Computing power scheduling method and device and computer readable storage medium
By obtaining the user's computing power needs and generating suggestions for allocating GPUs and power consumption periods, optimizing the use of power resources, solving the problem of high power consumption caused by the increase in computing power demand, and minimizing electricity consumption costs and improving energy efficiency.
Patent Information
- Application Number
- CN202411186655.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-29
AI Technical Summary
The rapid growth of computing power demand has led to huge problems of power consumption and high electricity consumption costs. In the existing technology, uneven computing power distribution and low task scheduling efficiency affect the overall efficiency.
By obtaining the computing power requirements of users, generating suggestions for allocating the number of GPUs and power consumption periods of graphics processors, optimizing the use of power resources, selecting green energy power supply or low peak time periods, and dynamically adjusting the number of GPUs to reduce power costs.
It has achieved the maximization of computing power use during the lowest power cost period, reduced non-emergency computing power demand, reduced electricity consumption costs, optimized computing power distribution and task scheduling, and improved energy utilization efficiency.
Smart Images

Figure CN119045973A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a computing power scheduling method, device and computer-readable storage medium. Background Art
[0002] The current rapid growth in computing power demand has led to huge electricity consumption, making electricity costs high. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a computing power scheduling method, device and computer-readable storage medium in response to the above-mentioned deficiencies in the prior art, so as to solve the problem in the prior art that the rapid growth of computing power demand leads to huge power consumption and high electricity costs.
[0005] In a first aspect, the present invention provides a computing power scheduling method, comprising:
[0006] Obtain the user's computing power requirements;
[0007] Generate, according to the computing power demand, a suggestion including the number of allocated graphics processing units (GPUs) and at least one computing power task execution suggestion for different power usage periods, wherein the power usage periods are divided according to electricity prices at different power supply times, and the at least one computing power task execution suggestion includes a suggestion for ensuring maximum computing power usage during the period with the lowest electricity cost;
[0008] Receive a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion;
[0009] Execute the computing task according to the suggestion for executing the target computing task.
[0010] Further, generating, according to the computing power demand, a suggestion for executing at least one computing power task including the number of allocated GPUs and different power usage periods specifically includes:
[0011] Obtaining the power resource usage and GPU usage of the computing center according to the computing power demand;
[0012] According to the power resource usage, a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period is selected as a time period for executing computing tasks;
[0013] Selecting a GPU for executing the computing task according to the GPU usage and a preset GPU priority principle;
[0014] Based on the time period selected as the time for executing the computing task and the GPU used to execute the computing task, a suggestion for executing the at least one computing task including the number of allocated GPUs and different power usage time periods is generated.
[0015] Furthermore, before selecting a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period as a time period for executing a computing task according to the power resource usage, the method further includes:
[0016] The electricity usage time periods are divided according to the electricity prices corresponding to different power supply time periods of different power supply modes of the computing power center, wherein the power supply modes at least include green energy power supply modes.
[0017] Furthermore, after executing the computing task according to the suggestion for executing the target computing task, the method further includes:
[0018] Monitor the execution status of the computing task according to the number of GPUs and power consumption period corresponding to the computing task.
[0019] Furthermore, the method further comprises:
[0020] The number of GPUs used during the execution of the computing task is dynamically increased or decreased according to the real-time changes in electricity rates.
[0021] Furthermore, the method further comprises:
[0022] Obtain the target number of GPUs and target power consumption period in the recommendation for executing the target computing power task;
[0023] The computing power cost corresponding to the computing power task is obtained according to the target number of GPUs and the target power consumption period.
[0024] Furthermore, obtaining the computing power cost corresponding to the computing power task according to the target number of GPUs and the target power consumption period specifically includes:
[0025] Calculate the total power consumption of all GPUs corresponding to the computing task according to the target number of GPUs and the power of a single GPU;
[0026] Calculate the total power consumption corresponding to the computing task according to the power consumption duration of the target power consumption period and the total power consumption of all GPUs;
[0027] The computing power cost corresponding to the computing power task is calculated according to the electricity price during the target electricity consumption period and the total electricity consumption corresponding to the computing power task.
[0028] In a second aspect, the present invention provides a computing power scheduling device, comprising:
[0029] The first acquisition module is used to obtain the computing power requirements of the user;
[0030] a generating module connected to the first acquiring module, configured to generate, according to the computing power demand, suggestions for executing at least one computing power task including the number of allocated graphics processing units (GPUs) and different power usage periods, wherein the power usage periods are divided according to electricity prices at different power supply times, and the at least one computing power task execution suggestion includes a suggestion for ensuring that computing power usage is maximized during the period with the lowest electricity cost;
[0031] A receiving module, connected to the generating module, configured to receive a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion;
[0032] An execution module is connected to the receiving module and is used to execute the computing task according to the suggestion for executing the target computing task.
[0033] In a third aspect, the present invention provides a computing power scheduling device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the computing power scheduling method described in the first aspect above.
[0034] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the computing power scheduling method described in the first aspect is implemented.
[0035] The computing power scheduling method, device and computer-readable storage medium provided by the present invention first obtain the computing power demand of the user; then generate suggestions including the number of allocated graphics processors GPU and at least one computing power task execution in different power consumption periods according to the computing power demand, wherein the power consumption period is divided according to the electricity price of different power supply times, and the at least one computing power task execution suggestion includes a suggestion to ensure the maximum computing power usage in the period with the lowest electricity cost; then receive the target computing power task execution suggestion selected by the user from the at least one computing power task execution suggestion; finally, execute the computing power task according to the target computing power task execution suggestion. The present invention optimizes computing power allocation and task scheduling, automatically generates at least one suggestion for allocating GPUs and power consumption periods to ensure the maximum computing power usage in the period with the lowest electricity cost according to the computing power demand of the user, and automatically executes computing power tasks according to the suggestions selected by the user, thereby reducing non-emergency computing power demand in the period with higher electricity cost, minimizing electricity cost, and solving the problem that the rapid growth of computing power demand in the prior art leads to huge electricity consumption, resulting in high electricity cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 This is a flow chart of a computing power scheduling method according to Embodiment 1 of the present invention;
[0037] Figure 2This is a schematic diagram of the structure of a computing power scheduling device according to Embodiment 2 of the present invention;
[0038] Figure 3 This is a structural diagram of a computing power scheduling device according to Example 3 of the present invention. DETAILED DESCRIPTION
[0039] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0040] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.
[0041] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.
[0042] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.
[0043] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.
[0044] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.
[0045] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.
[0046] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.
[0047] Application Overview
[0048] At present, in response to the rapid growth of computing power demand, the existing computing power scheduling methods have the following shortcomings:
[0049] (1) Energy consumption: With the rapid development of cloud computing and big data, energy consumption in data centers has become a major issue. The rapid growth in computing power demand has led to huge electricity consumption and corresponding carbon emissions, resulting in high electricity costs;
[0050] (2) Uneven distribution of computing power: In many systems, the allocation of computing power resources is often not efficient. Some nodes may be overloaded while other nodes are idle. This imbalance leads to energy waste.
[0051] (3) Low task scheduling efficiency: Traditional task scheduling algorithms may not be able to respond to load changes in real time or effectively utilize heterogeneous computing resources, affecting overall efficiency.
[0052] In response to the above technical problems, the present application provides a computing power scheduling method, device and computer-readable storage medium, which optimizes computing power allocation and task scheduling, automatically generates at least one suggestion for allocating GPUs and power consumption periods according to the user's computing power needs to ensure maximum computing power usage during the period with the lowest electricity cost, and automatically executes computing power tasks according to the suggestions selected by the user, thereby reducing non-emergency computing power demand during periods with higher electricity costs and minimizing electricity costs, so as to at least solve the problem in the prior art that the rapid growth of computing power demand has led to huge electricity consumption, resulting in high electricity costs.
[0053] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.
[0054] Embodiment 1:
[0055] This embodiment provides a computing power scheduling method, such as Figure 1 As shown, the method includes:
[0056] Step S101: Obtain the computing power requirements of the user.
[0057] It should be noted that the user's computing power demand includes the computing power required by the user and the duration of the rental computing power.
[0058] Step S102: Generate suggestions for executing at least one computing task according to the computing power demand, including the number of allocated GPUs (Graphics Processing Units) and different power consumption time periods, wherein the power consumption time periods are divided according to electricity prices at different power supply times, and the at least one computing task execution suggestion includes a suggestion for ensuring maximum computing power usage during the period with the lowest electricity cost.
[0059] In this embodiment, in order to reduce non-emergency computing power demand during periods of higher electricity costs, based on the user's computing power demand, a suggestion for executing at least one computing power task corresponding to the computing power demand is generated, including the number of allocated GPUs and different power consumption periods, wherein different power consumption periods are divided according to the electricity prices at different power supply times, and at least one computing power task execution suggestion includes a suggestion to ensure maximum computing power utilization during periods of lowest electricity costs, thereby minimizing electricity costs.
[0060] In an optional embodiment, generating, according to the computing power demand, a suggestion for executing at least one computing power task including the number of allocated GPUs and different power usage periods specifically includes:
[0061] Obtaining the power resource usage and GPU usage of the computing center according to the computing power demand;
[0062] According to the power resource usage, a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period is selected as a time period for executing computing tasks;
[0063] Selecting a GPU for executing the computing task according to the GPU usage and a preset GPU priority principle;
[0064] Based on the time period selected as the time for executing the computing task and the GPU used to execute the computing task, a suggestion for executing the at least one computing task including the number of allocated GPUs and different power usage time periods is generated.
[0065] Specifically, the power resource usage and GPU usage of the computing center are obtained according to the computing power demand of the user. Based on the power resource usage, the time period corresponding to the green energy power supply mode, or the low-peak power consumption time period, or the low-cost power consumption time period is preferentially selected as the time period for executing the computing task. In order to be able to dynamically adjust according to the real-time status and historical data to ensure efficient, fair and reliable GPU resource allocation, the GPU for executing the computing task is selected according to the preset GPU priority principle based on the GPU usage. Among them, the preset GPU priority principles include: idle GPU priority principle, minimum usage time principle, minimum number of errors principle, load balancing principle, health status priority principle, task matching principle, user / task priority principle, etc. Based on the selected time period for executing the computing task and the GPU used to execute the computing task, a suggestion for executing the computing task is generated, including the number of allocated GPUs and at least one computing task execution in different power consumption time periods.
[0066] In an optional embodiment, before selecting a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period as a time period for executing a computing task according to the power resource usage, the method further includes:
[0067] The electricity usage time periods are divided according to the electricity prices corresponding to different power supply time periods of different power supply modes of the computing power center, wherein the power supply modes at least include green energy power supply modes.
[0068] Specifically, based on different power supply modes of the computing power center, the electricity prices corresponding to the power supply in different power supply time periods are divided into electricity time periods, where the power supply modes include at least green energy power supply modes, and the green energy power supply modes include at least solar power supply, wind power supply, and stored green electricity power supply. For example, when solar power is used, the electricity price (that is, the electricity price) fluctuations in different power supply time periods are divided into electricity time periods; when wind power is used, the electricity price fluctuations in different power supply time periods are divided into electricity time periods; when stored green electricity is used, the electricity price fluctuations in different power supply time periods are divided into electricity time periods; when thermal power is used, the electricity price fluctuations in different power supply time periods are divided into electricity time periods.
[0069] Step S103: receiving a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion.
[0070] Specifically, the user selects a suggestion for executing a target computing power task from at least one suggestion for executing a computing power task, and receives the suggestion for executing the target computing power task selected by the user.
[0071] Step S104: Execute the computing task according to the suggestion for executing the target computing task.
[0072] Specifically, the computing task corresponding to the computing power demand is executed according to the target number of GPUs and the target power consumption period in the recommendation for executing the target computing power task.
[0073] In an optional embodiment, after executing the computing task according to the suggestion for executing the target computing task, the method further includes:
[0074] Monitor the execution status of the computing task according to the number of GPUs and power consumption period corresponding to the computing task.
[0075] In this embodiment, after the user chooses to execute the computing task corresponding to the computing demand according to the suggestion of the target computing task execution, the execution of the computing task is automatically monitored according to the number of GPUs and the power consumption period corresponding to the computing task (that is, the target number of GPUs and the target power consumption period), and the GPU is automatically switched and the computing task is automatically started and ended during the power consumption period. The details are as follows:
[0076] (1) Power consumption time period monitoring:
[0077] Objective: Track whether the current time is in a preset low-cost or high-cost electricity period.
[0078] Monitoring indicators: The system uses clock services or synchronizes with the smart interface of the power grid to identify in real time which time period the system belongs to.
[0079] (2) GPU usage status monitoring:
[0080] Goal: Ensure GPU usage efficiency and health, and adjust usage strategies based on power consumption periods.
[0081] Monitoring indicators:
[0082] GPU Load: Monitor the current workload of each GPU (e.g. utilization, temperature, power consumption).
[0083] Task progress: Track the completion percentage of computing tasks to ensure that tasks proceed as planned.
[0084] Errors and failures: Record the number of GPU errors and failure status to avoid using problematic devices.
[0085] (3) Automatic switching and adjustment strategy:
[0086] GPU switching: When a GPU is detected to be overloaded, overheated, or close to a failure threshold, tasks are automatically assigned to other healthy GPUs with lower loads to ensure task continuity and hardware safety.
[0087] (4) Task Scheduling:
[0088] Start and end: Before the low-cost electricity period begins, the system automatically wakes up or starts the planned computing-intensive tasks; before the high-cost period arrives, if the task allows interruption, it automatically saves the status and pauses, or prioritizes tasks with a high degree of completion and nearing the end to accelerate completion.
[0089] (5) Energy efficiency monitoring:
[0090] Power consumption: Continuously monitor the power consumption of the entire system to ensure that energy consumption is reduced as much as possible while meeting computing power requirements.
[0091] Energy efficiency ratio: Evaluate the energy efficiency performance of different GPUs when performing specific tasks and optimize the configuration to improve overall energy efficiency.
[0092] In an optional embodiment, the method further includes:
[0093] The number of GPUs used during the execution of the computing task is dynamically increased or decreased according to the real-time changes in electricity rates.
[0094] In this embodiment, in order to optimize cost-effectiveness, the number of GPUs used during the execution of the computing task is dynamically increased or decreased according to real-time changes in electricity rates.
[0095] In an optional embodiment, the method further includes:
[0096] Obtain the target number of GPUs and target power consumption period in the recommendation for executing the target computing power task;
[0097] The computing power cost corresponding to the computing power task is obtained according to the target number of GPUs and the target power consumption period.
[0098] Specifically, the target number of GPUs and the target power consumption period in the recommendation for executing the target computing task are obtained, and the total power consumption of all GPUs corresponding to the computing task is calculated based on the target number of GPUs and the power of a single GPU; the total power consumption corresponding to the computing task is calculated based on the power consumption duration of the target power consumption period and the total power consumption of all GPUs; the computing cost (i.e., electricity cost) corresponding to the computing task is calculated based on the electricity price of the target power consumption period and the total power consumption corresponding to the computing task.
[0099] It is worth mentioning that the present invention uses intelligent algorithms to optimize computing power allocation and task scheduling to ensure that computing power is maximized during periods of lowest electricity costs (such as peak solar energy during the day), thereby solving the pressure of green electricity consumption and turning "watts" into "bits"; and reducing non-emergency computing power demand during periods of higher electricity costs to minimize electricity costs. The present invention incorporates data bases into the peak and valley time-of-use electricity price policy and flexibly adjusts electricity consumption periods to improve the linkage mechanism between energy costs and electricity prices, guides data bases to consume more green electricity, and promotes the reduction of electricity costs in the green computing power industry. The present invention uses the highly adaptable and intelligent decision-making of computing and electricity collaboration between green computing power bases to better guide users to shave peaks and fill valleys, improve electricity supply and demand, and promote the consumption of new energy. The proposal of this invention is of great significance for building a rationally laid out, green and low-carbon computing power infrastructure and continuously expanding the scale of the green computing power industry. It can help companies significantly reduce operating costs while promoting environmental and ecological protection.
[0100] In a specific embodiment, a computing power scheduling method is provided and applied to a computing power scheduling platform of a computing power center. The platform is used to automatically generate recommendations for allocating GPUs and power usage time periods based on computing power requirements provided by users, and automatically execute computing power tasks based on the recommendations selected by users, thereby simplifying and automating the computing power scheduling process and improving the computing power scheduling efficiency.
[0101] The computing power scheduling platform includes a power supply module, a computing power equipment monitoring module, a computing power allocation module, etc. After obtaining the computing power demand input by the user, the computing power allocation module gives at least one computing power task execution suggestion based on the power resource usage provided by the power supply module and the GPU usage provided by the computing power equipment monitoring module.
[0102] Among them, the computing power allocation module gives priority to the time period corresponding to the green energy power supply mode, or the low-peak power consumption time period, or the low-cost power consumption time period as the time period for executing computing power tasks based on the power resource usage provided by the power supply module, and combines the GPU usage provided by the computing power equipment monitoring module to give priority to the number of idle GPUs, the cumulative GPU usage time, the cumulative number of GPU errors, etc. When allocating GPU resources to perform computing tasks, priority principles are formulated based on a variety of factors to ensure efficiency, stability and maximum utilization of resources: idle GPU priority principle, minimum usage time principle, minimum number of errors principle, load balancing principle, health status priority principle, task matching principle, user / task priority principle, etc., which can be dynamically adjusted according to real-time status and historical data to ensure efficient, fair and reliable GPU resource allocation.
[0103] For example, if the user's computing power requirement is no less than 10P and the rental time is 24h, the suggestions for computing power task execution under the same computing power supply are given (assuming that 8 GPUs can provide 10P computing power) as shown in Table 1:
[0104] Table 1: Suggestions for computing task execution
[0105]
[0106] After the user chooses to execute a computing task according to a certain suggestion, the execution of the computing task is automatically monitored according to the power consumption time period corresponding to the computing task and the number of GPUs used in different time periods, and the GPU is automatically switched and the computing task is automatically started and ended during the power consumption period. The details are as follows:
[0107] (1) Power consumption time period monitoring:
[0108] Objective: Track whether the current time is in a preset low-cost electricity period (such as the night off-peak period) or a high-cost period.
[0109] Monitoring indicators: The system uses clock services or synchronizes with the smart interface of the power grid to identify in real time which time period the system belongs to.
[0110] (2) GPU usage status monitoring:
[0111] Goal: Ensure GPU usage efficiency and health, and adjust usage strategies based on power consumption periods.
[0112] Monitoring indicators:
[0113] GPU Load: Monitor the current workload of each GPU (e.g. utilization, temperature, power consumption).
[0114] Task progress: Track the completion percentage of computing tasks to ensure that tasks proceed as planned.
[0115] Errors and failures: Record the number of GPU errors and failure status to avoid using problematic devices.
[0116] (3) Automatic switching and adjustment strategy:
[0117] GPU switching: When a GPU is detected to be overloaded, overheated, or close to a failure threshold, tasks are automatically assigned to other healthy GPUs with lower loads to ensure task continuity and hardware safety.
[0118] (4) Task Scheduling:
[0119] Start and end: Before the low-cost electricity period begins, the system automatically wakes up or starts the planned computing-intensive tasks; before the high-cost period arrives, if the task allows interruption, it automatically saves the status and pauses, or prioritizes tasks with a high degree of completion and nearing the end to accelerate completion.
[0120] (5) Energy efficiency monitoring:
[0121] Power consumption: Continuously monitor the power consumption of the entire system to ensure that energy consumption is reduced as much as possible while meeting computing power requirements.
[0122] Energy efficiency ratio: Evaluate the energy efficiency performance of different GPUs when performing specific tasks and optimize the configuration to improve overall energy efficiency.
[0123] In addition, the present invention can also dynamically adjust the GPU: according to the real-time changes in electricity rates (if the dynamic pricing information of the smart grid is connected), the number of GPUs used can be dynamically increased or decreased to optimize cost-effectiveness.
[0124] Through these detailed monitoring and automated adjustment strategies, the system can intelligently manage computing resources, which not only improves computing efficiency but also effectively controls energy costs, achieving a win-win situation in economic and environmental benefits.
[0125] Among them, the power supply module provides the changes in electricity prices under the mainstream power supply mode in the region. For example, a certain area can be fully powered by solar energy during the day, so the electricity price fluctuations in the region are generated according to the electricity price of solar power supply during the day and according to the price of thermal power supply at night.
[0126] In addition, the interface corresponding to the power supply module can also display the corresponding electricity prices in different time periods under different power supply methods in the region. For example, the electricity price fluctuations in different time periods when solar power is used; the electricity price fluctuations in different time periods when wind power is used; the electricity price fluctuations in different time periods when stored green electricity is used; and the electricity price fluctuations in different time periods when thermal power is used.
[0127] Among them, the amount of electricity that can be supplied in different time periods under the above different power supply modes and the corresponding electricity prices of different power supply modes will change with the weather. For example, on cloudy days and sunny days, the amount of electricity that can be supplied by solar power supply and the fluctuation of electricity prices will be different. For example, the corresponding electricity prices of different time periods under different power supply modes provided by the power supply module are shown in Table 2:
[0128] Table 2: Electricity prices corresponding to different time periods under different power supply modes
[0129]
[0130] The computing power equipment monitoring module can provide the idle status of each GPU in different time periods. For example, the idle status of each GPU in different time periods provided by the computing power equipment monitoring module is shown in Table 3:
[0131] Table 3: Idle status of each GPU in different time periods
[0132]
[0133] Therefore, the user can also design the computing task execution plan based on the power supply time and electricity price of different power supply modes provided by the power supply module and the GPU situation provided by the computing device monitoring module. For example, the computing task execution plan designed by the user based on the above information is shown in Table 4:
[0134] Table 4: Computing task execution plan
[0135]
[0136] After determining the execution plan designed by the user, it automatically controls the startup of the computing task, monitors the execution process, automatically switches the GPU, and automatically switches the power supply mode.
[0137] In addition, the computing power allocation module can also provide a calculation model based on the following calculation principle. Users can input information such as the number of GPUs required to perform a certain computing task, power time period, power supply method, etc. into the computing power allocation module, and then the computing power allocation module directly outputs the computing power cost of performing the computing task. The calculation model is calculated based on the following calculation principle:
[0138] 1. Basic parameter definition:
[0139] 1. Number of GPUs (N): The number of GPUs in use.
[0140] 2. Single GPU power (P, unit: Watt): The average power consumption of each GPU when running.
[0141] 3. Power usage time (T, unit: hours): the continuous usage time within the calculation period.
[0142] 4. Electricity price (E, unit: yuan / kWh): The price varies according to the time period of electricity consumption.
[0143] 2. Algorithm steps:
[0144] 1. Total power consumption (W_total): Total power consumption of all GPUs = N*P.
[0145] 2. Total power consumption (KWh_total): Total power consumption during the period = W_total*T / 1000 (converted to kilowatt-hours).
[0146] 3. Electricity cost (C_electricity): Electricity cost during the period = KWh_total*E.
[0147] 3. Example calculation:
[0148] 1. Assumptions:
[0149] - Number of GPUs N = 4.
[0150] - Single GPU power P = 300 watts.
[0151] - Electricity usage time T = 24 hours.
[0152] -Electricity price (peak period) E = 0.5 yuan / kWh.
[0153] 2. Calculation process:
[0154] Total power consumption = 4*300 = 1200 watts = 1.2 kilowatts.
[0155] Total electricity consumption = 1.2 kW*24 hours = 28.8 kWh.
[0156] Electricity cost = 28.8 kWh * 0.5 yuan / kWh = 14.4 yuan.
[0157] For example, taking the peak-valley time-of-use electricity price of a certain province and the mainstream GPU chip A100 as an example, the computing power costs corresponding to the computing power tasks recommended by several computing power scheduling methods given based on the above theory are shown in Table 5:
[0158] Table 5: Computing cost corresponding to computing tasks
[0159]
[0160] The computing power scheduling method provided by the embodiment of the present invention first obtains the computing power demand of the user; then generates suggestions including the number of allocated graphics processors GPU and at least one computing power task execution in different power consumption periods according to the computing power demand, wherein the power consumption period is divided according to the electricity price at different power supply times, and the at least one computing power task execution suggestion includes a suggestion to ensure the maximum computing power usage in the period with the lowest electricity cost; then receives the target computing power task execution suggestion selected by the user from the at least one computing power task execution suggestion; finally, executes the computing power task according to the target computing power task execution suggestion. The present invention optimizes computing power allocation and task scheduling, automatically generates at least one suggestion for allocating GPUs and power consumption periods to ensure the maximum computing power usage in the period with the lowest electricity cost according to the computing power demand of the user, and automatically executes computing power tasks according to the suggestions selected by the user, thereby reducing non-emergency computing power demand in the period with higher electricity cost, minimizing electricity cost, and solving the problem that the rapid growth of computing power demand in the prior art leads to huge electricity consumption, resulting in high electricity cost.
[0161] Embodiment 2:
[0162] like Figure 2 As shown, this embodiment provides a computing power scheduling device, which is used to execute the above computing power scheduling method, including:
[0163] A first acquisition module 11 is used to obtain the computing power requirements of the user;
[0164] A generating module 12, connected to the first acquiring module 11, for generating, according to the computing power demand, suggestions for executing at least one computing power task including the number of allocated graphics processing units (GPUs) and different power usage periods, wherein the power usage periods are divided according to electricity prices at different power supply times, and the at least one computing power task execution suggestion includes a suggestion for ensuring maximum computing power usage during the period with the lowest electricity cost;
[0165] A receiving module 13, connected to the generating module 12, for receiving a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion;
[0166] The execution module 14 is connected to the receiving module 13 and is used to execute the computing task according to the suggestion for executing the target computing task.
[0167] Furthermore, the generating module 12 specifically includes:
[0168] An acquisition unit, used to acquire the power resource usage and GPU usage of the computing center according to the computing power demand;
[0169] A first selection unit is used to select a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period as a time period for executing a computing task according to the power resource usage;
[0170] A second selection unit is used to select a GPU for executing a computing task according to a preset GPU priority principle based on the GPU usage;
[0171] A generating unit is used to generate a suggestion for executing the at least one computing task in different power usage periods, including the number of allocated GPUs and the execution of the at least one computing task according to the selected time period for executing the computing task and the GPU used to execute the computing task.
[0172] Furthermore, the generating module 12 further includes:
[0173] A division unit is used to divide the electricity consumption time periods according to the electricity prices corresponding to different power supply time periods of different power supply modes of the computing power center, wherein the power supply mode at least includes a green energy power supply mode.
[0174] Furthermore, the device also includes:
[0175] The monitoring module is used to monitor the execution status of the computing task according to the number of GPUs and power consumption period corresponding to the computing task.
[0176] Furthermore, the device also includes:
[0177] The dynamic increase and decrease module is used to dynamically increase and decrease the number of GPUs used during the execution of the computing task according to real-time changes in electricity rates.
[0178] Furthermore, the device also includes:
[0179] A second acquisition module is used to obtain the target number of GPUs and the target power consumption period in the suggestion for executing the target computing power task;
[0180] The obtaining module is used to obtain the computing power cost corresponding to the computing power task according to the target number of GPUs and the target power consumption period.
[0181] Furthermore, the obtaining module specifically includes:
[0182] A first calculation unit, configured to calculate the total power consumption of all GPUs corresponding to the computing task according to the target number of GPUs and the power of a single GPU;
[0183] A second calculation unit, configured to calculate a total power consumption corresponding to the computing task according to the power consumption duration of the target power consumption period and the total power consumption of all the GPUs;
[0184] The third calculation unit is used to calculate the computing power cost corresponding to the computing power task according to the electricity price of the target electricity consumption period and the total electricity consumption corresponding to the computing power task.
[0185] Embodiment 3:
[0186] refer to Figure 3 This embodiment provides a computing power scheduling device, including a memory 21 and a processor 22, wherein the memory 21 stores a computer program, and the processor 22 is configured to run the computer program to execute the computing power scheduling method in Example 1.
[0187] The memory 21 is connected to the processor 22. The memory 21 may be a flash memory, a read-only memory or other memory. The processor 22 may be a central processing unit or a single-chip microcomputer.
[0188] Embodiment 4:
[0189] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the computing power scheduling method in the above-mentioned embodiment 1 is implemented.
[0190] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.
[0191] In summary, the computing power scheduling method, device and computer-readable storage medium provided by the embodiments of the present invention first obtain the computing power demand of the user; then generate suggestions including the number of allocated graphics processors GPU and at least one computing power task execution in different power consumption periods according to the computing power demand, wherein the power consumption period is divided according to the electricity price of different power supply times, and the at least one computing power task execution suggestion includes a suggestion to ensure the maximum computing power usage in the period with the lowest electricity cost; then receive the target computing power task execution suggestion selected by the user from the at least one computing power task execution suggestion; finally, execute the computing power task according to the target computing power task execution suggestion. The present invention optimizes computing power allocation and task scheduling, automatically generates at least one suggestion for allocating GPUs and power consumption periods to ensure the maximum computing power usage in the period with the lowest electricity cost according to the computing power demand of the user, and automatically executes computing power tasks according to the suggestions selected by the user, thereby reducing non-emergency computing power demand in the period with higher electricity cost, minimizing electricity cost, and solving the problem that the rapid growth of computing power demand in the prior art leads to huge electricity consumption, resulting in high electricity cost.
[0192] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.
Claims
1. A computing power scheduling method, characterized in that: The method comprises: Obtain the user's computing power requirements; Generate, according to the computing power demand, a suggestion including the number of allocated graphics processing units (GPUs) and at least one computing power task execution suggestion for different power usage periods, wherein the power usage periods are divided according to electricity prices at different power supply times, and the at least one computing power task execution suggestion includes a suggestion for ensuring maximum computing power usage during the period with the lowest electricity cost; Receive a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion; Execute the computing task according to the suggestion for executing the target computing task.
2. The method according to claim 1, characterized in that: Generating, according to the computing power demand, a suggestion for executing at least one computing power task including the number of allocated GPUs and different power usage periods specifically includes: Obtaining the power resource usage and GPU usage of the computing center according to the computing power demand; According to the power resource usage, a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period is selected as a time period for executing computing tasks; Selecting a GPU for executing the computing task according to the GPU usage and a preset GPU priority principle; Based on the time period selected as the time for executing the computing task and the GPU used to execute the computing task, a suggestion for executing the at least one computing task including the number of allocated GPUs and different power usage time periods is generated.
3. The method according to claim 2, characterized in that Before selecting a time period corresponding to a green energy power supply mode, a low-peak power consumption time period, or a low-cost power consumption time period as a time period for executing a computing task according to the power resource usage, the method further includes: The electricity usage time periods are divided according to the electricity prices corresponding to different power supply time periods of different power supply modes of the computing power center, wherein the power supply modes at least include green energy power supply modes.
4. The method according to claim 1, characterized in that: After executing the computing task according to the suggestion for executing the target computing task, the method further includes: Monitor the execution status of the computing task according to the number of GPUs and power consumption period corresponding to the computing task.
5. The method according to claim 4, characterized in that The method further comprises: The number of GPUs used during the execution of the computing task is dynamically increased or decreased according to the real-time changes in electricity rates.
6. The method according to claim 1, characterized in that The method further comprises: Obtain the target number of GPUs and target power consumption period in the recommendation for executing the target computing power task; The computing power cost corresponding to the computing power task is obtained according to the target number of GPUs and the target power consumption period.
7. The method according to claim 6, characterized in that The step of obtaining the computing power cost corresponding to the computing power task according to the target number of GPUs and the target power consumption period specifically includes: Calculate the total power consumption of all GPUs corresponding to the computing task according to the target number of GPUs and the power of a single GPU; Calculate the total power consumption corresponding to the computing task according to the power consumption duration of the target power consumption period and the total power consumption of all GPUs; The computing power cost corresponding to the computing power task is calculated according to the electricity price during the target electricity consumption period and the total electricity consumption corresponding to the computing power task.
8. A computing power scheduling device, characterized in that: include: The first acquisition module is used to obtain the computing power requirements of the user; a generating module connected to the first acquiring module, configured to generate, according to the computing power demand, suggestions for executing at least one computing power task including the number of allocated graphics processing units (GPUs) and different power usage periods, wherein the power usage periods are divided according to electricity prices at different power supply times, and the at least one computing power task execution suggestion includes a suggestion for ensuring that computing power usage is maximized during the period with the lowest electricity cost; A receiving module, connected to the generating module, configured to receive a target computing task execution suggestion selected by a user from the at least one computing task execution suggestion; An execution module is connected to the receiving module and is used to execute the computing task according to the suggestion for executing the target computing task.
9. A computing power scheduling device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the computing power scheduling method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computing power scheduling method according to any one of claims 1 to 7 is implemented.
Citation Information
Cited By
Distribution method and system for improving GPU (Graphics Processing Unit) resource utilization rate, medium and program product
CN119415272A
Allocation method, system, medium and program product for improving GPU resource utilization
CN119415272B
Calculation power peak shifting scheduling method and device for intelligent calculation center of common calculation power
CN119883575A
Optimized scheduling method and device based on resource collaboration and electronic equipment
CN120430463A
