PCIE service cluster starting method and PCIE service cluster

Abstract

The invention provides a PCIE (Peripheral Component Interface Express) service cluster starting method and a PCIE service cluster, and the PCIE service cluster comprises a cluster management server, a plurality of management servers, a plurality of CPUs (Central Processing Unit) and a plurality of high-performance microprocessors; the physical topology of the PCIE service cluster is divided into four layers which are a high-performance microprocessor, a CPU (Central Processing Unit), a management server and a cluster management server in sequence from bottom to top; the method comprises the following steps: starting a cluster management server; after the cluster management server is started, the plurality of management servers are started; after the management server is started, the CPU is started; after the CPU is started, the high-performance microprocessor is started. According to the method provided by the invention, through four layers of physical topology, the problems that an existing PCIE service cluster is slow in starting, the starting of each PCIE device cannot be effectively monitored, the starting sequence is inconsistent, and network congestion is easily caused are effectively solved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method for starting a PCIE service cluster and a PCIE service cluster. Background Art

[0002] PCIE (Peripheral Component Interconnect Express) is a high-speed serial computer expansion bus standard, which is widely used in the connection of various computing hardware devices, especially in large computing servers with high speed and throughput requirements.

[0003] There are usually hundreds or even thousands of PCIE devices interconnected in large computing servers. With so many PCIE devices ready to start, there are problems such as slow startup of the existing PCIE service cluster, inability to effectively monitor the startup of each PCIE device, inconsistent startup order, and easy network congestion. Summary of the invention

[0004] In order to solve one of the above technical defects, the present application provides a method for starting a PCIE service cluster and a PCIE service cluster.

[0005] In a first aspect of the present application, a method for starting a PCIE service cluster is provided, wherein the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors;

[0006] The physical topology of the PCIE service cluster is 4 layers, from bottom to top, high-performance microprocessor, CPU, management server, cluster management server;

[0007] Methods include:

[0008] The cluster management server is started;

[0009] After the cluster management server is started, multiple management servers are started;

[0010] After the management server is started, the CPU starts;

[0011] After the CPU is started, the high-performance microprocessor starts.

[0012] Optionally, every eight high-performance microprocessors are interconnected with the same CPU via a PCIE Switch;

[0013] Every 8 CPUs are connected to the same management server through IB;

[0014] All management servers are connected to the cluster management server through IB.

[0015] Optionally, the logical topology of the PCIE service cluster is 5 layers;

[0016] Among them, the bottom layer of the logical topology corresponds to the bottom layer of the physical topology;

[0017] The second bottom layer of the logical topology corresponds to the second bottom layer of the physical topology;

[0018] The top level of the logical topology corresponds to the top level of the physical topology;

[0019] The second top layer of the logical topology and the next layer of the second top layer correspond to the second top layer of the physical topology.

[0020] Optionally, the next layer below the second top layer of the logical topology corresponds to all management servers;

[0021] The second top layer of the logical topology corresponds to one of each group of management servers, wherein each group of management servers includes eight management servers, and the management servers in each group of management servers are different from each other;

[0022] Each management server in the second top layer of the logical topology is used to manage other management servers in its group and interact with the top layer.

[0023] Optionally, after the cluster management server is started, multiple management servers are started, including:

[0024] After the cluster management server is started, the management server corresponding to the second top layer of the logical topology is started;

[0025] After the management server corresponding to the second top layer of the logical topology is started, the management server corresponding to the next layer of the second top layer of the logical topology is started.

[0026] Optionally, a high performance microprocessor is enabled, including:

[0027] High-performance microprocessor performs link training;

[0028] If the training is unsuccessful, the high-performance microprocessor determines the startup result as a startup failure;

[0029] If the training is successful, the high-performance microprocessor downloads the firmware and starts up to obtain a startup result; wherein the startup result is a startup success or a startup failure;

[0030] The high-performance microprocessor reports the startup results to the CPU;

[0031] The CPU reports the startup result to the management server;

[0032] The management server reports the startup result to the cluster management server;

[0033] After the cluster management server determines that all high-performance microprocessors have been successfully started according to the startup results, it determines that the startup is normal and the computing state is ready; otherwise, it determines that the startup is abnormal and reports the abnormality.

[0034] In a second aspect of the present application, a PCIE service cluster is provided, wherein the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors;

[0035] The physical topology of the PCIE service cluster is 4 layers, from bottom to top, high-performance microprocessor, CPU, management server, cluster management server;

[0036] The PCIE service cluster is started by the method according to any one of claims 1 to 6.

[0037] Optionally, every eight high-performance microprocessors are interconnected with the same CPU via a PCIE Switch;

[0038] Every 8 CPUs are connected to the same management server through IB;

[0039] All management servers are connected to the cluster management server through IB.

[0040] Optionally, the logical topology of the PCIE service cluster is 5 layers;

[0041] Among them, the bottom layer of the logical topology corresponds to the bottom layer of the physical topology;

[0042] The second bottom layer of the logical topology corresponds to the second bottom layer of the physical topology;

[0043] The top level of the logical topology corresponds to the top level of the physical topology;

[0044] The second top layer of the logical topology and the next layer of the second top layer correspond to the second top layer of the physical topology.

[0045] Optionally, the next layer below the second top layer of the logical topology corresponds to all management servers;

[0046] The second top layer of the logical topology corresponds to one of each group of management servers, wherein each group of management servers includes eight management servers, and the management servers in each group of management servers are different from each other;

[0047] Each management server in the second top layer of the logical topology is used to manage other management servers in its group and interact with the top layer.

[0048] The present application provides a method for starting a PCIE service cluster and a PCIE service cluster, wherein the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors; the physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessors, CPUs, management servers, and cluster management servers from bottom to top; the method includes: starting the cluster management server; after the cluster management server is started, multiple management servers are started; after the management server is started, the CPU is started; after the CPU is started, the high-performance microprocessor is started. The method provided by the present application effectively solves the problems of slow startup of the existing PCIE service cluster, inability to effectively monitor the startup of each PCIE device, inconsistent startup order, and easy network congestion through a 4-layer physical topology. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0050] Figure 1 A physical topology diagram of a PCIE service cluster provided in an embodiment of the present application;

[0051] Figure 2 A logical topology diagram of a PCIE service cluster provided in an embodiment of the present application;

[0052] Figure 3 A schematic diagram of the interconnection between a high-performance microprocessor and a CPU in a logical topology of a PCIE service cluster provided in an embodiment of the present application;

[0053] Figure 4 A schematic diagram of the interconnection between a CPU and a secondary management server in a logical topology of a PCIE service cluster provided in an embodiment of the present application;

[0054] Figure 5 A schematic diagram of the interconnection between a secondary management server and a primary management server in a logical topology of a PCIE service cluster provided in an embodiment of the present application;

[0055] Figure 6 A schematic diagram of the interconnection between a primary management server and a cluster management server in a logical topology of a PCIE service cluster provided in an embodiment of the present application;

[0056] Figure 7 A schematic diagram of a process for starting a PCIE service cluster provided in an embodiment of the present application;

[0057] Figure 8A schematic diagram of the startup sequence of a PCIE service cluster provided in an embodiment of the present application;

[0058] Fig. 9 A schematic diagram of a startup process of a high-performance microprocessor provided in an embodiment of the present application;

[0059] Fig.10 A schematic diagram of a startup status reporting process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the technical solutions and advantages in the embodiments of the present application more clearly understood, the exemplary embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0061] In the process of implementing this application, the inventor found that there are usually hundreds or thousands of PCIE devices interconnected in large computing servers. With so many PCIE devices ready to start, there are problems such as slow startup of the existing PCIE service cluster, inability to effectively monitor the startup of each PCIE device, inconsistent startup order, and easy network congestion.

[0062] In view of the above problems, the present application provides a method for starting a PCIE service cluster and a PCIE service cluster, wherein the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors; the physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessors, CPUs, management servers, and cluster management servers from bottom to top; the method includes: starting the cluster management server; after the cluster management server is started, multiple management servers are started; after the management server is started, the CPU is started; after the CPU is started, the high-performance microprocessor is started. The method provided by the present application effectively solves the problems of slow startup of the existing PCIE service cluster, inability to effectively monitor the startup of each PCIE device, inconsistent startup sequence, and easy network congestion through a 4-layer physical topology.

[0063] This embodiment provides a method for starting a PCIE (Peripheral Component Interconnect Express, a high-speed serial computer expansion bus standard) service cluster, and the method can be used to start the PCIE service cluster.

[0064] The physical topology of the PCIE service cluster is as follows: Figure 1As shown, the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs (Central Processing Units), and multiple high-performance microprocessors.

[0065] The physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessor, CPU, management server, and cluster management server from bottom to top.

[0066] The PCIE service cluster is started by the PCIE service cluster starting method provided by this embodiment.

[0067] Among them, every 8 high-performance microprocessors are connected to the same CPU through PCIE Switch.

[0068] Every 8 CPUs are connected to a management server via IB (InfiniBand Line).

[0069] All management servers are connected to the cluster management server through IB.

[0070] For example, the PCIE service cluster has a four-layer physical architecture. At the bottom layer are 2048 high-performance microprocessors. Among the 2048 high-performance microprocessors, every 8 high-performance microprocessors are connected to the same CPU through PCIE Switch. Among the 256 CPUs, every 8 CPUs are connected to the same management server through IB, and all 32 management servers are connected to the same cluster management server through IB.

[0071] The architecture can be effectively expanded by simply increasing the number of management servers, from 2048 to 3072, to 4096, and so on.

[0072] The logical topology of the PCIE service cluster is as follows: Figure 2 As shown, the logical topology of the PCIE service cluster is 5 layers.

[0073] Among them, the bottom layer of the logical topology corresponds to the bottom layer of the physical topology.

[0074] The second bottom layer of the logical topology corresponds to the second bottom layer of the physical topology.

[0075] The top level of the logical topology corresponds to the top level of the physical topology.

[0076] The second top layer of the logical topology and the next layer of the second top layer correspond to the second top layer of the physical topology.

[0077] In addition, the next layer of the second top layer of the logical topology corresponds to all management servers (i.e. Figure 2 Secondary management server in the .

[0078] The second top layer of the logical topology corresponds to one of each group of management servers, wherein each group of management servers includes eight management servers, and the management servers in each group of management servers are different from each other.

[0079] The management servers in the second top layer of the logical topology (i.e. Figure 2 A first-level management server in a group) that manages other management servers in its group and interacts with the top level.

[0080] That is to say, the high-performance microprocessor in the physical topology also constitutes the high-performance microprocessor in the logical topology, the CPU in the physical topology also constitutes the CPU in the logical topology, and the cluster management server in the physical topology also constitutes the cluster management server in the logical topology. The management servers in the physical topology also constitute the management servers of the next layer of the second top layer in the logical topology, but one management server out of every eight management servers also constitutes the management server of the second top layer in the logical topology. In order to facilitate the distinction between different management servers, this embodiment and subsequent embodiments name the management servers that do not constitute the second top layer in the logical topology (that is, the management servers that only constitute the next layer of the second top layer in the logical topology) as secondary management servers; and name the management servers that both constitute the next layer of the second top layer in the logical topology and constitute the second top layer in the logical topology as primary management servers.

[0081] The interconnection diagram of each layer in the logical topology is as follows Figures 3 to 6 As shown, Figure 3 The diagram is a schematic diagram of the interconnection between the high-performance microprocessor and the CPU, where one CPU node is interconnected with eight high-performance microprocessors through a PCIE Switch; Figure 4 The diagram is a diagram of the interconnection between the CPU and the secondary management server. Each management server is connected to 8 CPUs through IB. Logically, each management server is also called a secondary management server. Figure 5 The diagram is a diagram of the interconnection between the secondary management server and the primary management server. Each management server with the primary management service manages 8 management servers with secondary management services through IB. The management server with the primary management service function also manages the secondary management service on its own node. Figure 6 It is a schematic diagram of the interconnection between the primary management server and the cluster management server. The cluster management server manages four management servers with primary management services through IB.

[0082] The PCIE service cluster adopts a 5-layer logical topology, which can reduce network congestion and facilitate management. For example, all 32 management servers are abstracted as secondary management servers. Among all secondary management servers, one management server is extracted from every 8 management servers to be a primary management server. The four primary management servers are responsible for managing not only the 8 CPUs connected to themselves, but also the 7 secondary management servers adjacent to themselves, and are responsible for interacting with the cluster management server. In this way, there are only 4 management servers interacting with the cluster management server, and a large amount of interaction data is dispersed to each node.

[0083] Thanks to this network architecture, the startup time of 2048 high-performance microprocessors is basically the same as the startup time of 8 high-performance microprocessors under one CPU, and the startup acceleration ratio is as high as 256.

[0084] At the same time, through the control of the first and second level management services, it is possible to control whether the PCIE device is started or not, truly realizing the controllability of the startup.

[0085] Based on the above PCIE service cluster, the implementation process of the PCIE service cluster startup method provided in this embodiment is shown in Figure 7 .

[0086] 701, the cluster management server is started.

[0087] 702. After the cluster management server is started, multiple management servers are started.

[0088] The management server startup process is as follows:

[0089] 702-1, after the cluster management server is started, the management server corresponding to the second top layer of the logical topology is started.

[0090] 702-2, after the management server corresponding to the second top layer of the logical topology is started, the management server corresponding to the next layer of the second top layer of the logical topology is started.

[0091] 703. After the management server is started, the CPU is started.

[0092] 704, after the CPU is started, the high-performance microprocessor is started.

[0093] The high-performance microprocessor startup process is as follows:

[0094] 704-1, a high performance microprocessor performs link training.

[0095] 704-2, if the training is unsuccessful, the high-performance microprocessor determines that the startup result is a startup failure. If the training is successful, the high-performance microprocessor downloads the firmware and then starts up to obtain a startup result.

[0096] The startup result is startup success or startup failure.

[0097] 704-3, the high performance microprocessor reports the startup result to the CPU.

[0098] 704-4, the CPU reports the startup result to the management server.

[0099] 704-5, the management server reports the startup result to the cluster management server.

[0100] 704-6, after the cluster management server determines that all high-performance microprocessors have been successfully started according to the startup result, it determines that the startup is normal and the computing state is ready. Otherwise, it determines that the startup is abnormal and reports the abnormality.

[0101] In traditional PCIE service clusters, due to too many PCIE devices, the following problems may occur:

[0102] 1. Slow startup speed.

[0103] 2. The startup sequence is chaotic and inconsistent. The device that starts earlier usually has to wait for a long time for the device that starts later, resulting in a lot of time waste.

[0104] 3. The inability to effectively monitor the startup and operating status of each PCIE device will result in high labor costs.

[0105] 4. Poor scalability.

[0106] 5. Network communication congestion.

[0107] The PCIE service cluster startup method provided in this embodiment effectively solves the above-mentioned problems of the traditional PCIE service cluster through the 4-layer physical topology and 5-layer logical topology of the PCIE service cluster, and in a PCIE service cluster of 2048 PCIE devices, it has the same startup speed as that of 8 PCIE devices interconnected, and the startup acceleration ratio is as high as 256.

[0108] The PCIE service cluster startup method provided in this embodiment can be used for the rapid startup of a high-performance PCIE service cluster composed of 2048 high-performance PCIE devices, and has the characteristics of fast startup speed, safety and reliability, and low maintenance cost.

[0109] Specifically, to address the problem of slow startup speed of traditional PCIE service clusters, the startup method of the PCIE service cluster provided in this embodiment adopts a startup mechanism of 8 PCIE devices as a group, so that the overall startup speed is basically the same as the startup speed of a group of 8 PCIE devices.

[0110] In order to solve the problem of chaotic and inconsistent startup order of traditional PCIE service clusters, the startup method of the PCIE service cluster provided in this embodiment adopts a primary and secondary management service composed of 32 management servers to manage the startup of each PCIE device respectively.

[0111] As the traditional PCIE service cluster cannot effectively monitor each PCIE device, the PCIE service cluster startup method provided in this embodiment adopts a 4-layer physical architecture plus a 5-layer logical architecture, which can effectively manage each PCIE device.

[0112] In view of the problem of poor scalability of the traditional PCIE service cluster, the PCIE service cluster started in the method for starting the PCIE service cluster provided in this embodiment has a tree topology and has good scalability.

[0113] In response to the problem of network communication congestion in traditional PCIE service clusters, the PCIE service cluster startup method provided in this embodiment adopts hierarchical management. The first-level node only needs to receive the information integrated by the next-level node. The information of 2048 PCIE devices will not be collected to the cluster management server at one time to cause network congestion.

[0114] In the startup sequence of the high-performance PCIE service cluster of the startup method of the PCIE service cluster provided in this embodiment, as shown in FIG. Figure 8 As shown, first the cluster management server starts to start; then the four management servers with first-level management services start to start; followed by the remaining 28 management servers with only second-level management services start to start; the eight CPUs under the 32 management servers start to start; after the CPU starts, the high-performance microprocessor under each CPU starts to start; the nodes at each level are started step by step, with the advantages of high controllability, availability, scalability, reliability, extensibility and stability.

[0115] The startup process of the high-performance microprocessor of the startup method of the PCIE service cluster provided in this embodiment is as follows: Fig. 9 As shown. The startup of a high-performance microprocessor starts after the CPU is ready and the firmware is ready. At this time, link training is first performed. If the training is unsuccessful, the startup failure is directly reported. After the training is successful, the firmware download begins. After the high-performance microprocessor receives the firmware, it starts various startup tasks. After startup, the startup result is reported to the CPU, whether it is successful or failed, so that the PCIE startup status of each high-performance microprocessor can be efficiently monitored.

[0116] The startup status reporting process of the high-performance PCIE service cluster of the startup method of the PCIE service cluster provided in this embodiment is as follows: Fig.10As shown. The CPU traverses the startup status of the high-performance microprocessor, the secondary management server traverses, the primary management server traverses, and the cluster management server traverses to determine whether all 2048 high-performance microprocessors are started normally. If the startup is normal, the computing status is confirmed to be ready, otherwise an exception is reported. The startup status reporting process of the high-performance PCIE service cluster is a process in which the success or failure of the startup is reported step by step through the nodes of each layer. Each layer only receives the report of the next layer. Each level receives the data results sorted by the nodes of the next level. The data will not be collected on the cluster management server side, so as not to cause data congestion.

[0117] This embodiment provides a method for starting a PCIE service cluster, wherein the PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors; the physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessors, CPUs, management servers, and cluster management servers from bottom to top; the method includes: starting the cluster management server; after the cluster management server is started, multiple management servers are started; after the management server is started, the CPU is started; after the CPU is started, the high-performance microprocessor is started. The method provided in this embodiment effectively solves the problems of slow startup of the existing PCIE service cluster, inability to effectively monitor the startup of each PCIE device, inconsistent startup sequence, and easy network congestion through a 4-layer physical topology.

[0118] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0119] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0122] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0123] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for starting a PCIE service cluster, characterized in that: The PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors; The physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessor, CPU, management server, and cluster management server from bottom to top; The method comprises: The cluster management server is started; After the cluster management server is started, multiple management servers are started; After the management server is started, the CPU starts; After the CPU is started, the high-performance microprocessor starts.

2. The method according to claim 1, characterized in that Every 8 high-performance microprocessors are connected to the same CPU via PCIE Switch; Every 8 CPUs are connected to the same management server through IB; All management servers are connected to the cluster management server through IB.

3. The method according to claim 1, characterized in that The logical topology of the PCIE service cluster is 5 layers; Among them, the bottom layer of the logical topology corresponds to the bottom layer of the physical topology; The second bottom layer of the logical topology corresponds to the second bottom layer of the physical topology; The top level of the logical topology corresponds to the top level of the physical topology; The second top layer of the logical topology and the next layer of the second top layer correspond to the second top layer of the physical topology.

4. The method according to claim 3, characterized in that The next layer below the second top layer of the logical topology corresponds to all management servers; The second top layer of the logical topology corresponds to one of each group of management servers, wherein each group of management servers includes eight management servers, and the management servers in each group of management servers are different from each other; Each management server in the second top layer of the logical topology is used to manage other management servers in its group and interact with the top layer.

5. The method according to claim 4, characterized in that After the cluster management server is started, multiple management servers are started, including: After the cluster management server is started, the management server corresponding to the second top layer of the logical topology is started; After the management server corresponding to the second top layer of the logical topology is started, the management server corresponding to the next layer of the second top layer of the logical topology is started.

6. The method according to claim 1, characterized in that The high performance microprocessor starts up, including: High-performance microprocessor performs link training; If the training is unsuccessful, the high-performance microprocessor determines the startup result as a startup failure; If the training is successful, the high-performance microprocessor downloads the firmware and starts up to obtain a startup result; wherein the startup result is a startup success or a startup failure; The high-performance microprocessor reports the startup results to the CPU; The CPU reports the startup result to the management server; The management server reports the startup result to the cluster management server; After the cluster management server determines that all high-performance microprocessors have been successfully started according to the startup results, it determines that the startup is normal and the computing state is ready; otherwise, it determines that the startup is abnormal and reports the abnormality.

7. A PCIE service cluster, characterized in that: The PCIE service cluster includes a cluster management server, multiple management servers, multiple CPUs, and multiple high-performance microprocessors; The physical topology of the PCIE service cluster is 4 layers, which are high-performance microprocessor, CPU, management server, and cluster management server from bottom to top; The PCIE service cluster is started by the method according to any one of claims 1 to 6.

8. The PCIE service cluster according to claim 7, characterized in that: Every 8 high-performance microprocessors are interconnected with the same CPU via PCIESwitch; Every 8 CPUs are connected to the same management server through IB; All management servers are connected to the cluster management server through IB.

9. The PCIE service cluster according to claim 7, characterized in that: The logical topology of the PCIE service cluster is 5 layers; Among them, the bottom layer of the logical topology corresponds to the bottom layer of the physical topology; The second bottom layer of the logical topology corresponds to the second bottom layer of the physical topology; The top level of the logical topology corresponds to the top level of the physical topology; The second top layer of the logical topology and the next layer of the second top layer correspond to the second top layer of the physical topology.

10. The PCIE service cluster according to claim 9, characterized in that: The next layer below the second top layer of the logical topology corresponds to all management servers; The second top layer of the logical topology corresponds to one of each group of management servers, wherein each group of management servers includes eight management servers, and the management servers in each group of management servers are different from each other; Each management server in the second top layer of the logical topology is used to manage other management servers in its group and interact with the top layer.

Images