A cloud resource proportion balancing allocation method and device for optimizing national computing power scheduling efficiency

Abstract

The application discloses a cloud resource proportion balancing allocation method and device for optimizing national computing power scheduling efficiency, and belongs to the technical field of national computing power scheduling, cloud computing resource management, localized computing power operation and maintenance and special computing power chip. The application is based on the original proportion balancing theory and STE localized computing power scheduling code, carries the localized cloud computing power scheduling special chip and the CloudBalance-AI balancing scheduling model, constructs a whole-process system of global computing power investigation, proportion balancing modeling, dynamic allocation scheduling, real-time load calibration and localized execution, and sets core indexes such as scheduling efficiency improvement >=72%, resource idle rate <=12%, computing power utilization rate >=88%, cross-region time delay <=8ms and load deviation <=+ / -3%. Through global resource collection, proportion model construction, dynamic resource matching and load calibration, the problems of resource imbalance and idle overload are solved, and national-level cloud resource overall deployment is realized, and 100% of the whole process is localized without foreign dependence. The application is suitable for government cloud, industrial cloud, public computing power platform and other scenes, can save computing power resources and improve intensive operation efficiency, and fills the gap of localized cloud resource proportion balancing scheduling technology.

Description

Technical Field

[0001] This invention pertains to national computing infrastructure optimization, cloud computing resource scheduling, load balancing, and domestically produced computing chips. In the fields of chip design and AI intelligent computing power operation and maintenance technology, specifically involving a cloud computing power scheduling efficiency optimization method. Resource proportional balancing allocation methods and devices are particularly suitable for national government clouds, industrial internet clouds, and public computing. Power service platforms, cross-regional cloud computing clusters, and classified computing platforms all affect scheduling efficiency and resource utilization. National-level cloud computing power scheduling scenarios with stringent requirements for autonomy and controllability. Background Technology

[0002] Currently, the scheduling and allocation of national cloud computing resources face numerous intractable technical challenges, severely restricting the development of computing power. Infrastructure efficiency: Firstly, existing cloud resources mostly adopt a fixed allocation and temporary on-demand deployment model, without the construction of infrastructure. The fundamental principle of balanced resource allocation in science has been disrupted, resulting in a severe imbalance in resource allocation and overload of computing power on some nodes. Some nodes have been idle for a long time, with a computing power utilization rate of less than 50%, resulting in a huge waste of national computing resources; secondly... The lack of a comprehensive cloud resource allocation mechanism results in poor coordination and high latency in cross-regional computing power allocation, making it impossible to achieve full coverage. Third, insufficient dynamic scheduling capabilities and lack of a real-time load calibration mechanism mean that the allocation ratio cannot be quickly adjusted when the load fluctuates, resulting in poor business operation stability; fourth, core scheduling algorithms and cloud management... The management platform and scheduling chip both rely on overseas technology, posing risks of technological bottlenecks, data leaks, and malicious control. Fifth, the domestic production and controllability are extremely poor; and sixth, the scheduling instructions lack dedicated domestic coding, making instruction transmission susceptible to interference. Insufficient scheduling execution precision makes it impossible to achieve efficient, accurate, and secure national-level computing power scheduling, thus failing to meet national needs. The need for efficient and autonomous development of home computing power.

[0003] To address the aforementioned technical shortcomings, this invention, starting from the fundamental principle of proportional balance, constructs a domestically produced, full-domain cloud resource equalization system. The balanced scheduling system, combining dedicated domestic computing chips with an AI intelligent scheduling model, achieves precise allocation of cloud resources. Allocation, overall coordination and scheduling, and real-time balance calibration comprehensively improve the efficiency of national computing power scheduling and resource utilization. Summary of the Invention

[0004] 1. Core Definition 1. Fundamental Proportional Balancing Logic: Based on mathematical fundamental proportions, combined with cloud resource node computing power reserves, load, and business operations. Prioritize and establish a scientific allocation benchmark to achieve a balanced distribution of computing resources. 2. Computing power utilization: Actual effective computing power consumption of a node / Total computing power capacity × 100%, measuring the utilization of computing power resources. efficiency 3. Load balancing deviation: The difference between the actual load and the baseline load balancing ratio. The smaller the deviation, the better the load balancing. 4. STE Computing Power Scheduling Coding: Domestically developed dedicated cloud scheduling instruction coding, ensuring the security, accuracy, and high efficiency of scheduling instructions. Effective Execution 2. Key quantization thresholds (can be directly accessed by the programmer) 1. Improved computing power scheduling efficiency: ≥72%, a significant optimization compared to traditional scheduling modes. 2. Cloud resource idle rate: ≤12%, a reduction of ≥65% compared to the traditional model. 3. Overall computing power utilization: ≥88%, completely solving the problems of idle and overloaded computing power. 4. Cross-regional computing power scheduling latency: ≤8ms, significantly improving global scheduling coordination. 5. Load balancing deviation: ≤±3%, extremely high allocation accuracy. 6. Intrinsic proportional modeling latency: ≤2ms, dynamic scheduling response time ≤3ms 7. STE scheduling code bit width: 128 bits, instruction transmission bit error rate ≤ 10^-15 8. AI scheduling model inference latency: ≤1.5ms, load prediction accuracy ≥98.8%

[0005] 1. Global computing power resource statistics formula Ctotal =∑i=1n Ci ,Mtotal =∑i=1n Mi ,Btotal =∑i=1n Bi Among them: Ctotal / Mtotal / Btotal This refers to the total computing power resources of the entire domain, including CPU, memory, and bandwidth. Ci / Mi / Bi This represents the resource quantity corresponding to a single node, where n is the total number of cloud resource nodes. 2. Formula for the original proportion equilibrium distribution model Ri = ω1 . Ctotal Cused +ω2 . Mtotal Mused +ω3. Total Bused +ω4 . Pi Alloci =Ri . (Ctotal +Mtotal +Btotal ). η where: 1. Ri Assigning a proportionality coefficient to the source of node i 2. ω1 / ω2 / ω3 / ω4 The resource weighting coefficients are ω1 + ω2 + ω3 + ω4 = 1 3. Pi This is the business priority coefficient (core government / industrial business priority ≥ 0.8). 4. Alloci Equal allocation of resources to node i 5. η is the equalization calibration coefficient (dynamically adjusted from 0.95 to 1.05). 3. Load balancing deviation calculation formula ΔLi = |Lreal −Lalloc| × 100% When > , start real Timely calibration to ensure load deviation ≤ ±3%. 4. Formula for calculating computing power utilization rate Ui =Ci +Mi +Bi Cused +Mused +Bused ×100% target computing power Usage rate Ui ≥ 88% 5. STE Domestic Computing Power Scheduling Coding Rules STECloud =Hash(Alloci )⊕KCloud ⊕IDNode ⊕CRCSched 1. Hash(Alloci): A hash value for a balanced distribution instruction. 2. KCloud 128-bit domestically developed cloud scheduling dedicated key, generated by hardware TRNG. 3. IDNode Unique identifier for cloud resource nodes 4. CRCSched Dispatch instruction check code 5. ⊕: Low-level XOR encoding, instructions are irreversible, only domestically produced chips can decode and execute it.

[0006] Chip model: GZ-Cloud-RISC-V, main frequency 1.2GHz, dedicated to cloud computing power scheduling, computing power ≥3.0 TOPS. Supports multi-node parallel scheduling, wide-temperature industrial-grade adaptation, anti-interference and anti-attack, fully domestically produced hardware, and boundless possibilities. External components and technical backdoors.

[0007] asm; Global computing resource acquisition command CLOUD_COLLECT rD, rNode Global node computing power / load / demand collection, single cycle implement Original Proportional Equilibrium Modeling Command RATIO_MODEL rD, rRes Construct a proportional allocation model, calculate the equilibrium coefficient, and perform bi-weekly calculations. Execution period Dynamic resource allocation command RES_ALLOC rD, rRatio Dynamic allocation of CPU, memory, and bandwidth; dual-cycle execution. STE Scheduling Instruction Encoding Instruction SCHED_ENCODE rD, rCmd Scheduling instructions are STE-encoded and encrypted, and executed in two cycles. Real-time load monitoring commands LOAD_MON rD, rStatus Node load is monitored in real time, and execution is performed in a single cycle. Balance deviation calibration command BALANCE_CALIB rD, rErr Dynamic calibration of load deviation, executed in a single cycle. Dispatch instructions are issued and execution instructions are executed. SCHED_EXEC rD, rNode The encoded instructions are issued to the node for execution, and are executed in a single cycle. The chip integrates a hardware encryption module, a true random number generator, and a multi-node parallel scheduling unit, supporting scheduling instructions. Encrypted transmission, tamper-proof, and interference-proof, compatible with all domestic cloud platforms and computing nodes.

[0008] IV. CloudBalance-AI Cloud Resource Balancing Scheduling Model (In-depth Refinement) 1. Model Architecture: Lightweight Transformer + CNN cloud-specific hybrid architecture, only 9.6M parameters, INT8 Quantitative, cloud-distributed deployment, built on a domestically developed deep learning framework, with no dependence on foreign frameworks.

[0009] 2. Core Model Functions: Global computing power data fusion analysis, intelligent modeling of original proportions, load trend prediction, dynamic... Optimization of dynamic allocation ratio, cross-regional scheduling coordination, and real-time calibration of balance deviation.

[0010] 3. Inference Process: Global computing power data collection → Data preprocessing → AI proportional modeling → Dynamic resource allocation → Load balancing Real-time monitoring → Deviation prediction → Balance calibration → Scheduling execution → Effect feedback optimization.

[0011] 4. Core Model Metrics: Inference latency ≤ 1.5ms, load prediction accuracy ≥ 98.8%, proportional allocation accuracy ≥ 99%. Cross-regional collaborative scheduling efficiency improved by 75%.

[0012] 5. Loss Function: Proportional Allocation Loss + Load Balancing Loss + Computing Power Utilization Loss + Scheduling Latency Loss Four constraints ensure optimal scheduling efficiency and balance.

[0013] 1. Utilizing GZ-Cloud-RISC-V chip hardware acceleration, collect data on the CPU, memory, and other performance of national cloud resource nodes. Bandwidth availability, real-time load, service requirements, priority data 2. Calculate the total computing power resources across the entire domain according to the global computing power statistics formula, and establish a standardized global computing power resource ledger. 3. Classify and categorize nodes and services, prioritizing core government, industry, and public service services. 4. Comprehensive data collection with no missing nodes, 1ms collection cycle, and 100% data accuracy.

[0014] 1. Based on the fundamental proportional logic of mathematics, substitute the formula of the fundamental proportional equilibrium distribution model to calculate the distribution ratio of each node. coefficient 2. Based on business priorities and resource weights, establish a differentiated and balanced allocation benchmark to determine the resource allocation for each node. Quota 3. AI model optimization parameters adapt to load fluctuations and business changes, with modeling latency ≤2ms. 4. The entire modeling process utilizes domestically developed algorithms, with no intervention from foreign scheduling logic, ensuring independent and controllable modeling.

[0015] 1. Based on the balanced allocation criteria, CPU, memory, and bandwidth resources are categorized and dynamically matched, and allocated as needed. Each cloud node 2. Avoid resource idleness and overload, prioritize the computing power needs of high-priority services, and ensure scheduling response time ≤ 3ms. 3. In accordance with the STE scheduling coding rules, the allocation instructions are encrypted with 128 bits to ensure secure transmission of instructions. 4. Multi-node parallel scheduling, no resource contention or conflict, and highly efficient global scheduling coordination.

[0016] 1. Monitor the load status of each cloud node in real time and calculate the load imbalance value using the load imbalance formula. 2. When the deviation exceeds 3%, immediately initiate dynamic calibration, adjust resource allocation, and control the deviation within ±3%. Inside 3. AI models predict load fluctuation trends and optimize allocation ratios in advance to achieve proactive balancing. 4. The calibration cycle is 1ms, maintaining balanced computing power throughout the process, with a stable utilization rate of ≥88%.

[0017] 1. STE-encoded instructions are distributed to various cloud resource nodes via a domestically developed scheduling and execution module, and the chips decode and execute them. 2. The entire process is independent of foreign technologies, platforms, and algorithms, ensuring secure and controllable scheduling. 3. Real-time statistics on scheduling efficiency, resource utilization, and idle rate, generating scheduling logs. 4. Closed-loop optimization of the ratio model and scheduling parameters to continuously improve computing power scheduling efficiency.

[0018] The device described in this invention comprises five core modules, which are electrically connected to each other and work together to achieve cloud resource ratio calculation. Balanced allocation and global computing power scheduling: 1. Global Computing Power Acquisition Module: Equipped with the GZ-Cloud-RISC-V chip, it realizes global cloud node computing power, load, and... Real-time collection, statistics, and ledger establishment of demand data 2. Proportional Balancing Modeling Module: Based on the original proportional logic, this module constructs a cloud resource allocation model and calculates the balanced proportion system. Number and allocation quota, AI model-assisted optimization 3. Dynamic Scheduling and Allocation Module: Based on the load balancing ratio, it dynamically allocates CPU, memory, and bandwidth. STE refers to... Encoding, multi-node parallel scheduling 4. Load Monitoring and Calibration Module: Real-time monitoring of node load, calculation of balancing deviation, dynamic calibration of distribution ratio, and maintenance. Maintain global computing power balance 5. Domestically-developed scheduling execution module: decodes scheduling instructions, issues and executes them, ensuring fully domestically-developed scheduling throughout the entire process, and stores logs. Evidence and closed-loop optimization Beneficial effects

[0019] 1. Significantly improved computing power scheduling efficiency: Based on precise allocation according to the original ratio, scheduling efficiency is improved by more than 72%, and cross-platform... Regional scheduling latency ≤8ms, significantly enhancing global computing power synergy. 2. Extreme optimization of resource utilization: computing power utilization ≥ 88%, resource idle rate ≤ 12%, a 65% reduction compared to the traditional model. The above measures completely eliminate idle and overloaded computing power, significantly saving national computing resources. 3. Strong load balancing and stability: Load balancing deviation ≤ ±3%, real-time calibration mechanism ensures continuous load balancing. The service runs smoothly without lag or interruption, with a stability of ≥99.99%. 4. Fully Independent and Controllable Security: The scheduling algorithm, coding rules, dedicated chips, and execution modules are all domestically produced. No reliance on foreign technology, no security backdoors, ensuring national sovereignty over computing power scheduling. 5. High adaptability across all scenarios: Adaptable to national government cloud, industrial cloud, public computing power platforms, cross-regional computing power clusters, etc. The application is highly versatile and can be quickly compatible with existing domestic cloud infrastructure. Detailed Implementation

[0020] 1. Application Scenario: National-level government cloud platform, with multiple regional nodes and various types of concurrent government services, traditional scheduling methods have limitations. Node overload and idleness coexist, with a utilization rate of only 48%. 2. Implementation process: (1) Global data collection: Collect CPU, memory, and bandwidth data from 32 government cloud nodes nationwide. Establish a comprehensive computing power ledger (2) Proportional modeling: Construct a source proportion allocation model according to the priority of government affairs business. Calculate the equilibrium coefficient (3) Dynamic allocation: allocate resources dynamically according to the proportion, and prioritize high-priority approval business. (4) Monitoring and calibration of computing power: Real-time monitoring of load, immediate calibration if deviation exceeds 3%, maintaining balanced operation (5) Scheduling and Execution: Domestically produced chips issue STE-encoded instructions, ensuring full self-control throughout the entire process. 3. Implementation Results: The utilization rate of government cloud computing power increased to 89%, the idle rate decreased to 11%, and scheduling efficiency improved by 75%. Cross-regional dispatch latency is 6ms, business operations are stable with no overload or lag, and the entire process is made domestically produced to ensure the smooth operation of government data. Safety.

[0021] 1. Application Scenario: Industrial Internet cloud platform, concurrent production line control and industrial data analysis, computing power requirements. Large fluctuations and low efficiency of traditional scheduling 2. Implementation process: (1) Collect data on computing power, load, and business requirements of each node in the industrial cloud; (2) Build an industrial cloud platform. Service-specific proportional balancing model to adapt to computing power fluctuation requirements (3) Dynamically allocate CPU and memory resources to avoid (4) Real-time calibration of load deviation to ensure the real-time performance of industrial operations (5) Domestic scheduling execution Yes, no foreign dependence. 3. Implementation Results: Computing power utilization rate reached 88.5%, idle rate was 10%, scheduling efficiency improved by 73%, and industrial business operations were enhanced. It operates without delay, meeting the high real-time requirements of the Industrial Internet.

[0022] 1. Application Scenarios: A national-level public computing power service platform, providing computing power services to the public, scientific research, and small and medium-sized enterprises. Dispersed resource demand 2. Implementation process: (1) Conduct a comprehensive assessment of computing power across the entire domain and compile statistics on the resources and needs of each node. (2) Model the original proportions and public... (3) Dynamic scheduling and on-demand allocation of computing resources to avoid resource waste. (4) Real-time monitoring and calibration. Accurate, maintain overall platform balance (5) Domestic scheduling, ensure safe and efficient public computing power 3. Implementation Results: Computing power utilization rate reached 88%, idle rate was 12%, scheduling efficiency improved by 72%, and it can simultaneously handle more... The multi-service, intensive utilization of computing resources has yielded significant results.

Claims

1. A method for optimizing the proportional balanced allocation of cloud resources to improve the efficiency of national computing power scheduling, characterized in that, Including the following Steps: Collect data on computing power, load, and business requirements of national cloud resource nodes through hardware acceleration, and establish... Global computing power ledger; a balanced allocation model is constructed based on the mathematical principle of proportionality, and a computing node allocation ratio system is established. Number and resource quotas; dynamically allocate CPU, memory, and bandwidth cloud resources according to a balanced ratio, through STE country Production coding generates scheduling instructions; real-time monitoring of node load, calculation of load balancing deviation and dynamic calibration, maintaining negative load. Load deviation ≤ ±3%; scheduling instructions are executed using domestically produced dedicated chips and modules, with no foreign technology involved in the entire process. Dependencies; The entire process is configured with computing power utilization ≥88%, resource idle rate ≤12%, scheduling efficiency improvement ≥72%, and cross-platform compatibility. Regional scheduling latency is ≤8ms, enabling balanced and intensive scheduling of cloud resources across the entire region.

2. The method according to claim 1, characterized in that, Global computing power statistics use a formula Ctotal =∑i=1n Ci ,Mtotal =∑i=1n Mi ,Btotal =∑i=1n Bi, The original proportion is evenly distributed using the formula. Ri = ω1 . Ctotal Cused +ω2 . Mtotal Mused +ω3 Btotal Bused +ω4 . Pi And Alloci = Ri . (Ctotal + Mtotal + Btotal ). η, the sum of the weighting coefficients is 1, The business priority coefficient is dynamically adapted to the core business requirements.

3. The method according to claim 1, characterized in that, Load balancing deviation is expressed by the formula ΔLi = |Lreal −Lalloc| × 100%, calibration is initiated when the deviation > 3%, and the computing power utilization rate is adopted. Use the formula Ui =Ci +Mi +Bi Cused +Mused +Bused ×100%, mesh Standard utilization rate ≥ 88%.

4. The method according to claim 1, characterized in that, STE domestic scheduling coding rules are as follows STECloud =Hash(Alloci)⊕KCloud ⊕IDNode ⊕CRCSched, 128-bit hardware key encryption, instruction transmission error rate ≤10^-15.

5. The method according to claim 1, characterized in that, Powered by GZ-Cloud-RISC-V domestic cloud computing power A dedicated scheduling chip with an instruction set including CLOUD_COLLECT, RATIO_MODEL, RES_ALLOC, SCHED_ENCODE, LOAD_MON, BALANCE_CALIB, and SCHED_EXEC are dedicated instructions for the chip. Frequency ≥1.2GHz, supports multi-node parallel scheduling.

6. The method according to claim 1, characterized in that, CloudBalance-AI cloud resource balancing The scheduling model achieves load prediction and proportional optimization, with inference latency ≤1.5ms and load prediction accuracy ≥98.8%. Proactively optimize scheduling strategies.

7. The method according to claim 1, characterized in that, Prioritize ensuring high-priority national core business computing It meets power requirements, allows for parallel scheduling of multiple nodes without resource conflicts, and is suitable for government cloud, industrial cloud, and public computing power platforms. National-level scenario of cross-regional computing power clusters.

8. A cloud resource proportional equalization allocation device for optimizing national computing power scheduling efficiency, characterized in that, Including global Computing power acquisition module, proportional balancing modeling module, dynamic scheduling and allocation module, load monitoring and calibration module, national The production scheduling and execution module, with each module electrically connected and operating collaboratively, implements the process described in any one of claims 1-7. Cloud resource proportional balanced allocation method.

9. A cloud computing power scheduling terminal device, comprising a memory, a processor, and a device stored in the memory and operable on the processor. A computer program running on, characterized in that, The processor implements the right when executing the computer program. Requirement 1 to 7: Cloud resource proportional equalization allocation method for optimizing national computing power scheduling efficiency. The steps.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program When the sequence is executed by the processor, it implements the optimized national computing power scheduling efficiency as described in any one of claims 1 to 7. The steps of the cloud resource proportional equalization allocation method.