Centrless cloud computing architecture, method and apparatus

Abstract

This disclosure provides a decentralized cloud computing architecture, method, and apparatus, relating to the field of cloud computing technology. The architecture includes: a decentralized hardware layer comprising multiple hardware components, including a general-purpose computing processor and at least one domain-specific accelerator (DSA), with any two hardware components interconnected via a switch; and an operating system layer comprising: an operating system (OS) mounted on the general-purpose computing processor, and at least one OS agent, each corresponding to one of the at least one DSA. Through the collaboration of the decentralized hardware layer and the operating system layer at both the hardware and software levels, a centralized cloud computing architecture is constructed. Thus, neither the general-purpose computing processor nor the DSA needs to communicate via a central hub.

Description

Technical Field

[0001] This disclosure relates to the field of cloud computing technology, and in particular to a decentralized cloud computing architecture, method, and apparatus. Background Technology

[0002] With the continuous development of cloud computing, users' demand for high-quality cloud computing services is constantly growing. As a result, a series of Domain Customization Accelerators (DSAs) have emerged to improve the quality of cloud computing services. Summary of the Invention

[0003] In related technologies, cloud computing architecture requires a central processor such as a central processing unit (CPU) or a data processing unit (DPU). Data communication of DSAs needs to be proxied and controlled by the central processor. As the types and numbers of DSAs increase, the central processor needs to adapt to an increasing number of DSA types and quantities, and the central processor has gradually become a performance bottleneck for cloud computing services.

[0004] In view of this, embodiments of the present disclosure propose a decentralized cloud computing architecture.

[0005] According to one aspect of the present disclosure, a decentralized cloud computing architecture is provided, comprising: a decentralized hardware layer, including multiple hardware components, the multiple hardware components including a general-purpose computing processor and at least one domain-specific accelerator (DSA), wherein any two of the multiple hardware components are interconnected via a switch; and an operating system layer, including an operating system (OS) mounted on the general-purpose computing processor, and at least one OS agent, each corresponding to one of the at least one DSA.

[0006] In some embodiments, the at least one DSA includes multiple DSAs, and the at least one OS agent includes multiple OS agents.

[0007] In some embodiments, the architecture further includes a runtime system mounted on the OS and including an API interface configured to receive requests for cloud computing tasks, the requests including the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources; and a task mapper configured to analyze the request to determine the DSA to be invoked to satisfy the request, so as to invoke the required DSA to complete the cloud computing task.

[0008] In some embodiments, the runtime system further includes a manager, and the task mapper is configured to use the manager to call the corresponding control stack in the OS, and control the corresponding stack in the OS agent on which the DSA to be called is mounted, so as to call the DSA to be called.

[0009] In some embodiments, the manager includes a memory manager, a network manager, and a storage manager, and the control stack includes a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager. The stack includes a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

[0010] In some embodiments, the two hardware devices are interconnected using a compute fast link protocol or a cache coherent interconnect protocol.

[0011] In some embodiments, the general-purpose computing processor includes a central processing unit (CPU), and the at least one DSA includes one or more of a graphics processing unit (GPU), a tensor processor (TPU), a programmable gate array (FPGA), a video encoding processor (VCU), a reconfigurable data stream processor (RDU), a data processor (DPU), a solid-state drive (SSD), a cloud infrastructure processor (CIPU), and a heterogeneous processor (HPU).

[0012] According to another aspect of the embodiments of this disclosure, a decentralized cloud computing method is provided, comprising: analyzing a request for a cloud computing task from an API interface to determine a Domain-Customized Accelerator (DSA) that needs to be invoked to satisfy the request, the request including the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources, the API interface being located in a runtime system, the runtime system being mounted on an operating system (OS) in an operating system layer, the OS being mounted on a general-purpose computing processor in a plurality of hardware in a decentralized hardware layer, the plurality of hardware further including at least one DSA, the at least one DSA being mounted with at least one OS agent in the operating system layer, and any two hardware devices being interconnected via a switch; invoking the DSA to be invoked to complete the cloud computing task.

[0013] In some embodiments, the at least one DSA includes multiple DSAs, and the at least one OS agent includes multiple OS agents.

[0014] In some embodiments, the runtime system further includes a manager that uses the manager to call the corresponding control stack in the OS to control the corresponding stack in the OS agent on which the DSA to be called is mounted, so as to call the DSA to be called.

[0015] In some embodiments, the manager includes a memory manager, a network manager, and a storage manager, and the control stack includes a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager. The stack includes a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

[0016] In some embodiments, the two hardware devices are interconnected using a compute fast link protocol or a cache coherent interconnect protocol.

[0017] According to another aspect of the embodiments of this disclosure, a decentralized cloud computing device is provided, comprising: an analysis module configured to analyze a request for a cloud computing task from an API interface to determine a DSA that needs to be invoked to satisfy the request, the request including the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources, the API interface being located in a runtime system, the runtime system being mounted on an operating system OS in an operating system layer, the OS being mounted on a general-purpose computing processor in a plurality of hardware in a decentralized hardware layer, the plurality of hardware further including at least one DSA, each of the at least one DSA being mounted with at least one OS agent in the operating system layer, and any two of the plurality of hardware being interconnected via a switch; and an invocation module configured to invoke the DSA to be invoked to complete the cloud computing task.

[0018] According to another aspect of the present disclosure, a decentralized cloud computing device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the method described in any of the above embodiments based on instructions stored in the memory.

[0019] According to another aspect of the present disclosure, a computer-readable storage medium is provided, including computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method described in any of the above embodiments.

[0020] In this embodiment of the disclosure, a centralized cloud computing architecture is constructed through the collaboration of a decentralized hardware layer and an operating system layer at both the hardware and software levels. Thus, neither general-purpose computing processors nor DSAs can communicate data through a central hub.

[0021] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1A This is a schematic diagram of a decentralized cloud computing architecture according to some embodiments of the present disclosure;

[0024] Figure 1B These are detailed schematic diagrams of a decentralized cloud computing architecture according to some embodiments of this disclosure;

[0025] Figure 2 This is a schematic diagram of a decentralized cloud computing architecture according to other embodiments of this disclosure;

[0026] Figure 3 This is a schematic diagram of a runtime system according to some embodiments of the present disclosure;

[0027] Figure 4 This is a schematic diagram of an operating system (OS) and an OS agent according to some embodiments of this disclosure;

[0028] Figure 5 This is a flowchart illustrating a decentralized cloud computing method according to some embodiments of the present disclosure;

[0029] Figure 6 This is a schematic diagram of the structure of a decentralized cloud computing device according to some embodiments of the present disclosure;

[0030] Figure 7 This is a schematic diagram of the structure of a decentralized cloud computing device according to other embodiments of the present disclosure. Detailed Implementation

[0031] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0032] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0033] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0034] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0035] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0036] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0037] Furthermore, in the description of this disclosure, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or order. Similarly, although operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring such operations to be performed in the specific order shown or in sequential order, or requiring the execution of all illustrated operations to achieve the desired result. In some cases, multitasking and parallel processing can be advantageous.

[0038] Figure 1A This is a schematic diagram of a decentralized cloud computing architecture according to some embodiments of the present disclosure.

[0039] The decentralized cloud computing architecture 100 includes a decentralized hardware layer 101 and an operating system layer 102. Here, the decentralized hardware layer 101 includes multiple hardware components, including a general-purpose computing processor and at least one DSA, and any two of these hardware components are interconnected via a switch; the operating system layer 102 includes an OS mounted on the general-purpose computing processor and at least one OS agent that is mounted on at least one DSA.

[0040] In the above embodiments, a centralized cloud computing architecture is constructed through the collaboration of a decentralized hardware layer and an operating system layer at both the hardware and software levels. Thus, neither general-purpose computing processors nor DSAs need to go through a central hub for data communication.

[0041] In some embodiments, at least one DSA includes multiple DSAs, and at least one OS agent includes multiple OS agents.

[0042] It should be understood that OS and OS agents, as well as different OS agents, can communicate data through physical links formed by interconnecting corresponding hardware via switches, such as for the exchange of control flow and data flow.

[0043] In some embodiments, the general-purpose computing processor may include a central processing unit (CPU), and at least one DSA may include one or more of a graphics processing unit (GPU), a tensor processor (TPU), a programmable gate array (FPGA), a video encoding processor (VCU), a reconfigurable data stream processor (RDU), a digital processing unit (DPU), a solid-state drive (SSD), a cloud infrastructure processor (CIPU), and a heterogeneous processor (HPU).

[0044] In some embodiments, any two pieces of hardware among a plurality of hardware can be interconnected using the Compute Fast Link Protocol (CXL) or the Cache Coherent Interconnect Protocol (CCIX).

[0045] As some implementation methods, the hardware connectivity and OS / OS agent configuration in the decentralized cloud computing architecture 100 can be as follows: Figure 1B As shown in the diagram. The CPU carries a general-purpose OS (i.e., OS), while the VPU, GPU, FPGA, and DPU each carry an OS Agent (i.e., OS agent). Here, "other" refers to other DSAs.

[0046] Figure 2 This is a schematic diagram of a decentralized cloud computing architecture according to other embodiments of this disclosure.

[0047] Compared to the decentralized cloud computing architecture 100, the decentralized cloud computing architecture 200 also includes a runtime system 201 mounted on the OS.

[0048] The runtime system 201 includes an API interface and a task mapper. The API interface is configured to receive requests for cloud computing tasks. The task mapper is configured to analyze the request to determine the DSAs that need to be invoked to satisfy the request, so as to invoke the required DSAs to complete the cloud computing task. Here, the request includes the type of DSA to be invoked, the unit of the resource to be invoked, and the number of resources to be invoked.

[0049] Thus, by receiving and analyzing requests from cloud computing tasks to determine the DSA that needs to be invoked to satisfy the request, DSA invocation can be achieved without going through a central hub for data communication.

[0050] As one implementation method, the collaboration method and data flow links between different DSAs can be pre-set. That is, when the DSA that needs to be called to meet the request is determined, the cloud computing task can be completed with the pre-set collaboration method and data flow links.

[0051] In some embodiments, the type of DSA may include one or more of GPU, TPU, FPGA, VCU, RDU, DPU, SSD, CIPU, and HPU.

[0052] In some embodiments, the unit of the resource may be, for example, a half-precision floating-point number (FP16), a single-precision floating-point number (FP32), or a double-precision floating-point number (FP64).

[0053] In some embodiments, the runtime system may also include a manager, and the task mapper may be configured to use the manager to call the corresponding control stack in the OS, and control the corresponding stack in the OS agent on which the DSA to be called is carried, so as to call the DSA to be called.

[0054] In this way, by using the manager to call the control stack, the stack corresponding to the DSA that needs to be called can be controlled to call the required DSA. This allows for accurate calling of DSAs without having to go through a central hub for data communication.

[0055] As one implementation, the corresponding control stack in the OS is stored in the OS's DSA Lib library.

[0056] In some embodiments, the OS may store DSA drivers to drive the corresponding DSA to run.

[0057] In some embodiments, such as Figure 3 As shown, the managers can include a memory manager, a network manager, and a storage manager.

[0058] In some embodiments, such as Figure 4 As shown, the control stack (not shown) in the DSA Lib library may include a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager; the stack may include a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

[0059] Figure 5 This is a flowchart illustrating a decentralized cloud computing method according to some embodiments of the present disclosure. The cloud computing method may, for example, be executed by a task mapper.

[0060] In step 502, the requests from cloud computing tasks originating from the API interface are analyzed to determine the DSAs that need to be invoked to satisfy the requests. Here, the request includes the type of DSA to be invoked, the unit of the resource to be invoked, and the number of resources. The API interface resides in the runtime system, which is mounted on the OS in the operating system layer. The OS is mounted on a general-purpose computing processor in multiple hardware devices in the decentralized hardware layer. The multiple hardware devices also include at least one DSA, and each DSA has at least one OS agent in the operating system layer. Any two hardware devices are interconnected via a switch.

[0061] In step 504, the required DSA is invoked to complete the cloud computing task.

[0062] Thus, by analyzing requests from cloud computing tasks via API interfaces to determine the DSA that needs to be invoked to fulfill the request, DSA invocation can be achieved without requiring data communication through a central hub.

[0063] In some embodiments, at least one DSA may include multiple DSAs, and at least one OS agent may include multiple OS agents.

[0064] In some embodiments, the runtime system may further include a manager that can use the manager to call the corresponding control stack in the OS and control the corresponding stack in the OS agent of the DSA to be called, so as to call the DSA to be called.

[0065] In this way, by using the manager to call the control stack, the stack corresponding to the DSA that needs to be called can be controlled to call the required DSA. This allows for accurate calling of DSAs without having to go through a central hub for data communication.

[0066] In some embodiments, the manager may include a memory manager, a network manager, and a storage manager; the control stack may include a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager; the stack may include a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

[0067] In some embodiments, any two hardware devices are interconnected using CXL or CCIX.

[0068] Some embodiments of this disclosure also provide cloud computing methods executed by various components in the above-described cloud computing architecture. Specific operations can be referred to the functions executed by each component, and will not be repeated here.

[0069] The following describes a decentralized cloud computing process according to some embodiments of this disclosure.

[0070] Step 1, Initialization. Interconnect the general-purpose computing processor and any two DSAs via a switch according to CXL or CCIX, deploy the OS on the general-purpose computing processor, deploy the OS agent on the DSA, and install the runtime system on the OS.

[0071] Step 2: Users deploy cloud computing tasks via the DSA programming / configuration API interface. First, users can select the type of DSA to be invoked through the application via the DSA programming API interface. Then, users can select the unit and number of resources to be invoked through the DSA configuration API interface.

[0072] Step 3: The task mapper analyzes the cloud computing tasks, senses the remaining DSA resources, decomposes and maps the computing tasks, generates a mapping and scheduling strategy, and calls the network, memory and storage managers to schedule different DSA acceleration tasks to the corresponding DSAs according to their resource requirements.

[0073] Step 4: The network, memory, and storage managers call the DSA Lib library to interact with control flow data and determine the collaboration method and data flow path. When calling heterogeneous DSAs (such as FPGA+GPU), the heterogeneous DSA peer-to-peer interaction collaboration method is automatically enabled, without the need for a general-purpose computing processor agent.

[0074] Step 5: The general-purpose computing processor interacts with the DSA to perform data stream interactions and calculations, and returns the calculation results.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to mutually. For the apparatus embodiments, since they largely correspond to the method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0076] Figure 6 This is a schematic diagram of the structure of a decentralized cloud computing device according to some embodiments of the present disclosure.

[0077] like Figure 6 As shown, the decentralized cloud computing device includes an analysis module 601 and a calling module 602.

[0078] Analysis module 601 is configured to analyze requests from cloud computing tasks originating from the API interface to determine the DSAs that need to be invoked to satisfy the requests. Here, the request includes the type of DSA to be invoked, the unit of the resource to be invoked, and the number of resources. The API interface resides in a runtime system, which is mounted on an OS within the operating system layer. The OS is mounted on a general-purpose computing processor in multiple hardware components within a decentralized hardware layer. Each hardware component also includes at least one DSA, and each DSA has a corresponding OS agent within the operating system layer. Any two hardware components are interconnected via a switch.

[0079] Module 602 is configured to invoke the required DSA to complete the cloud computing task.

[0080] In some embodiments, the decentralized cloud computing device may further include other modules to execute the decentralized cloud computing method of any of the above embodiments. In some embodiments, the decentralized cloud computing device is located in a task mapper.

[0081] Figure 7 This is a schematic diagram of the structure of a decentralized cloud computing device according to some embodiments of the present disclosure.

[0082] like Figure 7 As shown, the decentralized cloud computing device 700 includes a memory 701 and a processor 702 coupled to the memory 701. The processor 702 is configured to execute the method of any of the foregoing embodiments based on instructions stored in the memory 701.

[0083] The memory 701 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, and other programs.

[0084] The decentralized cloud computing device 700 may also include an input / output interface 703, a network interface 704, and a storage interface 705. The input / output interface 703, network interface 704, and storage interface 705, as well as the memory 701 and processor 702, can be connected via, for example, a bus 706. The input / output interface 703 provides a connection interface for input / output devices such as monitors, mice, keyboards, and touchscreens. The network interface 704 provides a connection interface for various networked devices. The storage interface 705 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0085] This disclosure also provides a computer-readable storage medium including computer program instructions that, when executed by a processor, implement the method of any of the above embodiments.

[0086] The embodiments of this disclosure have now been described in detail. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

[0087] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that the functions specified in one or more flowchart illustrations and / or one or more block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0091] While specific embodiments of this disclosure have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments or equivalent substitutions can be made to some technical features without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A decentralized cloud computing architecture, comprising: A decentralized hardware layer includes multiple hardware components, including a general-purpose computing processor and at least one domain-specific accelerator (DSA), wherein any two hardware components are interconnected via a switch. The operating system layer includes: Operating system (OS), mounted on the general-purpose computing processor, and At least one OS agent is mounted on each of the at least one DSA; The runtime system, running on the OS, includes: API interfaces, a task mapper, and a manager. The API interface is configured to receive requests from cloud computing tasks, the requests including the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources. The task mapper is configured to analyze the request to determine the DSA that needs to be invoked to satisfy the request, and use the manager to invoke the corresponding control stack in the OS to control the corresponding stack in the OS agent on which the DSA needs to be invoked, so as to invoke the DSA to complete the cloud computing task.

2. The architecture of claim 1, wherein, The at least one DSA includes multiple DSAs, and the at least one OS agent includes multiple OS agents.

3. The architecture of claim 1, wherein, The manager includes a memory manager, a network manager, and a storage manager. The control stack includes a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager. The stack includes a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

4. The architecture according to claim 1, wherein, The two hardware devices are interconnected using a compute fast link protocol or a cache coherence interconnect protocol.

5. The architecture according to claim 1, wherein, The general-purpose computing processor includes a central processing unit (CPU), and the at least one DSA includes one or more of a graphics processing unit (GPU), a tensor processor (TPU), a programmable gate array (FPGA), a video encoding processor (VCU), a reconfigurable data stream processor (RDU), a data processing unit (DPU), a solid-state drive (SSD), a cloud infrastructure processor (CIPU), and a heterogeneous processor (HPU).

6. A decentralized cloud computing method, comprising: The system analyzes cloud computing task requests from an API interface to determine the domain-specific accelerator (DSA) that needs to be invoked to satisfy the request. The request includes the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources. The API interface is located in a runtime system, which is run on an operating system (OS) in the operating system layer. The OS is run on a general-purpose computing processor in a decentralized hardware layer. The multiple hardware components also include at least one DSA, and each DSA has at least one OS agent in the operating system layer. Any two hardware components are interconnected via a switch. The runtime system manager calls the corresponding control stack in the OS to control the corresponding stack in the OS agent of the DSA to be called, so as to call the DSA to complete the cloud computing task.

7. The method according to claim 6, wherein, The at least one DSA includes multiple DSAs, and the at least one OS agent includes multiple OS agents.

8. The method according to claim 6, wherein, The manager includes a memory manager, a network manager, and a storage manager. The control stack includes a network control stack corresponding to the network manager, a storage control stack corresponding to the storage manager, and a memory control stack corresponding to the memory manager. The stack includes a network stack corresponding to the network control stack, a storage stack corresponding to the storage control stack, and a memory stack corresponding to the memory control stack.

9. The method according to claim 6, wherein, The two hardware devices are interconnected using a compute fast link protocol or a cache coherence interconnect protocol.

10. A decentralized cloud computing device, comprising: An analysis module is configured to analyze requests from cloud computing tasks originating from an API interface to determine the DSAs that need to be invoked to satisfy the requests. The requests include the type of DSA to be invoked, the unit of the resource to be invoked, and the number of the resources. The API interface resides in a runtime system, which is mounted on an operating system (OS) in the operating system layer. The OS is mounted on a general-purpose computing processor in a decentralized hardware layer. The multiple hardware components also include at least one DSA, and each DSA has at least one OS agent in the operating system layer mounted on it. Any two of the multiple hardware components are interconnected via a switch. The calling module is configured to use the runtime system manager to call the corresponding control stack in the OS, control the corresponding stack in the OS agent carried by the DSA to be called, so as to call the DSA to complete the cloud computing task.

11. A decentralized cloud computing device, comprising: Memory; as well as A processor coupled to the memory is configured to perform the method of any one of claims 6-9 based on a request stored in the memory.

12. A computer-readable storage medium comprising computer program instructions, wherein, The computer program requests that, when executed by the processor, the method described in any one of claims 6-9 be implemented.

Images