Hyper-converged server system suitable for edge and end side data centers

Abstract

The invention discloses a hyper-converged server system suitable for edge and end side data centers, and relates to the field of cloud computing. A calculation module; the computing module is arranged at the front part of the case and comprises a plurality of CPU (Central Processing Unit) modules, and the CPU modules are used for providing computing power required by general services; a GPU module; the GPU module is arranged at the rear part of the case, the GPU module comprises a plurality of GPU modules, and the GPU modules are used for providing computing power required by services of cloud rendering, cloud games, AI reasoning and small model training; a fan module; a network switching module; an out-of-band management module; a redundant power source; and a middle plate. By implementing the method, a plurality of computing modules are fused in one server architecture, so that hardware construction is more flexible.

Description

Technical Field

[0001] The present application relates to the field of cloud computing, and in particular to a hyper-converged server system suitable for edge and end-side data centers. Background Art

[0002] With the development of edge computing and edge computing, diverse demands have emerged, such as cloud rendering, cloud gaming, cloud desktops, and edge AI applications. These application scenarios require the corresponding data centers to have high performance and low latency.

[0003] Existing data center solutions use an architecture that separates general-purpose servers, general-purpose storage, and network switches, which are combined in cabinets through different product forms. There are also many large and financially strong manufacturers at home and abroad that use workstations to meet the needs of end-side related business application scenarios.

[0004] However, for tasks that require an entire edge / end cabinet or workstation, such as building or rendering cloud games, it is usually necessary to purchase and configure a variety of high-performance hardware devices. The hardware configuration is relatively fixed and it is difficult to meet different consumer needs. Summary of the invention

[0005] The present application provides a hyper-converged server system suitable for edge and end-side data centers, which is used to integrate multiple computing modules into one server architecture, making hardware construction more flexible.

[0006] In a first aspect, the present application provides a hyper-converged server system suitable for edge and end-side data centers, characterized in that the hyper-converged server system comprises: a chassis; the chassis is the outer shell of the hyper-converged server system, used to accommodate various components; a computing module; the computing module is arranged at the front of the chassis, the computing module includes multiple CPU modules, and the CPU module is used to provide the computing power required for general services; a GPU module; the GPU module is arranged at the rear of the chassis, the GPU module includes multiple GPU modules, and the GPU module is used to provide the computing power required for cloud rendering, cloud gaming, AI reasoning and small model training services; a fan module; the fan module is arranged in the middle of the chassis, the fan module includes multiple cooling fans, and the cooling fans are used to form an air duct from the front of the chassis to the rear of the chassis; a network switching module ; The network switching module is arranged at the lower rear layer of the chassis, and is used to provide network signal exchange between the computing module and external devices; out-of-band management module; The out-of-band management module is arranged at the lower rear layer of the chassis, and is used to send control signals to the computing module, the GPU module and the fan module to realize control management of the computing module, the GPU module and the fan module; redundant power supply; The redundant power supply is arranged at the lower rear layer of the chassis, and is used to provide operating current to various components of the hyper-converged server system; midplane; The midplane is arranged in the middle of the chassis and is located below the fan module, and is respectively connected to the computing module, the GPU module, the fan module, the network switching module, the out-of-band management module and the redundant power supply; The midplane is used to provide transmission of the network signal, the control signal and the operating current.

[0007] In the above embodiment, the hyper-converged server system forms a highly integrated hyper-converged server system architecture by setting the computing module at the front of the chassis, the GPU module at the rear, the fan module in the middle, and the network switching module, out-of-band management module and redundant power supply at the lower layer of the rear, and connecting each module through the middle plate. This layout design can optimize the airflow channel inside the server and improve the heat dissipation efficiency; at the same time, the network signal, control signal and power supply are uniformly managed through the middle plate, simplifying the connection complexity within the system. This architecture allows different computing resources (CPU and GPU) to work together to meet the high performance and low latency requirements of edge computing and end-side computing, while ensuring the reliability and maintainability of the system.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the hyper-converged server system also includes a storage module; the storage module can be used to replace the GPU module, or replace part of the GPU module of the GPU module; the storage module is arranged at the rear of the chassis, and includes multiple storage media for providing data storage.

[0009] In the above embodiment, the hyper-converged server system has higher flexibility and scalability by designing replaceable storage modules. The storage module can completely replace the GPU module or part of the GPU module, allowing users to flexibly adjust the ratio of the system's computing power and storage capacity according to actual business needs. This modular design can not only optimize resource utilization, but also reduce the total cost of ownership of the system. When business needs shift from computing-intensive to storage-intensive, there is no need to re-purchase a complete server system, only the corresponding module needs to be replaced, which greatly improves the adaptability and economic benefits of the system.

[0010] In combination with some embodiments of the first aspect, in some embodiments, the computing module, the GPU module and the storage module all support hot-plug operations.

[0011] In the above embodiment, the hyper-converged server system improves the maintainability and availability of the system by supporting hot-swap operations of computing modules, GPU modules, and storage modules. The hot-swap function allows the replacement and maintenance of modules while the system is running, avoiding complete system downtime due to maintenance and reducing the impact of maintenance on the business. This design is particularly suitable for critical business scenarios that require continuous operation, and can achieve dynamic expansion of the system and rapid recovery from faults, thereby improving the overall reliability and service quality of the system.

[0012] In combination with some embodiments of the first aspect, in some embodiments, the network switching module includes multiple network interfaces, one of the CPU modules of the computing module is communicatively connected to the multiple network interfaces, and the network switching module can receive the network control signal of the out-of-band management module to realize network switching of the CPU module on the multiple network interfaces.

[0013] In the above embodiment, the hyper-converged server system realizes a flexible network management mechanism by setting multiple network interfaces in the network switching module and allowing the CPU module to communicate with these interfaces, while supporting network switching control through the out-of-band management module. This design not only provides network redundancy protection and ensures the network reliability of the system, but also supports dynamic adjustment of the network connection mode according to business needs. Through the control of the out-of-band management module, the system can realize intelligent network traffic scheduling and load balancing, optimize network resource utilization, and improve overall network performance.

[0014] In combination with some embodiments of the first aspect, in some embodiments, multiple video signal channels are set on the middle board, which are respectively used to receive video signals transmitted by the computing module and the GPU module; the computing module and the GPU module interact with each other through the middle board via high-speed signals.

[0015] In the above embodiment, the hyper-converged server system realizes high-speed signal interaction between the computing module and the GPU module by setting up multiple video signal channels on the midboard. This design provides data channels for high-bandwidth applications such as video processing and graphics rendering, reduces signal transmission delay, and improves data transmission efficiency. The dedicated video signal channel can not only ensure the real-time transmission of video data, but also avoid mutual interference with other data transmissions. It is suitable for application scenarios with high requirements for video processing performance, such as cloud games and visual computing.

[0016] In combination with some embodiments of the first aspect, in some embodiments, a CPU module of the computing module can simultaneously exchange information with one or more GPU modules of the GPU module through the midboard.

[0017] In the above embodiment, the hyper-converged server system realizes more flexible computing resource scheduling and more efficient parallel processing capabilities by supporting a single CPU module to interact with multiple GPU modules at the same time. This design enables the system to dynamically allocate GPU resources according to task characteristics, optimize computing load distribution, and improve the overall processing efficiency of the system. Especially in scenarios such as deep learning training and large-scale parallel computing, the advantages of multi-GPU collaborative computing can be fully utilized to improve computing performance and resource utilization.

[0018] In combination with some embodiments of the first aspect, in some embodiments, the out-of-band management module receives a data signal stream from the computing module through the midplane, and the out-of-band management module receives control management signals respectively from the computing module, the GPU module, the fan module, the network switching module and the redundant power supply through the midplane, and sends the data signal stream and the control management signal to the information monitoring terminal.

[0019] In the above embodiment, the hyper-converged server system receives and processes the data signal stream and control management signal from each functional module through the out-of-band management module, and transmits this information to the information monitoring terminal, thereby realizing comprehensive system monitoring and management functions. This centralized management mechanism enables system administrators to grasp the operating status of each module in real time, quickly discover and handle potential problems, and improve the manageability and maintenance efficiency of the system.

[0020] In the second aspect, an embodiment of the present application provides a data processing method suitable for edge and end-side data centers, characterized in that it is applied to a hyper-converged server system, and the method includes: receiving a computing task issued by a user terminal, and determining a task characteristic corresponding to the computing task; according to the task characteristic, determining multiple target computing platforms that match the computing task in all computing platform modules on the communication link; determining the resource configuration of the multiple target computing platforms, and generating corresponding computing subtasks based on the resource configuration list and the task characteristic of the computing task; issuing the computing subtask to the corresponding target computing platform; monitoring the operating data of the multiple target computing platforms, and generating an execution status report.

[0021] In the above embodiment, the hyper-converged server system realizes an efficient edge computing task processing mechanism by establishing a complete task processing process, including task feature identification, resource matching, task decomposition, execution monitoring, etc. It can accurately identify task requirements, optimize resource allocation, realize efficient execution of tasks, and improve overall computing efficiency and service quality.

[0022] In combination with some embodiments of the second aspect, in some embodiments, after the step of monitoring the operating data of the multiple target computing platforms and generating an execution status report, the method also includes: when a new computing platform is detected to be connected, obtaining the hardware configuration information and performance indicators of the new computing platform, and generating a resource feature vector; determining the priority score for task transfer of each target computing platform based on the execution status report and the resource feature vector; based on the priority score, screening the computing platforms to be transferred and the corresponding tasks to be transferred from the target computing platforms; and generating a task transfer list based on the task dependencies and resource requirements of the tasks to be transferred.

[0023] In the above embodiment, the hyper-converged server system implements a dynamic task migration mechanism. When a new computing platform is detected, the system can intelligently evaluate the migration possibility of existing tasks. By calculating priority scores, analyzing task dependencies and resource requirements, the system can formulate the optimal task migration plan. This dynamic adjustment capability can not only optimize resource utilization, but also achieve load balancing, improve the scalability and adaptability of the system, and is particularly suitable for dynamically changing business needs in edge computing environments.

[0024] In combination with some embodiments of the second aspect, in some embodiments, the step of monitoring the operating data of the multiple target computing platforms and generating an execution status report specifically includes: collecting the original monitoring data streams including CPU occupancy, memory usage, storage capacity, network bandwidth and energy consumption data of the multiple target computing platforms; performing standardization processing and time series storage on the original monitoring data streams to generate a standard performance indicator data set containing timestamps; based on the standard performance indicator data set, generating an execution status report including performance scores, anomaly detection results and optimization suggestions.

[0025] In the above embodiment, the hyper-converged server system establishes a comprehensive platform monitoring mechanism, and the system can collect multi-dimensional monitoring data including CPU occupancy, memory usage, storage capacity, network bandwidth and energy consumption data. By standardizing and analyzing these data, an execution status report containing performance scores, anomaly detection results and optimization suggestions is generated, which helps to improve system operation efficiency and reduce operation and maintenance costs.

[0026] It can be understood that the hyper-converged server system provided in the first aspect is used to execute the method provided in the second aspect of the present application. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be repeated here.

[0027] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Due to the modular layout design inside the chassis and the unified connection management mechanism of the middle board, the optimal layout and efficient interconnection of core functional modules such as computing modules, GPU modules, and fan modules are achieved. The layout of the computing module at the front of the chassis and the GPU module at the rear, combined with the fan module in the middle, forms a heat dissipation airflow channel that runs from front to back; at the same time, the network signals, control signals, and power supply between modules are uniformly managed through the middle board, simplifying the internal connections of the system. It effectively solves the problems of low heat dissipation efficiency of modular servers in the existing technology, thereby improving the system's heat dissipation efficiency and system maintainability, enabling the system to maintain stable and efficient operation in an edge computing environment.

[0028] 2. Due to the adoption of a resource matching mechanism based on task feature vectors and a multi-level task processing flow, a complete processing chain from task feature identification, resource matching, task decomposition to execution monitoring is established. This mechanism can accurately analyze the resource requirements and time constraints of computing tasks, and accordingly find the most matching computing platform on the communication link, and then generate fine-grained computing subtasks and distribute them for execution. This effectively solves the problem of unreasonable computing resource allocation in the prior art, thereby achieving accurate matching and distribution of computing tasks and efficient use of computing resources, and improving the task processing efficiency in the edge computing environment.

[0029] 3. Due to the use of a dynamic task migration mechanism based on priority scoring and a resource feature vector analysis method, the migration possibility of existing tasks can be intelligently evaluated when a new computing platform is connected. The system calculates the priority score of task migration by analyzing the execution status report and the resource characteristics of the new platform, and formulates the optimal migration plan based on the task dependency. This effectively solves the problems of poor system scalability and low resource utilization in the existing technology, and then realizes the dynamic optimization allocation of computing tasks and the elastic expansion of the computing platform, which greatly improves the adaptability and resource utilization efficiency of the edge computing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a structural diagram of a hyper-converged server system in an embodiment of the present application; Figure 2 is another structural diagram of the hyper-converged server system in an embodiment of the present application; Figure 3 It is a schematic diagram of a virtual device of a hyper-converged server system in an embodiment of the present application; Figure 4 This is a network topology diagram of a hyper-converged server system in an embodiment of the present application; Figure 5 It is a computing topology diagram of a hyper-converged server system in an embodiment of the present application; Figure 6 It is a management topology diagram of a hyper-converged server system in an embodiment of the present application; Figure 7 This is a data collection schematic diagram of a hyper-converged server system in an embodiment of the present application; Figure 8 It is a flow chart of a data processing method suitable for edge and end-side data centers in an embodiment of the present application.

[0031] Description of reference numerals: 101. Chassis; 102. Computing module; 103. GPU module; 104. Fan module; 105. Network switching module; 106. Out-of-band management module; 107. Redundant power supply; 108. Midplane; 201. Storage module. DETAILED DESCRIPTION

[0032] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification of the present application, the singular expressions "one", "a kind of", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more of the listed items.

[0033] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0034] The following describes scenarios in which the hyper-converged server system in this application is suitable for edge and end-side data centers.

[0035] With the rapid development of 5G networks, the Internet of Things, and smart cities, the demand for edge computing has shown explosive growth. In a smart park, it is necessary to deploy edge servers in the park for AI applications such as video surveillance analysis, face recognition, and vehicle management. The server is required to have powerful GPU computing power and low latency characteristics. In a cloud gaming data center, it is necessary to deploy a game rendering server at an edge location close to the user, which requires powerful GPU rendering capabilities and high-speed network transmission capabilities. Existing solutions often require the deployment of computing servers, storage devices, and network devices separately, resulting in a wide variety of equipment, complex management, and high costs. To this end, the present application proposes a new type of hyper-converged server architecture that highly integrates computing, storage, network and other resources in a chassis 101 to better meet the needs of edge computing scenarios.

[0036] See also Figure 1, is a structural schematic diagram of a hyper-converged server system in an embodiment of the present application, including a front view (upper left), a rear view (upper right), a side view (lower left) and a top view (lower right) of the hyper-converged server system. As shown in the figure, the hyper-converged server system in the present application includes: a chassis 101; the chassis 101 is a shell of the hyper-converged server system, used to accommodate various components; a computing module 102; the computing module 102 is arranged at the front of the chassis 101, and the computing module 102 includes multiple CPU modules, the CP The U module is used to provide the computing power required for general services; GPU module 103; the GPU module 103 is arranged at the rear of the chassis 101, and the GPU module 103 includes multiple GPU modules, and the GPU module is used to provide the computing power required for cloud rendering, cloud gaming, AI reasoning and small model training services; fan module 104; the fan module 104 is arranged in the middle of the chassis 101, and the fan module 104 includes multiple cooling fans, and the cooling fans are used to form an air duct from the front of the chassis 101 to the rear of the chassis 101; network The switching module 105 is provided at the lower rear layer of the chassis 101, and is used to provide the network signal exchange between the computing module 102 and the external device; the out-of-band management module 106 is provided at the lower rear layer of the chassis 101, and is used to send control signals to the computing module 102, the GPU module 103 and the fan module 104, so as to realize the control management of the computing module 102, the GPU module 103 and the fan module 104; the redundant power supply 107; the redundant power supply 108 is provided at the lower rear layer of the chassis 101, and is used to send control signals to the computing module 102, the GPU module 103 and the fan module 104, so as to realize the control management of the computing module 102, the GPU module 103 and the fan module 104; the redundant power supply 108 is provided at the lower rear layer of the chassis 101, and is used to send control signals to the computing module 102, the GPU module 103 and the fan module 104, and is used to control ... The power source 107 is arranged at the lower rear layer of the chassis 101, and is used to provide operating current to various components of the hyper-converged server system; the midplane 108; the midplane 108 is arranged in the middle of the chassis 101 and is located below the fan module 104, and is respectively connected to the computing module 102, the GPU module 103, the fan module 104, the network switching module 105, the out-of-band management module 106 and the redundant power supply 107; the midplane 108 is used to provide transmission of the network signal, the control signal and the operating current.

[0037] Specifically, the hyper-converged server system adopts the U standard chassis 101 design (expandable to U, U and other specifications), with an overall size of mm (width) × mm (height) × mm (length). The chassis 101 adopts a modular design concept, with a computing module 102 installed in the front, a fan module 104 and a mid-board 108 in the middle, and a GPU module 103, a network switching module 105, an out-of-band management module 106 and a power module installed in the rear. The chassis 101 is made of high-quality steel and has dust-proof, anti-static, anti-interference and other characteristics, and is equipped with a top cover and side panels that can be quickly disassembled for easy daily maintenance.

[0038] The computing module 102 of the hyper-converged server system adopts a modular tray design, which supports the installation of an independent CPU computing unit at the same time. Each CPU computing unit can choose Intel Core series or AMD Ryzen series processors, support DDR, DDR memory, and a single computing unit supports a maximum memory capacity of GB. The computing module 102 adopts a pull-out design, equipped with a rail system, and supports hot-swap function. Each computing unit is equipped with an independent heat dissipation system and temperature monitoring sensor, and exchanges data and manages signal transmission with other modules of the system through the middle board 108.

[0039] The GPU module 103 of the hyper-converged server system is located at the rear of the chassis 101 and supports the installation of a standard GPU card. It can be equipped with NVIDIA RTX series or AMD Radeon Pro series graphics cards and connected to the system through the PCIe . x interface. Each GPU module 103 is equipped with an independent radiator and heat-conducting copper pipe to ensure that the temperature is maintained at a stable level under high-load working conditions. The GPU module 103 supports hot-swappable design and can be flexibly replaced or upgraded according to business needs, and realizes high-speed data interaction with the CPU module through the midplane 108.

[0040] The fan module 104 of the hyper-converged server system includes redundant fans of different specifications, which can be intelligently adjusted in speed through PWM signals. The fan module 104 is designed as an integral pull-out structure, which is convenient for maintenance and replacement. Each fan adopts a double ball bearing design and has an anti-reverse connection protection function. The temperature sensor data is received through the middle board 108 to achieve dynamic adjustment of the fan speed and ensure the heat dissipation efficiency of the system. The failure of a single fan does not affect the overall heat dissipation effect, and online replacement is supported.

[0041] The network switch module 105 of the hyper-converged server system adopts a rear-mounted design and provides 10 GbE SFP uplink ports and 10 internal GbE switch ports. It supports network features such as VLAN, QoS, ACL, and has LACP link aggregation function. The module has a built-in network processing chip, supports non-blocking line-speed forwarding, and can realize high-speed data exchange between components inside the system. Remote management and monitoring of network configuration can be achieved through the out-of-band management module 106.

[0042] The out-of-band management module 106 of the hyper-converged server system adopts an independent management processor design and provides a dedicated Gbps management network port. It supports the IPMI protocol and provides a web-based graphical management interface. It has a remote KVM function and can realize remote power on / off, system restart, hardware monitoring and other operations. The module collects the operating status data of each functional module through the middle board 108 and provides real-time alarm and log recording functions. It supports the SNMP protocol and can be integrated with a third-party monitoring platform.

[0043] The PCB design of the middle board 108 of the hyper-converged server system integrates a high-speed PCIe switching chip, providing up to 100 PCIe channels. As the core interconnection component of the system, the middle board 108 provides power distribution, signal transmission and management control functions for each functional module. The onboard BMC management chip is responsible for the underlying hardware monitoring of the system and communicates with the out-of-band management module 106 through an independent management bus. There are multiple high-speed connectors on the middle board 108 for docking of each functional module and providing redundant power management circuits.

[0044] The hyper-converged server system adopts a + redundant power supply 107 design. Each power supply module supports a rated power of W and supports -VAC input. The power supply module has multiple protection mechanisms such as overcurrent protection, overvoltage protection, and overtemperature protection. It supports hot-swap replacement and realizes power status monitoring and alarm functions through the out-of-band management module 106. The system can realize the coordinated work of multiple power supply modules through load balancing technology.

[0045] See also Figure 2 , is another structural schematic diagram of the hyper-converged server system in an embodiment of the present application, including a front view (upper left), a rear view (upper right), a side view (lower left) and a top view (lower right) of the hyper-converged server system, and the hyper-converged server system also includes a storage module 201; the storage module 201 can be used to replace the GPU module 103, or replace part of the GPU modules of the GPU module 103; the storage module 201 is arranged at the rear of the chassis 101, and includes multiple storage media for providing data storage.

[0046] Specifically, the storage module 201 of the hyper-converged server system adopts a standard .-inch hard disk tray design, which can install up to multiple HDD mechanical hard disks, and is also compatible with .-inch mechanical or solid-state hard disks. The storage module 201 adopts a U.4 interface design, supports the PCIe .4 transmission protocol, and the maximum transmission rate of a single disk reaches GB / s. The module is equipped with an independent storage controller and supports multiple array modes such as RAID / / / / . Storage resources can be flexibly allocated to different computing units according to business needs, and storage pool management is supported. When the system has a low demand for GPU computing power, some or all of the GPU module positions can be replaced with storage module 201 to achieve dynamic expansion of storage capacity.

[0047] Based on the above, please refer to Figure 3 , is a schematic diagram of a virtual device of a hyper-converged server system in an embodiment of the present application.

[0048] The hyper-converged server system adopts a virtualized infrastructure design to achieve flexible allocation and unified management of hardware resources. The system uses the physical computing module 102, GPU module 103 and storage module 201 as resource pools, which can be dynamically allocated according to business needs. At the computing level, the system supports virtualizing a single physical CPU module into multiple virtual CPU instances, or combining multiple physical CPU modules into a high-performance computing cluster. Each virtual CPU instance can independently allocate memory resources and network bandwidth, and supports real-time migration and load balancing. At the GPU level, the system adopts GPU virtualization technology to support the division of a single physical GPU into multiple virtual GPU instances to meet the computing power requirements of different application scenarios. It supports GPU pass-through mode to directly allocate specific GPUs to virtual machines for use; it also supports GPU sharing mode to enable multiple virtual machines to share the same GPU resources. At the storage level, the system manages physical storage devices in a pooled manner, supports the creation of storage volumes with different performance levels, and can dynamically adjust storage capacity and performance configuration according to application requirements. It supports online expansion and storage migration to ensure the continuity of data services. The system transmits various types of signals, such as control signals, high-speed signals, network signals, power supply, etc., based on the middleboard 108 through a unified virtualization management platform (out-of-band management module 106), to achieve unified resource scheduling and monitoring. The management platform provides functions such as resource usage analysis, performance monitoring, and capacity planning, and supports policy-based automated resource allocation and load balancing. At the same time, the system supports multi-tenant isolation, and the resources and data between different tenants are completely isolated to ensure business security.

[0049] At the same time, the computing module 102, the GPU module 103 and the storage module 201 in the hyper-converged server system all support hot-swap operations.

[0050] Specifically, the computing module 102, GPU module 103 and storage module 201 of the hyper-converged server system all adopt a standardized tray design, equipped with a guide rail system and a locking mechanism. The power and signal connectors of each module adopt a hot-swappable level design, with a first-in-last-out feature to ensure the connection security during live plugging and unplugging. An LED status indicator is provided at the front end of the module tray to display the current operating status and whether it can be safely unplugged. The system monitors the status of each module in real time through the out-of-band management module 106, and supports online replacement of modules without shutting down the power of the entire machine.

[0051] See also Figure 4 , is a network topology diagram of a hyper-converged server system in an embodiment of the present application. The network switching module 105 in the hyper-converged server system includes multiple network interfaces, one of the CPU modules of the computing module 102 is communicatively connected with the multiple network interfaces, and the network switching module 105 can receive the network control signal of the out-of-band management module 106 to implement network switching of the CPU module on the multiple network interfaces.

[0052] Specifically, the network switching module 105 of the hyper-converged server system adopts a dual-switch redundant design, and each switching module can support network redundancy function and also support Boning function (double network bandwidth). The network architecture provides *Gbps (total Gbps) external data path. The CPU module can connect to two switching modules at the same time through a dual-port network card, supporting network redundancy protection or link aggregation mode. The switching module has a built-in network processing chip, which supports VLAN-based traffic isolation, QoS service quality assurance and ACL access control. The out-of-band management module 106 can realize dynamic switching and load balancing of the CPU module between multiple network ports by sending network control instructions.

[0053] See also Figure 5 , is a computing topology diagram of a hyper-converged server system in an embodiment of the present application. A plurality of video signal channels are arranged on the middle board 108 in the hyper-converged server system, which are respectively used to receive the video signals transmitted by the computing module 102 and the GPU module 103; the computing module 102 and the GPU module 103 perform high-speed signal interaction through the middle board 108.

[0054] Specifically, the middle board 108 of the hyper-converged server system provides multiple independent PCIe .x channels for high-speed data transmission between the CPU module and the GPU module 103. The middle board 108 is designed with multiple video signal adapters to support the transmission and conversion of DisplayPort . and HDMI . signals. Through the crossbar switch matrix of the middle board 108, the video signal connection between any CPU module and the GPU module 103 can be realized, supporting functions such as GPU pass-through and GPU virtualization.

[0055] A CPU module of the computing module 102 in the hyper-converged server system can simultaneously exchange information with one or more GPU modules of the GPU module 103 through the midplane 108 .

[0056] Specifically, the hyper-converged server system realizes flexible interconnection between the CPU module and the GPU module 103 through the midplane 108. A single CPU module can establish PCIe .x connections with a maximum of GPU modules at the same time, with a total bandwidth of GB / s. The system supports GPU pass-through mode, which directly allocates a specific GPU to a specified CPU for use; it also supports GPU sharing mode, which allows multiple CPU modules to share the same GPU resources. The connection relationship between the CPU and the GPU can be dynamically adjusted through the out-of-band management module 106 to achieve flexible allocation of resources.

[0057] See also Figure 6, is a management topology diagram of a hyper-converged server system in an embodiment of the present application. The out-of-band management module 106 in the hyper-converged server system receives a data signal stream from the computing module 102 through the midboard 108, and the out-of-band management module 106 receives control management signals from the computing module 102, the GPU module 103, the fan module 104, the network switching module 105, and the redundant power supply 107 through the midboard 108, and sends the data signal stream and the control management signal to the information monitoring terminal.

[0058] Specifically, the out-of-band management module 106 of the hyper-converged server system adopts a dual-core ARM processor design and is equipped with independent system memory and storage space. Through the management bus on the midboard 108, the operating parameters of each functional module are collected in real time, including temperature, voltage, power consumption, fan speed and other data. The management module provides a Web-based graphical management interface and supports functions such as remote KVM, virtual media mounting, and system log recording. The system status information can be pushed to an external monitoring platform, supporting the SNMP protocol and RESTful API interface, which is convenient for integration with third-party management systems.

[0059] Based on the above, please refer to Figure 7 , which is a data collection schematic diagram of a hyper-converged server system in an embodiment of the present application.

[0060] like Figure 7 As shown, the hyper-converged server cluster includes multiple hyper-converged server systems. The hyper-converged server system adopts a distributed data collection architecture and deploys data collection agents in each functional module 103. The collected data includes indicators such as CPU utilization, memory usage, storage capacity, energy consumption load, network bandwidth, etc. (in Figure 7 The system supports setting multi-level alarm thresholds. When the monitoring indicators exceed the preset range, the administrator can be notified by email, SMS, etc. The management platform can Figure 7 The Dashboard shown displays the overall resource usage, host resource usage ranking, and the operating status of each server host, and can also provide rich data visualization functions, including resource usage trend charts, performance analysis reports, capacity planning recommendations, etc.

[0061] The following is a description of the process of the method provided in this application. Figure 8 , which is a flow chart of a data processing method suitable for edge and end-side data centers in an embodiment of the present application.

[0062] S801: Receive a computing task sent by a user terminal, and determine a task feature corresponding to the computing task.

[0063] Among them, the user terminal refers to the client device that initiates the computing task request; the computing task refers to the specific business load that needs to be processed; the task characteristics represent the key attributes that describe the computing task, including but not limited to characteristic parameters such as computing complexity, resource requirements, latency requirements, and data scale.

[0064] This step usually occurs when a user submits a new computing task to the hyper-converged system through a terminal. Specifically, the hyper-converged server system first receives the computing task request from the user terminal, then extracts features of the computing task through the task analysis module, identifies and determines the key attribute parameters of the task, and provides a decision basis for subsequent task allocation and resource scheduling.

[0065] In some embodiments, the reception and feature determination of computing tasks can be implemented in a variety of ways: optionally, receiving task requests through a RESTful API interface, parsing the task description file in the request, using a feature extraction algorithm to analyze task attributes, and generating a standardized task feature description; optionally, receiving tasks through a message queue, preprocessing and formatting tasks, using a machine learning model for feature recognition, and outputting structured feature data. It is understandable that other protocols and processing methods can also be used to implement task reception and feature extraction, which are not limited here.

[0066] S802: According to the task characteristics, determine multiple target computing platforms that match the computing task among all computing platform modules on the communication link.

[0067] Among them, the communication link refers to the network channel connecting each computing platform; the computing platform module refers to the hardware facility unit with computing processing capabilities, including the CPU module in the computing module and the GPU module in the GPU module; the target computing platform refers to the computing processing unit that meets the task characteristic requirements.

[0068] This step is performed after the task characteristics are determined. Specifically, the system traverses all computing platforms on the communication link according to the determined task characteristics, compares the matching degree between the processing capabilities of each computing platform and the task requirements through a matching algorithm, and finally selects multiple target computing platforms suitable for executing the computing task.

[0069] In some embodiments, the target computing platform can be determined in a variety of ways: optionally, a computing platform capability index library is constructed, platform screening rules are designed, matching degree calculation is performed, and the target platform is selected according to matching degree sorting; optionally, a dynamic resource pool is established, platform status data is collected in real time, an intelligent scheduling algorithm is executed, and the optimal platform combination is output. It is understandable that other matching algorithms and selection strategies can also be used to determine the target platform, which is not limited here.

[0070] S803: Determine the resource configuration of the multiple target computing platforms, and generate corresponding computing subtasks based on the resource configuration list and the task characteristics of the computing task.

[0071] Among them, resource configuration refers to the specific configuration information of computing, storage, network and other resources of the target computing platform; computing subtask refers to the smaller-granularity task unit obtained by decomposing the original computing task according to resource characteristics; configuration list refers to a detailed list recording resource configuration.

[0072] This step is performed after the target computing platform is determined. Specifically, the system first obtains the detailed resource configuration information of each target computing platform, including the number of CPU cores, memory capacity, storage space, etc., and then, based on this resource configuration information and task characteristics, reasonably splits the original computing task into multiple computing subtasks with appropriate granularity to ensure that each subtask matches the resource capabilities of the target platform.

[0073] In some embodiments, resource configuration determination and task splitting can be achieved in a variety of ways: optionally, collecting platform resource information, generating a standardized resource list, executing a task decomposition algorithm, and establishing a task dependency graph; optionally, establishing a resource portrait model, calculating the optimal task allocation plan, generating a subtask description file, and formulating a task scheduling strategy. It is understandable that other resource management and task decomposition methods can also be used to implement this process, which is not limited here.

[0074] S804: Send the computing subtask to the corresponding target computing platform.

[0075] This step is performed after the subtask generation is completed. Specifically, the system accurately transmits each computing subtask to the specified target computing platform through a reliable task distribution mechanism based on the previously established correspondence between the subtask and the target computing platform, and ensures the integrity and reliability of the task distribution process.

[0076] In some embodiments, the subtask can be issued in a variety of ways: optionally, establish a secure communication channel, package subtask data, perform task transmission, and verify transmission integrity; optionally, configure task queues, implement load balancing, perform parallel distribution, and confirm task reception status. It is understandable that other task distribution mechanisms can also be used to implement the subtask issuance process, which is not limited here.

[0077] S805: Monitor the operating data of the multiple target computing platforms and generate an execution status report.

[0078] Among them, the operation data represents the performance indicators and status information of the target computing platform when executing subtasks; the execution status report refers to a comprehensive report document that describes the progress and results of task execution.

[0079] This step continues after the subtask starts executing. Specifically, the system collects the operating status data of each target computing platform in real time through the monitoring module, including resource utilization, execution progress, error logs and other information, analyzes and integrates these data, and generates a detailed execution status report to provide a basis for task management and exception handling.

[0080] In some embodiments, the hyper-converged server system collects the original monitoring data streams of the multiple target computing platforms, including CPU occupancy, memory usage, storage capacity, network bandwidth, and energy consumption data; performs standardization processing and time series storage on the original monitoring data streams to generate a standard performance indicator data set containing timestamps; based on the standard performance indicator data set, generates an execution status report containing performance scores, anomaly detection results, and optimization suggestions.

[0081] Among them, the original monitoring data stream represents the unprocessed performance monitoring data collected directly from the target computing platform; the standard performance indicator data set refers to a set of performance data that has been normalized and has a unified format; the execution status report represents a comprehensive evaluation document generated based on performance data analysis; the CPU occupancy rate represents the degree of processor resource utilization; the memory utilization rate refers to the occupancy level of memory resources; the storage capacity is used to represent the usage of data storage space; the network bandwidth represents the occupancy status of data transmission capacity; and the energy consumption data refers to the energy consumption indicators during the operation of the platform.

[0082] Specifically, the system first continuously collects multiple raw performance data through monitoring components, then standardizes these data to make them have a unified format and metric, stores the processed data in chronological order to form a standardized data set, and finally uses the data analysis engine to conduct in-depth analysis of the standardized data to generate a status report containing multi-dimensional evaluation results.

[0083] In some embodiments, the collection and analysis of performance data can be achieved in a variety of ways: optionally, deploying distributed data collection agents, establishing real-time data transmission channels, performing data cleaning and standardization, and applying machine learning models for analysis; optionally, configuring performance monitoring probes, implementing data aggregation processing, establishing time series database storage, and generating visual reports. It is understandable that other monitoring and analysis methods can also be used to achieve platform performance evaluation, which is not limited here.

[0084] In some embodiments, when a new computing platform is detected to be connected, the hyper-converged server system obtains the hardware configuration information and performance indicators of the new computing platform and generates a resource feature vector; determines the priority score for task transfer to each target computing platform based on the execution status report and the resource feature vector; based on the priority score, screens the computing platforms to be transferred and the corresponding tasks to be transferred from the target computing platforms; and generates a task transfer list based on the task dependencies and resource requirements of the tasks to be transferred.

[0085] Among them, the resource feature vector represents the numerical representation of the resource characteristics of the computing platform; the priority score refers to a quantitative indicator to measure the urgency of task transfer; the tasks to be transferred represent the computing tasks that need to be migrated; the task dependency refers to the logical relationship between computing tasks; and the task transfer list refers to a document that records the task migration plan in detail.

[0086] The operations described in this paragraph are performed when the system detects that a new computing platform has been added or that tasks need to be rescheduled. Specifically, the system first obtains the detailed configuration information of the newly added platform and converts it into a feature vector, calculates the priority of task transfer based on the execution status report of the existing platform, and then selects the platforms and tasks that need to be transferred. Finally, a detailed transfer plan is formulated considering the dependencies between tasks.

[0087] In some embodiments, the formulation of the task transfer plan can be implemented in a variety of ways: optionally, constructing a resource feature extraction model, calculating the platform load balance, performing task dependency analysis, and generating an optimal transfer plan; optionally, establishing a platform scoring system, identifying performance bottlenecks, formulating a task migration strategy, and verifying the feasibility of the plan. It is understandable that other decision algorithms can also be used to implement the formulation of the task transfer plan, which is not limited here.

[0088] In the embodiments of the present application, due to the adoption of a dynamic task decomposition mechanism based on task characteristics and computing platform resource configuration, as well as a real-time monitoring and adaptive task transfer strategy, it is possible to accurately identify the resource demand characteristics of computing tasks, and perform intelligent task allocation and dynamic adjustment according to the real-time status of multiple target computing platforms, thereby effectively solving the problems of unbalanced task allocation and low resource utilization in traditional hyper-converged systems, and thus achieving efficient utilization of computing resources, which not only improves the operating efficiency and reliability of the entire hyper-converged system, but also reduces the complexity of system maintenance and management, and provides users with more stable and efficient computing service support.

[0089] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram may represent a module, a program segment, or a portion of a code, and the above module, program segment, or a portion of a code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings.

[0090] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A hyper-converged server system suitable for edge and end-side data centers, characterized in that: The hyper-converged server system comprises: Chassis; the chassis is the shell of the hyper-converged server system, used to accommodate various components; Computing module; the computing module is arranged at the front of the chassis, the computing module comprises a plurality of CPU modules, and the CPU modules are used to provide computing power required for general business; GPU module; the GPU module is arranged at the rear of the chassis, the GPU module comprises a plurality of GPU modules, and the GPU module is used to provide the computing power required for cloud rendering, cloud gaming, AI reasoning and small model training services; A fan module; the fan module is arranged in the middle of the chassis, and the fan module includes a plurality of cooling fans, and the cooling fans are used to form an air duct from the front of the chassis to the rear of the chassis; Network switching module; the network switching module is arranged at the rear lower layer of the chassis, and is used to provide network signal exchange between the computing module and external devices; An out-of-band management module; the out-of-band management module is arranged at the rear lower layer of the chassis, and is used to send control signals to the computing module, the GPU module and the fan module to realize control management of the computing module, the GPU module and the fan module; Redundant power supply; the redundant power supply is arranged at the rear lower layer of the chassis, and is used to provide operating current to various components of the hyper-converged server system; The middle plate is arranged in the middle of the chassis and located at the bottom of the fan module, and is respectively connected to the computing module, the GPU module, the fan module, the network switching module, the out-of-band management module and the redundant power supply; the middle plate is used to provide transmission of the network signal, the control signal and the operating current.

2. The hyper-converged server system according to claim 1, characterized in that: The hyper-converged server system also includes a storage module; the storage module can be used to replace the GPU module, or replace part of the GPU module of the GPU module; the storage module is arranged at the rear of the chassis, and includes multiple storage media for providing data storage.

3. The hyper-converged server system according to claim 2, characterized in that: The computing module, the GPU module and the storage module all support hot-plug operations.

4. The hyper-converged server system according to claim 1, characterized in that: The network switching module includes multiple network interfaces, and one of the CPU modules of the computing module is communicatively connected to the multiple network interfaces. The network switching module can receive the network control signal of the out-of-band management module to realize network switching of the CPU module on the multiple network interfaces.

5. The hyper-converged server system according to claim 1, characterized in that: The middle board is provided with a plurality of video signal channels, which are respectively used to receive the video signals transmitted by the computing module and the GPU module; the computing module and the GPU module perform high-speed signal interaction through the middle board.

6. The hyper-converged server system according to claim 5, characterized in that: A CPU module of the computing module can simultaneously exchange information with one or more GPU modules of the GPU module through the midplane.

7. The hyper-converged server system according to claim 1, characterized in that: The out-of-band management module receives the data signal stream from the computing module through the midplane, and the out-of-band management module receives the control management signals respectively from the computing module, the GPU module, the fan module, the network switching module and the redundant power supply through the midplane, and sends the data signal stream and the control management signal to the information monitoring terminal.

8. A data processing method suitable for edge and end-side data centers, characterized in that: Applied to a hyper-converged server system, the method comprises: Receiving a computing task sent by a user terminal, and determining a task feature corresponding to the computing task; According to the task characteristics, determining multiple target computing platforms matching the computing task among all computing platform modules on the communication link; Determining resource configurations of the multiple target computing platforms, and generating corresponding computing subtasks based on the resource configuration list and the task characteristics of the computing tasks; Sending the computing subtask to the corresponding target computing platform; Monitor the operating data of the multiple target computing platforms and generate an execution status report.

9. The method according to claim 8, characterized in that After the step of monitoring the operation data of the plurality of target computing platforms and generating an execution status report, the method further includes: When detecting that a new computing platform is connected, obtaining hardware configuration information and performance indicators of the new computing platform, and generating a resource feature vector; Determining, according to the execution status report and the resource feature vector, a priority score for task transfer for each of the target computing platforms; Based on the priority score, screening the computing platform to be transferred and the corresponding task to be transferred from the target computing platform; A task transfer list is generated according to the task dependencies and resource requirements of the tasks to be transferred.

10. The method according to claim 8, characterized in that The step of monitoring the operation data of the multiple target computing platforms and generating an execution status report specifically includes: Collecting raw monitoring data streams including CPU occupancy, memory usage, storage capacity, network bandwidth, and energy consumption data of the multiple target computing platforms; The original monitoring data stream is subjected to standardization processing and time-series storage to generate a standard performance indicator data set containing a timestamp; Based on the standard performance indicator data set, an execution status report including performance scores, anomaly detection results and optimization suggestions is generated.

Images