Server and server design method

Abstract

The invention relates to the technical field of computer servers, in particular to a server and a server design method.The server comprises a basic framework unit, a heterogeneous computing module, an intelligent storage module, a configurable network module, an intelligent management control unit and a system firmware layer which are interconnected through standardized interfaces; the management control unit drives calculation, storage and network resources to perform collaborative dynamic reconstruction through real-time monitoring and AI prediction; the design method comprises six systematic stages of demand analysis, modular design, resource virtualization, dynamic reconstruction mechanism design, energy efficiency optimization and reliability design. According to the method, the problems of low utilization rate, poor expansibility and low energy efficiency caused by static configuration of traditional server resources are solved, on-demand accurate supply of the resources is realized, the resource utilization rate, the energy efficiency ratio and the cross-scene adaptive capacity are remarkably improved, and the method can be widely applied to the fields of cloud computing, high-performance computing, edge computing and the like.

Description

Technical Field

[0001] This invention relates to the field of computer server technology, and more specifically to servers and server design methods. Background Technology

[0002] With the widespread adoption of cloud computing, big data, and artificial intelligence applications, data centers are facing increasingly diverse and dynamic workloads. Traditional servers typically employ fixed hardware configurations, with the ratio of computing, storage, and network resources determined at the factory. This makes it difficult to flexibly adapt to changes in the needs of different application scenarios or different stages of the same application. This leads to two common problems: when the load is low, a large number of resources are idle, resulting in low overall utilization; and when the load is at its peak, performance is limited because a certain type of resource becomes a bottleneck.

[0003] Existing technologies include some modular or scalable server solutions. Some servers support expanding computing power by adding CPUs or memory boards, or expanding storage capacity by adding hard disk backplanes. However, these solutions are mostly limited to vertical expansion of a single type of resource and lack the ability to horizontally coordinate and reconfigure across computing, storage, and network resources. Adjustments to resource configurations often rely on manual intervention by administrators, resulting in slow response times and an inability to achieve forward-looking resource scheduling. Furthermore, most server architectures are designed for specific processors or network protocols, making it difficult to adapt to the needs of heterogeneous computing and hybrid network environments.

[0004] Therefore, those skilled in the art have provided servers and server design methods to address the problems raised in the background section. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a server and server design method to solve the technical issues of low utilization, inflexible expansion, poor scenario adaptability, and delayed resource configuration response caused by the static configuration of traditional server resources. This invention achieves real-time monitoring, forward prediction, and collaborative dynamic reconstruction of computing, storage, and network resources through the deep integration of hardware modular design and software-defined intelligent scheduling.

[0006] Servers and server design methods, including basic framework units, at least one computing resource module, at least one storage resource module, at least one network interconnection module, management and control unit, and system firmware layer; The basic framework unit is used to provide standardized physical installation interfaces, power supply and heat dissipation basic environment for each module; The computing resource module is a hot-swappable heterogeneous computing unit that supports selective configuration of x86, ARM or RISC-V architecture processors and integrates an out-of-band management controller. The storage resource module is a hot-swappable storage unit that supports NVMe, SAS, or SATA hybrid interfaces and has storage resource pooling and intelligent data tiering based on access characteristics. The network interconnection module is a hot-swappable network unit that supports selective configuration of Ethernet, InfiniBand, or RoCE network protocols and integrates a software-defined network controller. The management and control unit includes a central resource scheduler, an AI load prediction engine, and a dynamic reconfiguration controller. The central resource scheduler is used to monitor the resource status of each module in real time. The AI ​​load prediction engine is used to predict future resource requirements based on historical load data. The dynamic reconfiguration controller is used to coordinate the dynamic configuration adjustment of computing, storage, and network resources by sending instructions to the computing resource module, storage resource module, and network interconnection module according to the prediction results and real-time status. The system firmware layer is used to provide a unified hardware abstraction interface, secure boot, and firmware hot upgrade support for upper-layer software. The computing resource module, storage resource module, and network interconnection module are connected to the basic framework unit through standardized electrical and mechanical interfaces, and interact with the management control unit and system firmware layer through the system bus.

[0007] Preferably, the AI ​​load prediction engine uses a long short-term memory neural network algorithm to build a prediction model with a prediction accuracy of no less than 94%, and can predict the load peak 300 milliseconds to 5 seconds in advance.

[0008] Preferably, the intelligent data tiering function of the storage resource module automatically migrates data between NVMe SSDs, SAS SSDs and SATA HDDs based on the configured access frequency threshold and access latency threshold, and the data migration process has less than 5% impact on the input / output performance of the front-end business.

[0009] Preferably, the software-defined network controller of the network interconnection module is based on the OpenFlow protocol and can complete the dynamic reconstruction of the network topology in no more than 50 milliseconds.

[0010] Preferably, during the dynamic resource configuration adjustment process executed by the dynamic reconfiguration controller, business services are not interrupted, and the application performance fluctuation is less than 3%.

[0011] Server design methodology includes the following steps: S1. Requirements Analysis Phase: Collect the resource requirements of the target workload for computing, storage, and networking, establish a multi-dimensional resource requirements model, and determine performance, power consumption, and cost constraints. S2, Modular Design Phase: Define standardized electrical interfaces, mechanical dimensions, and heat dissipation specifications for computing, storage, and network modules; design low-latency communication protocols and hot-swap management mechanisms between modules. S3. Resource Virtualization Design Phase: Design a hardware resource abstraction layer to map physical resources into virtual resource pools, providing a unified resource scheduling interface and a multi-tenant resource isolation mechanism. S4. Dynamic Reconfiguration Mechanism Design Phase: Deploy resource monitoring probes to collect resource status in real time, establish a resource demand prediction model based on machine learning algorithms, and design an online resource migration algorithm that supports service continuity. S5. Energy efficiency optimization design stage: Establish the power consumption-utilization relationship model of each module, design dynamic voltage and frequency adjustment strategies and pre-scheduling energy-saving mechanisms based on load forecasting; S6. Reliability Design Phase: Implement N+1 redundancy configuration for critical modules, design fault prediction and health management modules, and implement rapid fault isolation and automatic recovery mechanisms.

[0012] Preferably, the resource demand prediction model is trained using a long short-term memory neural network algorithm, with a learning rate ranging from 0.001 to 0.01 and an iteration count of 1000 to 5000.

[0013] Preferably, the online resource migration algorithm includes: memory pre-copying technology for computing task migration, incremental migration technology for storage data migration, and session persistence technology for network connection migration.

[0014] Preferably, in step S5, the dynamic voltage frequency adjustment strategy is as follows: when the central processing unit utilization is less than 20%, its operating frequency is adjusted to 50% of the reference frequency; when the utilization is greater than 80%, its operating frequency is increased to 115% of the reference frequency.

[0015] The technical effects and advantages of this invention are as follows: Through AI prediction and collaborative dynamic reconstruction, resource supply is made to closely follow the load curve. Real-world tests show that in a hybrid load cloud platform, the average resource utilization rate can be increased from about 42.5% of traditional servers to more than 78.2%, an increase of more than 84%.

[0016] The combination of intelligent power management strategies and efficient heat dissipation design enables the overall power efficiency to be as low as 1.12, which is significantly better than the industry average, and the energy-saving effect is particularly obvious in scenarios such as edge computing.

[0017] It supports heterogeneous processors and multi-protocol networks. Through different combinations of modules and dynamic software configuration, the same hardware platform can be efficiently adapted to diverse scenarios from cloud computing and high-performance computing to edge computing, greatly enhancing deployment flexibility.

[0018] The combination of hardware redundancy and software self-healing mechanisms can achieve 99.999% system availability. The modular hot-swappable design makes it extremely easy to replace faulty modules and upgrade hardware, greatly reducing the complexity and cost of operation and maintenance.

[0019] Improved resource utilization and energy efficiency directly reduce the number of servers required and electricity costs. Modular upgrades extend the lifespan of the infrastructure, avoiding frequent replacement of entire machines and significantly reducing the total cost of ownership in the long run. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the server structure provided in an embodiment of this application; Figure 2 This is a framework diagram of the server provided in the embodiments of this application; Figure 3 This is a flowchart of the server design method provided in the embodiments of this application. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and design various embodiments with various modifications suitable for a particular purpose.

[0022] Example 1 Please see Figures 1-3 This embodiment provides a server and a server design method, including... A server comprising six core components: a basic framework unit, a computing resource module, a storage resource module, a network interconnection module, a management and control unit, and a system firmware layer.

[0023] Basic frame unit: Serving as the physical skeleton of the server, it adopts a standard 19-inch rack-mount design. Its core includes: Standardized backplane: Provides high-density, high-reliability gold finger or high-speed connector slots for mounting other functional modules. The backplane integrates power bus and data bus.

[0024] Redundant power supply system: Supports multiple hot-swappable redundant power supply modules, with a single module power of no less than 1600W, and supports zero-second switching in case of failure.

[0025] Intelligent cooling system: It adopts a temperature-controlled fan array with independent zone control, which can accurately adjust the fan speed according to the real-time temperature of each module to improve heat dissipation efficiency.

[0026] Computing resource module: a hot-swappable computing card.

[0027] Its innovation lies in: Heterogeneous compatibility: A single computing card can be flexibly configured with x86, ARM or RISC-V architecture processors, supporting the construction of heterogeneous computing environments within a single server.

[0028] Independent management and configuration: Each module integrates an independent basic management controller, supports the IPMI protocol, and can report data such as temperature and power consumption in real time. The memory capacity can be dynamically configured as needed.

[0029] Storage resource module: a hot-swappable storage expansion unit.

[0030] Its core functions include: Hybrid interface support: Supports multiple storage media such as NVMe, SAS, and SATA to meet different performance and capacity requirements.

[0031] Resource pooling and intelligent tiering: The built-in storage controller virtualizes physical storage devices into a unified resource pool and can automatically migrate data between high-speed NVMe SSDs, mid-speed SAS SSDs and large-capacity SATA HDDs based on data access frequency and latency requirements, achieving the optimal balance between performance and cost.

[0032] Network interconnection module: a hot-swappable network daughter card.

[0033] Its main characteristics are: Configurable with multiple protocols: The same hardware platform can support high-speed Ethernet, InfiniBand or RoCE network protocols by loading different firmware.

[0034] Software-defined networking capabilities: Integrated SDN controller, which can dynamically adjust network bandwidth allocation, VLAN division and other topologies within milliseconds according to management commands.

[0035] Network function hardware acceleration: By integrating FPGA, network functions such as virtual firewall and load balancing are offloaded to hardware, significantly improving processing efficiency.

[0036] Management and control unit: This is the "intelligent brain" of the entire server and the core of the invention's dynamic reconfiguration.

[0037] It contains: Central Resource Scheduler: Collects real-time operating status data of all modules at a high frequency through system buses such as I2C and SMBus, including CPU utilization, memory usage, IOPS, and bandwidth.

[0038] AI load prediction engine: It uses machine learning algorithms such as long short-term memory neural networks to analyze historical load sequences and can predict the peak resource demand in the near future with high accuracy, thus achieving forward-looking scheduling.

[0039] Dynamic Reconfiguration Controller: Receives prediction results and real-time status, generates collaborative reconfiguration strategies, and issues specific instructions; for example, it wakes up dormant computing cores in advance, allocates more storage cache to specific applications, and adjusts network port bandwidth before predicting an increase in computing load.

[0040] System firmware layer: Provides a unified operating foundation for software, including: Unified Hardware Abstraction Layer: Based on standards such as ACPI, it shields the differences between underlying heterogeneous hardware, enabling operating systems and virtualization software to identify and manage processors and hardware devices of different architectures without modification.

[0041] Enhanced security and maintainability: The integrated TPM security chip supports trusted boot and provides online hot-upgrade capabilities for firmware of each module, ensuring uninterrupted service.

[0042] A server design methodology: This method is a systematic engineering process that consists of six ordered phases: Step S1: Requirements Analysis Phase; Collect detailed load characteristics of the target application scenario through stress testing tools, quantify the resource requirements of computing, storage, and network, and establish a three-dimensional resource model to clarify the constraints of performance, power consumption, and cost.

[0043] Step S2: Modular design phase; Define standardized interface specifications for each functional module - electrical, mechanical, and heat dissipation, and design inter-module communication protocols and hot-swap management circuits to ensure low latency and high reliability.

[0044] Step S3: Resource virtualization design phase; design a software layer that abstracts and pools physical resources, provides a concise resource scheduling API, and implements strict multi-tenant resource isolation to ensure security and performance.

[0045] Step S4: Dynamic Reconfiguration Mechanism Design Phase; This is the key to achieving "intelligence". It requires designing and deploying lightweight monitoring probes in each module, building and training high-precision AI prediction models, and developing online migration algorithms that can maintain application continuity when adjusting resources - such as memory pre-copying and incremental data synchronization.

[0046] Step S5: Energy efficiency optimization design stage; establish the power consumption model of each module through experimental measurement, find the operating point with optimal energy efficiency, and on this basis, design a dynamic voltage and frequency adjustment strategy and a pre-scheduled energy-saving strategy that combines real-time load and prediction information.

[0047] Step S6: Reliability design phase; implement N+1 redundancy design for key components such as power supply and fans from a hardware perspective; develop fault prediction algorithms from a software perspective to achieve early warning of hardware aging; and design an automatic fault switching process to ensure high system availability.

[0048] Applications of this invention: Application 1: Application in cloud computing virtualization platforms Deployment Configuration: A resource pool is built using 3 servers based on this invention, with the following configuration for each server: Four compute modules – each containing two Intel Xeon Gold 6338 processors; Two storage modules - each containing 12 3.84TB NVMe SSDs; One network module (dual-port 100GbE), redundant power supply and fan; Deploy Kubernetes and OpenStack at the software layer.

[0049] Test load: Simulates a hybrid cloud load including web services, databases, and batch jobs, running continuously for 72 hours.

[0050] Implementation Process and Results: During the test, the system continuously monitored the load. When the AI ​​engine predicted that the Web service traffic would surge after 300ms, the dynamic reconfiguration controller coordinated to complete the following operations within 50ms: switched the CPUs of the two computing modules from power saving mode to high performance mode; expanded the data cache capacity allocated to the Web service in the storage cache pool from 128GB to 256GB; and increased the virtual network bandwidth corresponding to the Web service from 10Gbps to 25Gbps through the SDN controller.

[0051] Key data comparison: Average resource utilization: The server of this invention achieved 78.2%, compared to 42.5% for the traditional static server in the control group.

[0052] Virtual machine startup time: average 4.2 seconds, compared to 7.8 seconds in the control group.

[0053] Energy efficiency ratio: 15.2 SPECcpu2017 / W, compared to 9.8 for the control group.

[0054] The entire reconstruction process was uninterrupted, with performance jitter less than 2%.

[0055] Application 2: Applications in high-performance computing clusters Deployment configuration: A cluster is formed by 16 servers of this invention; each server is equipped with a GPU accelerated computing module, a high-speed NVMe-oF storage module and a 200Gb / s InfiniBand network module.

[0056] Test load: Run computational fluid dynamics and molecular dynamics simulation tasks.

[0057] Implementation Process and Results: During the execution of the GROMACS task, the system detected the start of the communication-intensive phase. The dynamic reconfiguration controller automatically switched the network topology from fully connected to a 3D toroidal topology more suitable for this communication mode, reducing the number of communication hops. Simultaneously, checkpoint data for the computation phase was migrated from high-performance NVMe disks to high-capacity QLC SSDs, freeing up high-speed space for hot data.

[0058] Key data comparison: GROMACS simulation performance: up to 2.15 μs / day, a 45% improvement over traditional clusters.

[0059] Cluster energy efficiency: 8.7 PFLOPS·h of computing power per kilowatt-hour, an improvement of 67%.

[0060] MPI communication overhead: reduced to 12.3%.

[0061] Application 3: Applications in edge computing nodes Deployment and Configuration: Deployed on the 5G base station side, it adopts a compact design and is equipped with a low-power ARM computing module, an industrial-grade wide-temperature storage module, and a multi-mode network module that supports 5G / LTE / Ethernet.

[0062] Test scenario: Intelligent transportation video AI analysis.

[0063] Implementation process and results: The system automatically executes daily cycle scheduling strategies based on historical patterns: video analysis resources are prioritized during the morning peak hours; at night, it enters a deep energy-saving mode, reducing power consumption to 85W; when the 5G main link is interrupted, the network module switches to the LTE backup link within 120ms and automatically enables data compression.

[0064] Key data: Typical load power consumption: 245W, which is 36% lower than that of traditional edge servers.

[0065] Link switching time: 85ms.

[0066] Space density: Each rack unit can support 128 video streams for analysis.

[0067] Application 4: Reliability and Energy Efficiency Special Verification Reliability testing: A 30-day continuous stress test was conducted on a single server, and anomalies such as module failure and power outage were injected.

[0068] The results show that: The overall system availability reached 99.999%.

[0069] The average isolation and recovery time for faulty modules is 1.2 minutes.

[0070] The fault prediction module can provide early warning of potential storage media failures up to 72 hours in advance, with an accuracy rate of >96%.

[0071] Energy efficiency test: Measure the power consumption of the whole machine under different loads.

[0072] The results show that under 10% light load, power consumption can be reduced to 32% of that under full load through dynamic energy-saving technology. In a typical data center environment, the overall PUE value can reach 1.12.

[0073] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention, unless otherwise specified or limited, shall be implemented according to conventional means in the art.

Claims

1. A server, characterized in that, include The basic framework unit provides standardized physical mounting interfaces, power supply, and heat dissipation basic environment for each module; At least one computing resource module is a hot-swappable heterogeneous computing unit that supports selective configuration of x86, ARM or RISC-V architecture processors and integrates an out-of-band management controller. At least one storage resource module is a hot-swappable storage unit that supports a hybrid NVMe, SAS, or SATA interface and features storage resource pooling and intelligent data tiering based on access characteristics. At least one network interconnect module is a hot-swappable network unit that supports selective configuration of Ethernet, InfiniBand, or RoCE network protocols and integrates a software-defined network controller. The management and control unit includes a central resource scheduler, an AI load prediction engine, and a dynamic reconfiguration controller; The system firmware layer provides a unified hardware abstraction interface, secure boot, and firmware hot-upgrade support for upper-layer software. The computing resource module, storage resource module, and network interconnection module are connected to the basic framework unit through standardized electrical and mechanical interfaces, and interact with the management control unit and system firmware layer through the system bus.

2. The server according to claim 1, characterized in that, The central resource scheduler is used to monitor the resource status of each module in real time. The AI ​​load prediction engine is used to predict future resource needs based on historical load data; The dynamic reconfiguration controller is used to coordinate the dynamic configuration adjustment of computing, storage and network resources by sending instructions to the computing resource module, storage resource module and network interconnection module according to the prediction results and real-time status. The AI ​​load prediction engine uses a long short-term memory neural network algorithm to build a prediction model with a prediction accuracy of no less than 94%, and can predict load peaks 300 milliseconds to 5 seconds in advance.

3. The server and server design method according to claim 1, characterized in that, The intelligent data tiering function of the storage resource module automatically migrates data between NVMe SSDs, SASSSDs, and SATA HDDs based on the configured access frequency threshold and access latency threshold, and the data migration process has less than 5% impact on the input / output performance of the front-end business.

4. The server and server design method according to claim 1, characterized in that, The software-defined network controller of the network interconnection module is based on the OpenFlow protocol and can complete the dynamic reconstruction of the network topology in no more than 50 milliseconds.

5. The server and server design method according to claim 1, characterized in that, During the dynamic resource configuration adjustment process executed by the dynamic reconfiguration controller, business services are not interrupted, and application performance fluctuations are less than 3%.

6. A server design method, characterized in that, Includes the following steps: S1. Requirements Analysis Phase: Collect the resource requirements of the target workload for computing, storage, and networking, establish a multi-dimensional resource requirements model, and determine performance, power consumption, and cost constraints. S2, Modular Design Phase: Define standardized electrical interfaces, mechanical dimensions, and heat dissipation specifications for computing, storage, and network modules; design low-latency communication protocols and hot-swap management mechanisms between modules. S3. Resource Virtualization Design Phase: Design a hardware resource abstraction layer to map physical resources into virtual resource pools, providing a unified resource scheduling interface and a multi-tenant resource isolation mechanism. S4. Dynamic Reconfiguration Mechanism Design Phase: Deploy resource monitoring probes to collect resource status in real time, establish a resource demand prediction model based on machine learning algorithms, and design an online resource migration algorithm that supports service continuity. S5. Energy efficiency optimization design stage: Establish the power consumption-utilization relationship model of each module, design dynamic voltage and frequency adjustment strategies and pre-scheduling energy-saving mechanisms based on load forecasting; S6. Reliability Design Phase: Implement N+1 redundancy configuration for critical modules, design fault prediction and health management modules, and implement rapid fault isolation and automatic recovery mechanisms.

7. The server design method according to claim 6, characterized in that, In step S4, the resource demand prediction model is trained using a long short-term memory neural network algorithm with a learning rate ranging from 0.001 to 0.01 and an iteration count of 1000 to 5000.

8. The server design method according to claim 6, characterized in that, In step S4, the online resource migration algorithm includes: memory pre-copying technology for computing task migration, incremental migration technology for storage data migration, and session persistence technology for network connection migration.

9. The server design method according to claim 6, characterized in that, In step S5, the dynamic voltage frequency adjustment strategy is as follows: when the central processing unit utilization is below 20%, its operating frequency is adjusted to 50% of the reference frequency; when the utilization is above 80%, its operating frequency is increased to 115% of the reference frequency.

Images