and a second resource pool according to the plurality of physical processors, wherein each physical processor in the first resource pool supports the exclusive use of the virtual equipment instance, and each physical processor in the second resource pool supports the shared use of the virtual equipment instance. At this time, if the target physical processor is located in the first resource pool, the target virtual device instance monopolizes the target physical processor; if the target physical processor is located in the second resource pool, the target virtual device instance includes at least two virtual device instances, and the at least two virtual device instances share the target physical processor.

Optionally, the apparatus further comprises: the creation module is used for responding to the creation request of the target virtual equipment instance, determining a target resource pool used for creating the target virtual equipment instance according to the application characteristic information of the target virtual equipment instance, wherein the target resource pool is the first resource pool or the second resource pool; the target physical processor is determined in the target resource pool to create the target virtual device instance based on the target physical processor.

Optionally, the scheduling module 12 is further configured to: and responding to the update of the application characteristic information of the target virtual equipment instance, and determining to migrate the target virtual equipment instance to another physical processor according to the updated application characteristic information, wherein the other physical processor and the target physical processor are positioned in different resource pools or the same resource pool.

Wherein optionally, the application feature information includes: delay sensitivity and/or loading.

Optionally, the scheduling module 12 is specifically configured to: if the application characteristic information of the target virtual device instance changes from delay sensitivity to delay insensitivity and/or the application characteristic information of the target virtual device instance changes from high load to low load, determining to migrate the target virtual device instance to the other physical processor in the second resource pool, wherein the delay sensitivity and the load are divided by corresponding thresholds, and the target physical processor is located in the first resource pool.

Optionally, the scheduling module 12 is specifically configured to: and if the application characteristic information of the target virtual equipment instance is changed from delay insensitive to delay sensitive and/or the application characteristic information of the target virtual equipment instance is changed from low load to high load, determining to migrate the target virtual equipment instance to the other physical processor in the first resource pool, wherein the target physical processor is positioned in the second resource pool.

The apparatus shown in fig. 7 may perform the steps in the foregoing embodiments, and the detailed execution and technical effects are referred to the descriptions in the foregoing embodiments, which are not repeated herein.

In one possible design, the structure of the resource scheduling device shown in fig. 7 may be implemented as an electronic device. As shown in fig. 8, the electronic device may include: a processor 21, a memory 22, a communication interface 23. Wherein the memory 22 has stored thereon executable code which, when executed by the processor 21, causes the processor 21 to at least implement the resource scheduling method as provided in the previous embodiments.

Additionally, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code that, when executed by a processor of an electronic device, causes the processor to at least implement a resource scheduling method as provided in the previous embodiments.

The apparatus embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by adding necessary general purpose hardware platforms, or may be implemented by a combination of hardware and software. Based on such understanding, the foregoing aspects, in essence and portions contributing to the art, may be embodied in the form of a computer program product, which may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (14)

1. A method for scheduling resources, the method comprising:

determining actual computing power performance values corresponding to each of the heterogeneous physical processors;

determining a reference computing power performance value according to the actual computing power performance values corresponding to the physical processors respectively;

and determining the computing power resource occupation amount of a target virtual device instance running on a target physical processor to the target physical processor according to the reference computing power performance value, wherein the computing power performance value reached by the computing power resource occupation amount corresponding to the target physical processor consumed by the target virtual device instance does not exceed the reference computing power performance value, and the target physical processor is any one of the physical processors.

2. The method of claim 1, wherein after determining the reference calculated force performance value, further comprising:

limiting the computational power performance values that the plurality of physical processors can provide according to the reference computational power performance values;

the determining, according to the reference computing power performance value, the computing power resource occupation amount of the target virtual device instance running on the target physical processor to the target physical processor includes:

And determining the occupation amount of the computing power resources of the target virtual device instance running on the target physical processor to the target physical processor after the limitation.

3. The method of claim 2, wherein the target virtual device instance monopolizes the target physical processor;

the determining the occupation amount of the computing power resource of the target virtual device instance running on the target physical processor to the target physical processor after the limitation comprises the following steps:

and determining the occupation amount of the computing power resources of the target virtual equipment instance to the target physical processor, wherein the total amount of the computing power resources which can be provided by the target physical processor after the limitation is taken as an upper limit.

4. The method of claim 2, wherein the target virtual device instance is at least two virtual device instances;

the determining the occupation amount of the computing power resource of the target virtual device instance running on the target physical processor to the target physical processor after the limitation comprises the following steps:

determining the occupation amount of computing power resources of the at least two virtual device instances to the target physical processor respectively according to the current application characteristic information of the at least two virtual device instances; and the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource which can be provided by the target physical processor after the limitation.

5. The method of claim 1, wherein the target virtual device instance monopolizes the target physical processor;

the determining the computing power resource occupation amount of the target virtual device instance running on the target physical processor to the target physical processor comprises the following steps:

determining a target amount of computing power resources that the target physical processor consumes to reach the reference computing power performance value;

and determining that the computing power resource occupation amount of the target virtual equipment instance on the target physical processor takes the target computing power resource amount as an upper limit.

6. The method of claim 1, wherein the target virtual device instance is at least two virtual device instances;

the determining the computing power resource occupation amount of the target virtual device instance running on the target physical processor to the target physical processor comprises the following steps:

determining the occupation amount of computing power resources of the at least two virtual device instances to the target physical processor respectively according to the current application characteristic information of the at least two virtual device instances; the sum of the computing power resource occupation amounts of the at least two virtual device instances to the target physical processor corresponds to the total computing power resource which can be provided by the target physical processor, and the computing power performance value which can be achieved by the target physical processor consuming the computing power resource occupation amount corresponding to any one of the at least two virtual device instances does not exceed the reference computing power performance value.

7. The method according to any one of claims 1 to 6, further comprising:

forming a first resource pool and a second resource pool according to the plurality of physical processors, wherein each physical processor in the first resource pool supports the exclusive use of the virtual equipment instance, and each physical processor in the second resource pool supports the shared use of the virtual equipment instance;

if the target physical processor is located in the first resource pool, the target virtual device instance monopolizes the target physical processor; if the target physical processor is located in the second resource pool, the target virtual device instance includes at least two virtual device instances, and the at least two virtual device instances share the target physical processor.

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

responding to the creation request of the target virtual equipment instance, and determining a target resource pool used for creating the target virtual equipment instance according to the application characteristic information of the target virtual equipment instance, wherein the target resource pool is the first resource pool or the second resource pool;

the target physical processor is determined in the target resource pool to create the target virtual device instance based on the target physical processor.

9. The method of claim 8, wherein the method further comprises:

and responding to the update of the application characteristic information of the target virtual equipment instance, and determining to migrate the target virtual equipment instance to another physical processor according to the updated application characteristic information, wherein the other physical processor and the target physical processor are positioned in different resource pools or the same resource pool.

10. The method of claim 9, wherein the application characteristic information comprises: delay sensitivity and/or loading.

11. The method of claim 10, wherein the determining to migrate the target virtual device instance to another physical processor based on the updated application characteristic information in response to the updating of the application characteristic information of the target virtual device instance comprises:

if the application characteristic information of the target virtual device instance changes from delay sensitivity to delay insensitivity and/or the application characteristic information of the target virtual device instance changes from high load to low load, determining to migrate the target virtual device instance to the other physical processor in the second resource pool, wherein the delay sensitivity and the load are divided by corresponding thresholds, and the target physical processor is located in the first resource pool.

12. The method of claim 10, wherein the determining to migrate the target virtual device instance to another physical processor based on the updated application characteristic information in response to the updating of the application characteristic information of the target virtual device instance comprises:

and if the application characteristic information of the target virtual equipment instance is changed from delay insensitive to delay sensitive and/or the application characteristic information of the target virtual equipment instance is changed from low load to high load, determining to migrate the target virtual equipment instance to the other physical processor in the first resource pool, wherein the target physical processor is positioned in the second resource pool.

13. An electronic device, comprising: a memory, a processor, a communication interface; wherein the memory has stored thereon executable code which, when executed by the processor, causes the processor to perform the resource scheduling method of any one of claims 1 to 12.

14. A non-transitory machine-readable storage medium having stored thereon executable code which, when executed by a processor of an electronic device, causes the processor to perform the resource scheduling method of any of claims 1 to 12.

Images