CN115080242A

CN115080242A - Method, device and medium for unified scheduling of PCI equipment resources

Info

Publication number: CN115080242A
Application number: CN202210764562.8A
Authority: CN
Inventors: 张百林; 亓开元; 苏志远
Original assignee: Jinan Inspur Data Technology Co Ltd
Current assignee: Jinan Inspur Data Technology Co Ltd
Priority date: 2022-06-30
Filing date: 2022-06-30
Publication date: 2022-09-20

Abstract

The application relates to the technical field of resource management, and discloses a method, a device and a medium for unified scheduling of PCI equipment resources, which comprise the following steps: and modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model, after a PCI device resource request instruction is obtained, performing resource allocation based on each PCI device model, and updating relevant information of each current PCI device in a PCI device resource pool, wherein the relevant information at least comprises the current use state of each PCI device and the number of available PCI devices. And uploading the relevant information of each PCI device to an Open Stack cloud platform. Therefore, unified scheduling of the PCI devices is realized based on each PCI device model. In addition, the information such as the current use state of each PCI device, the number of available PCI devices and the like can be visualized on the Open Stack cloud platform, and therefore resource consumption and operation and maintenance cost are reduced.

Description

Method, device and medium for unified scheduling of PCI equipment resources

Technical Field

The present application relates to the field of resource management technologies, and in particular, to a method, an apparatus, and a medium for unified scheduling of PCI device resources.

Background

The Open Stack is a combination of a series of software Open source projects, and is the most popular Open source cloud computing management platform project at present. The computing module of the Open Stack cloud platform tracks and manages basic resources such as a CPU, a memory, and a disk by using a resource service module, however, it is impossible to perform unified scheduling and management on Peripheral Component Interconnect (PCI) device resources, where PCI devices refer to devices connected to a motherboard through a PCI interface, such as a Field-Programmable Gate Array (FPGA), a display card, a Graphics Processing Unit (GPU), a QAT, and the like.

Currently, when an Open Stack cloud platform creates a cloud host instance to use PCI devices, PCI resources of a corresponding specification are selected by setting "PCI _ passhrrough: alias" metadata in a resource specification. When the PCI device resources are used in this manner, since it is necessary to declare the basic resources and then apply for the corresponding PCI device to create the cloud host instance on the Open Stack cloud platform, although the PCI device may be used in this manner, there may be a case where multiple cloud host instances compete for the same PCI device, which may cause the PCI device resources to be rescheduled, and increase resource consumption.

In addition, PCI resources can be allocated to the cloud host instance through a network service mode based on the SR-IOV technology, however, in such a mode, the allocated and available PCI device resources cannot be uniformly viewed through a visual interface because the PCI device resources are not stored in the database.

Therefore, how to uniformly manage and schedule PCI device resources, and to realize PCI device resource state visualization, and reduce operation and maintenance costs of PCI devices when used on an Open Stack cloud platform while reducing resource consumption are problems to be solved by those skilled in the art.

Disclosure of Invention

The application aims to provide a method, a device and a medium for unified scheduling of PCI equipment resources, so that unified management and scheduling of the PCI equipment resources are realized, visualization of the PCI equipment resources is realized, and the maximum utilization rate of the resources is realized.

In order to solve the above technical problem, the present application provides a method for uniformly scheduling PCI device resources, including:

modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model;

after a PCI equipment resource request instruction is obtained, PCI equipment resource allocation is carried out on the basis of each PCI equipment model;

updating the relevant information of each PCI device in a PCI device resource pool; the relevant information of the PCI equipment at least comprises the current use state of each PCI equipment and the number of available PCI equipment;

and uploading the relevant information of each PCI device in the current PCI device resource pool to an Open Stack cloud platform.

Preferably, the modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model includes:

dividing each PCI device into a physical unit and a virtual unit; wherein one of the PCI devices is one of the physical units, and one of the physical units corresponds to a plurality of the virtual units;

setting the same label for each virtual unit with the same attribute in the same physical unit;

setting different labels for each virtual unit with different attributes in the same physical unit;

and modeling based on the mapping relation among the physical units, the virtual units corresponding to the physical units and the corresponding labels to obtain the PCI equipment models.

Preferably, before the modeling each PCI device according to the virtualization attribute of the PCI device to obtain the corresponding PCI device model, the method further includes:

acquiring each PCI device, and analyzing each PCI device to determine the type of each PCI device;

and configuring the virtualization attributes corresponding to the PCI equipment in advance according to the type of the PCI equipment.

Preferably, after the PCI device resource request instruction is obtained, allocating PCI device resources based on each PCI device model includes:

analyzing the PCI equipment resource request instruction to determine the type of the PCI equipment resource request;

positioning a target computing host according to the PCI equipment resource request type;

judging whether available target PCI equipment exists in the target computing host; wherein, the target PCI equipment is the equipment requested by the PCI equipment resource request instruction;

if the PCI equipment model exists, resource allocation is carried out according to the target PCI equipment model corresponding to the target PCI equipment;

if not, ending the resource allocation.

Preferably, the allocating resources according to the target PCI device model corresponding to the target PCI device includes:

judging whether borrowed units exist in virtual units corresponding to physical units of the target PCI equipment model;

if the borrowed unit exists, determining whether an available unit exists in each virtual unit which is not borrowed, if so, allocating the corresponding virtual unit, and if not, rejecting the resource allocation request;

if there are no borrowed units, then the physical units are allocated directly.

Preferably, the allocating PCI device resources based on each PCI device model includes:

when the virtual unit is allocated, resource allocation is carried out according to the identifier corresponding to the virtual unit and the address code of the corresponding physical unit;

and when the physical unit is allocated, performing resource allocation according to the address code corresponding to the physical unit.

Preferably, before said locating the target computing host according to the PCI device resource request type, further comprises:

filtering all the calculation hosts through a NUMA affinity strategy to obtain a first calculation host list;

acquiring a second computing host list meeting the PCI equipment resource request from the first computing host list;

correspondingly, the positioning the target computing host according to the PCI device resource request type includes:

and positioning the target computing host in the second computing host list according to the PCI equipment resource request type.

Preferably, said locating the target computing host in the second computing host list according to the PCI device resource request type comprises:

and taking the host with the highest utilization rate in the second computing host list as the target computing host.

Preferably, the uploading the relevant information of each PCI device in the current PCI device resource pool to the Open Stack cloud platform includes:

and uploading the relevant information of each PCI device in the current PCI device resource pool to an Open Stack cloud platform every other preset time length.

Preferably, the method for uniformly scheduling PCI device resources further includes:

and after determining that the PCI equipment resources are used, releasing the PCI equipment resources, and entering the step of updating the related information of the current PCI equipment in the PCI equipment resource pool.

In order to solve the above technical problem, the present application further provides a device for uniformly scheduling PCI device resources, including:

the modeling unit is used for modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model;

the distribution unit is used for distributing PCI equipment resources based on each PCI equipment model after a PCI equipment resource request instruction is obtained;

the updating unit is used for updating the related information of each PCI device in the PCI device resource pool; the relevant information of the PCI equipment at least comprises the current use state of each PCI equipment and the number of available PCI equipment;

and the uploading unit is used for uploading the relevant information of each PCI device in the current PCI device resource pool to an Open Stack cloud platform.

In order to solve the above technical problem, the present application further provides a device for uniformly scheduling PCI device resources, including a memory for storing a computer program;

and the processor is used for realizing the steps of the method for uniformly scheduling the PCI equipment resources when executing the computer program.

In order to solve the above technical problem, the present application further provides a computer-readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the computer program implements the steps of the method for unified scheduling of PCI device resources.

The invention provides a method for uniformly scheduling PCI equipment resources, which comprises the following steps: and modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model, after obtaining the PCI device resource request instruction, distributing the PCI device resources based on each PCI device model corresponding to each PCI device, and updating the related information of each current PCI device in the PCI device resource pool after distribution, wherein the related information of the PCI devices at least comprises the current use state of each PCI device and the number of available PCI devices. And after updating, uploading the relevant information of each PCI device in the current PCI device resource pool to an Open Stack cloud platform. Therefore, according to the technical scheme provided by the application, each PCI device is modeled based on the virtualization attribute of the PCI device, and unified scheduling and distribution of all PCI devices are realized according to each PCI device model. In addition, the relevant information of each PCI device in the current PCI device resource pool is uploaded to the Open Stack cloud platform, and the Open Stack cloud platform visualizes the relevant information of each PCI device, namely, visualizes the information such as the current use state of each PCI device and the number of available PCI devices, so that the visualization of the PCI device resource state and the unified management and scheduling of the PCI device resources are realized, the resource consumption is reduced, and the operation and maintenance cost of the PCI devices when used on the Open Stack cloud platform is reduced.

In addition, the application also provides a device and a medium for uniformly scheduling the PCI equipment resources, which correspond to the method for uniformly scheduling the PCI equipment resources, and the effect is the same as the effect.

Drawings

In order to more clearly illustrate the embodiments of the present application, the drawings needed for the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained by those skilled in the art without inventive effort.

Fig. 1 is a flowchart of a method for uniformly scheduling PCI device resources according to an embodiment of the present application;

fig. 2 is a structural diagram of a first PCI device model according to an embodiment of the present disclosure;

FIG. 3 is a block diagram of another first PCI device model according to an embodiment of the present application;

FIG. 4 is a diagram illustrating a second PCI device model according to an embodiment of the present disclosure;

FIG. 5 is a block diagram of another second PCI device model according to an embodiment of the present application;

fig. 6 is a structural diagram of a PCI device resource according to an embodiment of the present application;

fig. 7 is a structural diagram of an apparatus for unified scheduling of PCI device resources according to an embodiment of the present application;

fig. 8 is a block diagram of an apparatus for unified scheduling of PCI device resources according to another embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without any creative effort belong to the protection scope of the present application.

The core of the application is to provide a method, a device and a medium for unified scheduling of PCI equipment resources, so that unified scheduling and visualization of the PCI equipment resources are realized, and the PCI equipment resources are utilized to the maximum extent.

In order that those skilled in the art will better understand the disclosure, the following detailed description will be given with reference to the accompanying drawings.

In order to perform unified management and scheduling on PCI device resources, implement PCI device resource state visualization, reduce resource consumption, and reduce operation and maintenance costs of PCI devices when used on an Open Stack cloud platform, embodiments of the present application provide a method for PCI device resource unified scheduling, which models each PCI device according to virtualization attributes of each PCI device to obtain a PCI device model, thereby implementing PCI device unified scheduling based on the PCI device model, and transmitting information such as a current use state and an available PCI device number of each PCI device in a PCI device resource pool to the Open Stack cloud platform so as to implement PCI device resource visualization.

Fig. 1 is a flowchart of a method for uniformly scheduling PCI device resources according to an embodiment of the present application, where as shown in fig. 1, the method includes:

s10: and modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model.

In a specific embodiment, after the PCI devices in all the computing hosts are acquired, each PCI device is analyzed to determine the type corresponding to the PCI device, and the virtualization attributes of each PCI device are configured in advance based on the type of the PCI device.

And after the virtualization attributes of the PCI devices are configured in advance, modeling the PCI devices according to the virtualization attributes of the PCI devices to obtain corresponding PCI device models. During modeling, each PCI device is divided into a physical unit and a virtual unit, wherein one PCI device is a physical unit, and one physical unit corresponds to a plurality of virtual units. If the attributes of the virtual units in the same physical unit are the same, setting the same label for each virtual unit, and if the attributes of the virtual units in the same physical unit are different, setting the different labels for each virtual unit device. Thus, the PCI device model is established based on the mapping relationship between the physical units, the virtual units corresponding to the physical units and the labels corresponding to the physical units. It is understood that a PCI device with the same tag is built as one PCI device model, and a PCI device with a different tag is built as another PCI device model. For ease of understanding, the following will be exemplified.

For example, the PCI device is a Graphics Processing Unit (GPU), and the GPU may be configured in advance as different virtual units with the same video memory size, for example, the 6G video memory size is configured as 3 2G video memory units, obviously, the virtualization attributes of the 3 units are all 2G video memories, and therefore, the same tag is set for each 2G video memory. Of course, the virtual units with different video memory sizes may also be configured in advance, for example, the configured 6G video memory size is configured as one 2G video memory virtual unit and one 4G video memory virtual unit. Obviously, the virtualization attributes of the 3 units are different, and are 2G video memory and 4G video memory, so that different tags are set for the 2G video memory and the 4G video memory.

It is noted that, because the virtualization attributes of different types of PCI devices are different, a corresponding PCI device model is established for each PCI device.

S11: and after the PCI equipment resource request instruction is obtained, PCI equipment resource allocation is carried out based on each PCI equipment model.

After each PCI device is modeled to obtain a PCI device model, if a PCI device resource request instruction is received, the PCI device resource request instruction is analyzed to determine the PCI device resource request type, a target computing host is positioned according to the PCI device resource request type, and when it is determined that available target PCI devices exist in the target computing host, PCI device allocation is carried out based on the target PCI device model corresponding to the target PCI devices. Of course, if there is no available target PCI device in the target computing host, it is determined that the PCI device resource allocation fails, that is, the resource allocation is ended.

When the virtual units are distributed based on the target PCI equipment model, if certain virtual unit exists in the virtual units corresponding to the physical units in the target PCI equipment model, whether available units exist in the remaining virtual units which are not borrowed, namely whether virtual units meeting PCI equipment resource requests exist is judged, if yes, the corresponding virtual units are distributed, if not, the distribution is finished, and the resource distribution is determined to fail. Of course, if none of the virtual units corresponding to a physical unit in the target PCI device model is borrowed, the physical unit may be directly allocated, and the manner of directly allocating the physical unit is referred to as through.

It should be noted that, when a physical unit is directly allocated, the physical unit is allocated by using the address code corresponding to the physical unit, and when a virtual unit is allocated, the virtual unit needs to be allocated by combining the identifier corresponding to the virtual unit and the address code corresponding to the physical unit to locate and allocate the resource.

In addition, it should be noted that, when a target computing host is located, all computing hosts are filtered through the NUMA affinity policy, a first computing host list is obtained by filtering out computing hosts that do not satisfy affinity, a second computing host list that satisfies the PCI device resource request is obtained from the first computing host list, and finally, the target computing host is determined from the second computing host list according to the PCI device resource request type and the resource utilization rate.

S12: updating the relevant information of each current PCI device in the PCI device resource pool; the relevant information of the PCI devices at least comprises the current use state of each PCI device and the number of available PCI devices.

After the PCI device resources are allocated in step S11, the relevant information of each PCI device in the PCI device resource pool is updated, that is, the information such as the current usage status of each PCI device and the number of available PCI devices is updated, it should be noted that the relevant information of the PCI device may include, but is not limited to, the current usage status of each PCI device and the number of available PCI devices.

It should be noted that the updating of the PCI device resource pool may be triggering the resource pool updating every time resource allocation is performed, or may be updating after a fixed time period, which is not limited in this application.

S13: and uploading the relevant information of each PCI device in the current PCI device resource pool to an Open Stack cloud platform.

In order to enable a user to visualize information such as the current use state of each PCI device, after updating the relevant information of each PCI device in the PCI device resource pool, the relevant information of each PCI device in the PCI device resource pool is uploaded to the Open Stack cloud platform, so that the user can visually observe the current state of each PCI device from the Open Stack cloud platform.

It should be noted that, the uploading of the relevant information of each PCI device may be performed after each update of the PCI device resource pool, or may be performed when it is determined that the updated resource pool is different from the resource pool before the update, that is, when it is determined that the relevant information of the PCI device in the PCI device resource pool changes, the relevant information is uploaded. The relevant information of each PCI device may also be uploaded once every preset time period, which is not limited in this application.

It can be understood that when it is determined that there is a PCI device that is used, the PCI device resource is immediately released, and the relevant information of the PCI device is updated, so that other tasks can invoke the PCI device in time.

The method for uniformly scheduling the PCI equipment resources provided by the embodiment of the application comprises the following steps: and modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model, after obtaining the PCI device resource request instruction, distributing the PCI device resources based on each PCI device model corresponding to each PCI device, and updating the related information of each current PCI device in the PCI device resource pool after distribution, wherein the related information of the PCI devices at least comprises the current use state of each PCI device and the number of available PCI devices. And after updating, uploading the relevant information of each PCI device in the current PCI device resource pool to the Open Stack cloud platform. Therefore, according to the technical scheme provided by the application, each PCI device is modeled based on the virtualization attribute of the PCI device, and unified scheduling and distribution of all PCI devices are realized according to each PCI device model. In addition, the relevant information of each PCI device in the current PCI device resource pool is uploaded to the Open Stack cloud platform, and the Open Stack cloud platform visualizes the relevant information of each PCI device, namely, visualizes the information such as the current use state of each PCI device and the number of available PCI devices, so that the visualization of the PCI device resource state and the unified management and scheduling of the PCI device resources are realized, the resource consumption is reduced, and the operation and maintenance cost of the PCI devices when used on the Open Stack cloud platform is reduced.

In a specific embodiment, when modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model, each PCI device needs to be divided into a physical unit and a virtual unit, where one PCI device is a physical unit, that is, one PCI device corresponds to one physical unit, and one physical unit corresponds to multiple virtual units. In modeling, the same label is set for each virtual unit with the same virtualization attribute corresponding to the physical unit, so that a PCI device model is obtained by modeling based on the mapping relationship among the physical unit, each virtual unit corresponding to the physical unit, and the corresponding same label, and the PCI device model with the same label can be referred to as a first PCI device model.

Fig. 2 is a structural diagram of a first PCI device model according to an embodiment of the present application, and as shown in fig. 2, one PCI device corresponds to one physical unit PF, and one physical unit PF corresponds to a plurality of virtual units VF, and in the first PCI device model, the virtualization attributes of each virtual unit VF are the same under the same physical unit PF, so that the same tag TtaitsA needs to be set for each virtual unit VF with the same virtualization attribute, that is, the tag TtaitsA is used to represent the virtualization attributes of each virtualization unit, and when scheduling the PCI device, the virtual unit can be quickly located according to the tag. In fact, one physical unit PF may correspond to multiple virtual units VF, or one physical unit PF may correspond to one virtual unit VF, which is not limited to this. Wherein each virtual unit VF can be assigned to a different cloud host instance. For ease of understanding, the following will be exemplified.

Fig. 3 is a structural diagram of another first PCI device model provided in this embodiment of the present application, for example, a physical unit PF is a GPU, and as shown in fig. 3, each virtual unit VF corresponding to the GPU is a 2G video memory, and a label of each virtual unit VF is Ttaits1, which is used to represent that each virtual unit VF is a 2G video memory.

In addition, different labels are set for the virtual units with different virtualization attributes corresponding to the physical unit, and a PCI device model can also be obtained by modeling based on the mapping relationship among the physical unit, the virtual units corresponding to the physical unit, and the corresponding different labels, and the PCI device model is referred to as a second PCI device model.

Fig. 4 is a structural diagram of a second PCI device model provided in the embodiment of the present application, and as shown in fig. 4, similarly, one PCI device corresponds to one physical unit PF, one physical unit PF corresponds to multiple virtual units VF, and each virtual unit VF may be allocated to a different cloud host instance. In the second PCI device model, the virtualization attributes of each virtual unit VF are different, and therefore, different labels need to be set for each virtual unit VF with different virtualization attributes. As shown in fig. 4, the virtualization attributes of the virtual units VF corresponding to TtaitsB are the same. The virtualization attributes of the virtual elements VF corresponding to TtaitsB to TtaitsE are different. And when each virtual unit VF is called, quickly positioning and calling the virtual unit according to the label. For ease of understanding, the following will be exemplified.

Fig. 5 is a structural diagram of another second PCI device model provided in this embodiment of the present application, for example, where a physical unit PF is a GPU, and as shown in fig. 5, virtual units VF corresponding to the GPU are respectively a 2G video memory, a 4G video memory, an 8G video memory, a 16G video memory, and the like, where the 2G video memory is provided with the same tags TtaitsB, and the 4G video memory, the 8G video memory, and the 16G video memory are respectively provided with TtaitsC, TtaitsD, and TtaitsE.

The label may be a number, a word, or a letter, and the specific form of the label is not limited in the present application.

It can be understood that, compared with the second PCI device model, the first PCI device model has general versatility, and the same type of virtual unit VF sets the same tag, which is convenient for managing and maintaining PCI device resources and PCI device usage. The second PCI device model is flexible in configuration, and can be used in a specific application scene when specific devices are needed.

According to the method for uniformly scheduling the PCI equipment resources, when modeling is performed on each PCI equipment according to the virtualization attribute of the PCI equipment to obtain the corresponding PCI equipment model, each PCI equipment is divided into a physical unit and a virtual unit, wherein one PCI equipment is a physical unit, and one physical unit corresponds to a plurality of virtual units. Then, the same label is set for each virtual unit with the same virtualization property corresponding to the physical unit, and different labels are set for each virtual unit with different virtualization property corresponding to the physical unit, so that modeling is performed based on the mapping relationship among each physical unit, the virtual unit corresponding to each physical unit, and the corresponding label to obtain each PCI device model. When PCI equipment is scheduled, the PCI equipment can be called based on each established PCI equipment model, and unified scheduling of all PCI equipment is realized.

In fact, before modeling each PCI device according to its virtualization attribute to obtain a corresponding PCI device model, the virtualization attribute of each PCI device needs to be configured in advance. During configuration, a PCI _ device _ list configuration item is newly added, and it should be noted that the PCI _ device _ list configuration item supports resource _ class and transactions tags of a resource management service module, where the resource management service module is used to create a PCI device model, and uniformly schedule each PCI device resource, and the like. In addition, the PCI _ device _ list configuration item supports PCI device custom resource _ class tags and supports the presentation and management of PCI devices with such tags. It is understood that the transactions tag is used to mark virtualization attributes corresponding to each virtual unit in the PCI device model, and the resource _ class tag is used to mark a group or class of PCI device resources.

In implementation, if the resource requested by the PCI device is not matched in the resource pool, that is, there is no resource _ class matched with the PCI request, the request is rejected, otherwise, the corresponding PCI device resource is provided for the PCI device resource request.

And utilizing the PCI _ device _ list configuration to acquire each PCI device, analyzing each PCI device to determine the type of each PCI device, and configuring the virtualization attribute corresponding to each PCI device in advance according to the type of each PCI device.

It is to be noted that, when the virtualization attributes of the virtual units VF corresponding to the same physical unit PF are the same, the virtual units VF need to be configured in the same pci _ device _ list configuration item, that is, the same class of virtual units VF uses the same tracks tag to track and call. If the virtualization attributes of the virtual units VF corresponding to the same physical unit PF are different, the virtual units VF of the same class need to be configured in the same pci _ device _ list configuration item.

The method for uniformly scheduling the PCI device resources provided in the embodiment of the present application analyzes the obtained PCI devices to determine the types of the PCI devices, and configures the virtualization attributes corresponding to the PCI devices in advance according to the types of the PCI devices, so as to model the PCI devices based on the virtualization attributes corresponding to the PCI devices.

In implementation, after the PCI device resource request instruction is obtained, when PCI device resource allocation is performed based on each PCI device model, the PCI device resource request instruction is firstly analyzed to determine the PCI device resource request type, and the target computing host is located according to the PCI device resource request type. After the target computing host is obtained, it is determined whether an available target PCI device exists in the target computing host, and it is understood that the target PCI device is a device requested by the PCI device resource request instruction. When it is determined that available target PCI equipment exists in the target computing host, a target PCI equipment model corresponding to the target PCI equipment is called, and resource allocation is carried out according to the target PCI equipment model. Of course, if the target PCI device does not exist, the resource allocation is ended.

According to the method for uniformly scheduling the PCI equipment resources, when the target computing host is obtained and the available target PCI equipment is determined to exist based on the target computing host, resource allocation is performed according to the target PCI equipment model corresponding to the target PCI equipment, so that uniform scheduling and allocation of the PCI equipment are achieved based on the PCI equipment model, and scheduling efficiency of the PCI equipment is improved.

Specifically, when resource allocation is performed according to a target PCI device model corresponding to a target PCI device, it is first determined whether a borrowed unit exists in each virtual unit corresponding to a physical unit of the target PCI device model, and if a borrowed unit exists, it is determined whether an available unit exists in each virtual unit that is not borrowed, and if an available unit exists, the corresponding virtual unit is allocated, and if an available unit does not exist, the resource allocation request is rejected. If there is no borrowed unit, the physical unit is directly allocated, and the method of directly allocating the physical unit is called through.

In fact, some cloud host instances need to use the PCI device in a pass-through manner, at which point the virtual unit VF in the PCI device can no longer be borrowed, i.e., the PCI device will track as a separate resource.

That is, when PCI devices are directly used, that is, when a physical unit is directly called, the corresponding virtual unit is not reusable, and therefore, when resource allocation is performed, it is necessary to determine whether a virtual unit corresponding to the physical unit exists and has been borrowed, and if the virtual unit exists and has been borrowed, the physical unit cannot be called in a direct mode.

In the method for uniformly scheduling PCI device resources provided in the embodiments of the present application, when resource allocation is performed according to a target PCI device model corresponding to a target PCI device, if it is determined that a borrowed unit exists in virtual units corresponding to physical units of the target PCI device model, it is further determined whether an available unit exists in virtual units that are not borrowed, if an available unit exists, a corresponding virtual unit is allocated, and if an available unit does not exist, a resource allocation request is rejected. If it is determined that no borrowed unit exists among the virtual units, the physical unit is directly allocated. Therefore, the target PCI equipment is distributed and scheduled based on the target PCI equipment model, and the scheduling reliability of the PCI equipment is improved.

In implementation, when allocating virtual units and physical units, allocation is performed based on a physical unit PF resource tree in a PCI device resource pool. If the virtual unit is allocated, resource allocation is performed according to the identifier corresponding to the virtual unit and the address code (e.g., 0000:84:00:0) of the corresponding physical unit, it should be noted that the identifier corresponding to the virtual unit may be a character string, or may be an alphabet or a number, and the specific form of the identifier is not limited in this application. If the physical unit is allocated, namely the direct connection is used, the resource allocation is carried out according to the address code corresponding to the physical unit.

According to the method for uniformly scheduling the PCI equipment resources, when the virtual units are allocated, the resources are allocated according to the identifiers corresponding to the virtual units and the address codes of the corresponding physical units, and when the physical units are allocated, the resources are allocated according to the address codes corresponding to the physical units, so that the use conditions of the physical units and the virtual units of the PCI equipment are effectively tracked.

In the above embodiment, the target computing host is located according to the PCI device resource request type, and in practice, all the computing hosts need to be filtered through a NUMA affinity policy first, so as to filter the computing hosts that do not satisfy the affinity to obtain the first computing host list, where NUMA affinity refers to deploying cloud host instances to a specific one or more CPUs as dispersedly as possible, so that a memory access path is the shortest, and a server group is set, through which cloud host instances in the group are deployed to the same NUMA node.

And after the first computing host list is obtained, a second computing host list meeting the PCI equipment resource request is obtained from the first computing host list, namely, a target computing host is positioned in the second computing host list according to the PCI equipment resource request type and the utilization rate of each computing host in the second computing host list.

It should be noted that, when the target computing host is located, the host with the highest utilization rate in the second computing host list is used as the target computing host, so that the PCI device resource utilization rate is maximized. For ease of understanding, the following will be exemplified.

Fig. 6 is a structural diagram of a PCI device resource according to an embodiment of the present application, and as shown in fig. 6, for example, the PCI device is a GPU device, the obtained computing hosts that satisfy the request in the second computing host list include a computing host 1 and a computing host 2, and the computing host 1 and the computing host 2 have 3 GPU devices and 2 GPU devices, respectively. The virtual unit 2G video memory and the virtual unit 4G video memory in the GPU0 of the host 1 are occupied, and the virtual unit 4G video memory in the GPU1 and the GPU2 are occupied.

When a user requests GPU resources of 2GB video memory, in order to improve the resource utilization rate of the PCI device resource pool, it is preferable to use the computing host 1 with a high utilization rate as a target computing host according to the usage conditions of the virtual units in the computing host 1 and the computing host 2.

That is, following the principle of maximizing resource utilization, both GPU0 and GPU1 satisfy the PCI device resource request, but since GPU0 has 75% resource utilization and GPU1 has 50% resource utilization, the resources on GPU0 are preferentially allocated.

According to the method for uniformly scheduling the PCI equipment resources, when a target computing host is located, all computing hosts are filtered through a NUMA affinity strategy to obtain a first computing host list, then a second computing host list meeting PCI equipment resource requests is obtained from the first computing host list, and finally the target computing host is located in the second computing host list according to the PCI equipment resource request type, wherein the host with the highest utilization rate in the second computing host list is used as the target computing host. Therefore, when the PCI equipment meeting the request is quickly positioned, the resource utilization rate of the PCI equipment is maximized, and the scheduling reliability of the PCI equipment resources is improved.

In specific implementation, the Open Stack implements a virtualized layer resource management cloud platform based on KVM and Libvirt, and when PCI device resource visualization is implemented, the relevant information of each PCI device in the PCI device resource pool needs to be uploaded to the Open Stack cloud platform every preset time period, so that a user can directly observe the current state of each PCI device through the Open Stack cloud platform.

When the relevant information of each PCI device in the PCI device resource pool is uploaded, a PCI device resource updating interface driven by Libvirt VIRT is called, the PCI device resource updating interface reports the created physical unit PF resources and the corresponding virtual unit VF resources to a resource management service module, and the resource management service module records the current use state of each PCI device, the number of available PCI devices and other information.

When the computing service is restarted, a PCI equipment resource updating interface is called, the Libvirt VIRT driver is used for checking and updating the resource management service module, and therefore the cloud host instance with PCI allocation is ensured to be recorded or allocated to the corresponding PCI equipment resource in the resource management service module.

According to the method for uniformly scheduling the PCI equipment resources, the relevant information of each PCI equipment in the current PCI equipment resource pool is uploaded to the OpenStack cloud platform every other preset time, and therefore visualization of each PCI equipment resource is achieved.

It can be understood that, in order to improve the resource utilization rate, whenever the PCI device resource is used up, the resource needs to be released accordingly, and the information such as the current usage state of the PCI device in the PCI device resource pool, the current available PCI device number, and the like is updated.

According to the method for uniformly scheduling the PCI equipment resources, after the use of each PCI equipment resource is determined to be finished, each PCI equipment resource is released, the related information of each current PCI equipment in the PCI equipment resource pool is updated, and the reduction of the use rate of the PCI equipment caused by the fact that the used PCI equipment resource is not released in time or the related information of each PCI equipment is not updated in time is avoided.

In the foregoing embodiment, a method for uniformly scheduling PCI device resources is described in detail, and the present application also provides an embodiment corresponding to a device for uniformly scheduling PCI device resources. It should be noted that the present application describes the embodiments of the apparatus portion from two perspectives, one is based on the functional module, and the other is based on the hardware structure.

Fig. 7 is a structural diagram of an apparatus for unified scheduling of PCI device resources according to an embodiment of the present application, and as shown in fig. 7, the apparatus includes:

the modeling unit 10 is configured to model each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model.

And the allocating unit 11 is configured to, after the PCI device resource request instruction is obtained, allocate the PCI device resources based on each PCI device model.

An updating unit 12, configured to update relevant information of each PCI device currently in the PCI device resource pool; the relevant information of the PCI devices at least comprises the current use state of each PCI device and the number of available PCI devices.

An uploading unit 13, configured to upload relevant information of each PCI device in the current PCI device resource pool to the Open Stack cloud platform.

Since the embodiments of the apparatus portion and the method portion correspond to each other, please refer to the description of the embodiments of the method portion for the embodiments of the apparatus portion, which is not repeated here.

The device for uniformly scheduling the PCI equipment resources provided by the embodiment of the application comprises: and modeling each PCI device according to the virtualization attribute of the PCI device to obtain a corresponding PCI device model, after obtaining the PCI device resource request instruction, distributing the PCI device resources based on each PCI device model corresponding to each PCI device, and updating the related information of each current PCI device in the PCI device resource pool after distribution, wherein the related information of the PCI devices at least comprises the current use state of each PCI device and the number of available PCI devices. And uploading the relevant information of each PCI device in the current PCI device resource pool to the OpenStack cloud platform after updating. Therefore, according to the technical scheme provided by the application, each PCI device is modeled based on the virtualization attribute of the PCI device, and unified scheduling and distribution of all PCI devices are realized according to each PCI device model. In addition, the relevant information of each PCI device in the current PCI device resource pool is uploaded to the OpenStack cloud platform, and the relevant information of each PCI device is visualized on the OpenStack cloud platform, namely the information such as the current use state and the number of available PCI devices of each PCI device is visualized, so that the visualization of the PCI device resource state and the unified management and scheduling of the PCI device resources are realized, the resource consumption is reduced, and the operation and maintenance cost of the PCI devices when the PCI devices are used on the OpenStack cloud platform is reduced.

Fig. 8 is a structural diagram of an apparatus for uniformly scheduling PCI device resources according to another embodiment of the present application, and as shown in fig. 8, the apparatus for uniformly scheduling PCI device resources includes: a memory 20 for storing a computer program;

the processor 21 is configured to implement the steps of the method for uniformly scheduling PCI device resources according to the above embodiment when executing the computer program.

The device for uniformly scheduling PCI device resources provided in this embodiment may include, but is not limited to, a smart phone, a tablet computer, a notebook computer, or a desktop computer.

The processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The Processor 21 may be implemented in at least one hardware form of a Digital Signal Processor (DSP), a Field-Programmable Gate Array (FPGA), and a Programmable Logic Array (PLA). The processor 21 may also include a main processor and a coprocessor, where the main processor is a processor for Processing data in an awake state, and is also called a Central Processing Unit (CPU); a coprocessor is a low power processor for processing data in a standby state. In some embodiments, the processor 21 may be integrated with an image processor (GPU), and the GPU is responsible for rendering and drawing the content required to be displayed by the display screen. In some embodiments, the processor 21 may further include an Artificial Intelligence (AI) processor for processing computing operations related to machine learning.

The memory 20 may include one or more computer-readable storage media, which may be non-transitory. Memory 20 may also include high speed random access memory, as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used for storing the following computer program 201, where after being loaded and executed by the processor 21, the computer program is capable of implementing relevant steps of the method for unified scheduling of PCI device resources disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202, data 203, and the like, and the storage manner may be a transient storage manner or a permanent storage manner. Operating system 202 may include, among others, Windows, Unix, Linux, and the like. Data 203 may include, but is not limited to, relevant data involved in the method of unified scheduling of PCI device resources.

In some embodiments, the PCI device resource uniform scheduling apparatus may further include a display 22, an input/output interface 23, a communication interface 24, a power supply 25, and a communication bus 26.

Those skilled in the art will appreciate that the architecture shown in FIG. 8 does not constitute a limitation of an apparatus for unified scheduling of PCI device resources and may include more or fewer components than those shown.

The device for uniformly scheduling the PCI device resources provided by the embodiment of the present application includes a memory and a processor, and when the processor executes a program stored in the memory, the following method can be implemented: a method for unified scheduling of PCI device resources.

The device for uniformly scheduling the PCI equipment resources, provided by the embodiment of the application, models each PCI equipment based on the virtualization attribute of the PCI equipment, and uniformly schedules and allocates all the PCI equipment according to each PCI equipment model. In addition, relevant information of each PCI device in the current PCI device resource pool is uploaded to the OpenStack cloud platform, and the OpenStack cloud platform visualizes the relevant information of each PCI device, namely visualizes the current use state of each PCI device, the number of available PCI devices and other information, so that the visualization of the PCI device resource state and the unified management and scheduling of the PCI device resources are realized, the resource consumption is reduced, and the operation and maintenance cost of the PCI devices when the PCI devices are used on the OpenStack cloud platform is reduced.

Finally, the application also provides a corresponding embodiment of the computer readable storage medium. The computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps as set forth in the above-mentioned method embodiments.

It is to be understood that if the method in the above embodiments is implemented in the form of software functional units and sold or used as a stand-alone product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, which are essential or part of the prior art, or all or part of the technical solutions may be embodied in the form of a software product, which is stored in a storage medium and executes all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The method, the apparatus, and the medium for unified scheduling of PCI device resources provided in the present application are described in detail above. The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description. It should be noted that, for those skilled in the art, it is possible to make several improvements and modifications to the present application without departing from the principle of the present application, and such improvements and modifications also fall within the scope of the claims of the present application.

It is further noted that, in the present specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

Claims

1. A method for unified scheduling of PCI device resources is characterized by comprising the following steps:

2. The method of claim 1, wherein the modeling each PCI device according to its virtualization attributes to obtain a corresponding PCI device model comprises:

3. The method for uniformly scheduling PCI device resources according to claim 2, wherein before the modeling each PCI device according to the virtualization attributes of the PCI device to obtain the corresponding PCI device model, the method further comprises:

4. The method according to claim 2, wherein the allocating PCI device resources based on each PCI device model after the PCI device resource request instruction is obtained comprises:

judging whether available target PCI equipment exists in the target computing host; wherein the target PCI device is a device requested by the PCI device resource request instruction;

if not, ending the resource allocation.

5. The method according to claim 4, wherein the allocating resources according to the target PCI device model corresponding to the target PCI device comprises:

judging whether borrowed units exist in virtual units corresponding to physical units of the target PCI equipment model or not;

if there are no borrowed units, then the physical units are allocated directly.

6. The method according to claim 5, wherein the allocating PCI device resources based on each PCI device model comprises:

7. The method of claim 4, wherein prior to said locating a target computing host according to the PCI device resource request type, further comprising:

8. The method of claim 7, wherein the locating the target computing host in the second computing host list according to the PCI device resource request type comprises:

9. The method according to claim 1, wherein the uploading the relevant information of each PCI device in the PCI device resource pool to an Open Stack cloud platform comprises:

10. The method for unified scheduling of PCI device resources according to claim 1, further comprising:

11. An apparatus for unified scheduling of PCI device resources, comprising:

12. An apparatus for unified scheduling of PCI device resources, comprising a memory for storing a computer program;

a processor for implementing the steps of the method of PCI device resource unified scheduling according to any of claims 1 to 10 when executing said computer program.

13. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method for unified scheduling of PCI device resources according to any of claims 1 to 10.