Three-phase power load balancing method and system based on HPC adaptive scheduling strategy

Abstract

The invention discloses a three-phase power load balancing method and system based on an HPC adaptive scheduling strategy. The method comprises the following steps: acquiring computational node information and three-phase power load information when a data center is idle; designing a test job and submitting the test job to a data center for operation, and obtaining three-phase power load information during operation; extracting a three-phase power load variable quantity from the three-phase power load information during the operation period, and establishing a mapping relation between a calculation node and a three-phase power load based on the three-phase power load variable quantity; based on the mapping relation, obtaining an imbalance degree of calculation node distribution, and based on the imbalance degree, evaluating whether the mapping relation is reasonable; and if the mapping relation is reasonable, dynamically distributing the computing nodes based on the mapping relation, and realizing three-phase power load balancing during daily operation. And adaptively adjusting an HPC job scheduling strategy based on the mapping relation between the computing node and the three-phase power supply to realize three-phase power load balancing.

Description

Technical Field

[0001] The present invention relates to the field of intelligent power distribution technology, and in particular to a three-phase power load balancing method and system based on an HPC adaptive scheduling strategy. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] In data centers, three-phase power load balancing is crucial for improving the efficiency of uninterruptible power supplies (UPS), reducing power consumption, and extending the service life of UPSs. Information technology (IT) equipment is usually powered by several phases of a three-phase power supply. Traditional three-phase power load balancing methods rely heavily on hardware devices and are difficult to adapt to the dynamic changes in the workload of high-performance computing centers. Therefore, three-phase power load imbalance often occurs. The main reasons are the following factors: (1) Unbalanced load distribution: Many data centers will plan the layout of terminal loads in the early stages of construction. However, with the requirements of project implementation schedules and the influence of factors such as construction technology in IT cabinets, the three-phase balance of the data center power supply and distribution system is often reduced. (2) Dynamic load changes: In data centers, multiple computing nodes generally execute computing tasks at the same time. However, because the power phase lines of the computing nodes are not matched with the task running servers, the current of one phase of the three-phase power supply system may suddenly increase during task operation, thereby affecting the output quality of the power supply voltage and causing unbalanced workload of the power supply equipment.

[0004] In existing HPC job scheduling technology, the scheduling strategy is generally based on computing node number, performance load, network structure, etc., and the factor of three-phase power load balancing has not been considered. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a three-phase power load balancing method and system based on an HPC adaptive scheduling strategy, which adaptively adjusts the HPC job scheduling strategy based on the mapping relationship between computing nodes and three-phase power supplies to achieve three-phase power load balancing.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a three-phase power load balancing method based on an HPC adaptive scheduling strategy, comprising:

[0008] Obtain computing node information and three-phase power load information when the data center is idle;

[0009] Design test jobs and submit them to the data center for execution, obtaining three-phase power load information during operation;

[0010] Extracting a three-phase power load variation from the three-phase power load information during the operation, and establishing a mapping relationship between the computing node and the three-phase power load based on the three-phase power load variation;

[0011] Obtaining an imbalance degree of computing node allocation based on the mapping relationship, and evaluating whether the mapping relationship is reasonable based on the imbalance degree;

[0012] If the mapping relationship is reasonable, the computing nodes are dynamically allocated based on the mapping relationship to achieve three-phase power load balancing during daily operation.

[0013] A further technical solution is to extract the three-phase power load change as follows:

[0014] Compute job J on computing node N j The change of three-phase power load during operation;

[0015] Determine the computing node N j The set of phases that causes the load to increase among the three phases.

[0016] A further technical solution is to establish a mapping relationship between computing nodes and three-phase power loads as follows:

[0017] Obtaining a total three-phase power load variation based on the three-phase power load variation of each computing node;

[0018] Cleaning the total three-phase power load change based on a load change threshold to obtain cleaned data;

[0019] A mapping relationship between computing nodes and phases is established based on the cleaned data.

[0020] In a further technical solution, the mapping relationship is expressed as:

[0021] P j ={a,b,c|ΔLj k >0,k=a,b,c}

[0022] Among them, P j Indicates computing node N j The phase set that causes the load to increase in phases A, B, or C. a, b, and c represent phases A, B, and C respectively. ΔLj k Indicates computing node N j The resulting change in the power load of phase k.

[0023] A further technical solution is to evaluate whether the mapping relationship is reasonable based on the imbalance degree, specifically:

[0024] The imbalance degree of computing node allocation is calculated based on the mapping relationship;

[0025] Comparing and evaluating the imbalance degree with a reasonable threshold;

[0026] When the imbalance is greater than a reasonable threshold, the number of computing nodes assigned to the three phases is adjusted and a new imbalance is calculated. When the imbalance is less than or equal to a reasonable threshold, the mapping relationship is evaluated to be reasonable.

[0027] A further technical solution is to dynamically allocate computing nodes based on the mapping relationship if the mapping relationship is reasonable, specifically as follows:

[0028] The new job J n Submit to the data center;

[0029] Count the set of all computing nodes running jobs at the current moment and the actual three-phase power load;

[0030] Using the greedy algorithm, we can n Conduct resource selection;

[0031] Adopting adaptive scheduling strategy for job J n The operation is allocated to the computing nodes.

[0032] A further technical solution is to use a greedy algorithm to n The specific steps for resource selection are:

[0033] Get the set of idle computing nodes based on the set of all computing nodes running jobs;

[0034] If the idle computing node set is empty or contains fewer computing nodes than job J n The required number of computing nodes is determined by waiting for other jobs to complete and then recalculating the set of all computing nodes currently running jobs and the actual three-phase power load.

[0035] If the number of computing nodes contained in the idle computing node set is greater than or equal to job J n The number of computing nodes required is used to determine the phase k with the minimum current load. j , select the node with phase k from the set of idle computing nodes j The compute node with the smallest number for which a mapping relationship exists

[0036] Combine the idle compute node set with the compute node Take the difference set to get a new set of idle computing nodes and recalculate the actual three-phase power load;

[0037] Repeat the above two steps until the job Jn Allocate all required compute nodes.

[0038] In a second aspect, the present invention provides a three-phase power load balancing system based on an HPC adaptive scheduling strategy, comprising:

[0039] A data acquisition module is configured to: acquire computing node information and three-phase power load information when the data center is idle;

[0040] An operation data acquisition module is configured to: design a test job and submit it to the data center for operation, and obtain three-phase power load information during the operation;

[0041] a mapping relationship building module, configured to: extract a three-phase power load variation from the three-phase power load information during the operation, and establish a mapping relationship between the computing node and the three-phase power load based on the three-phase power load variation;

[0042] a mapping relationship evaluation module configured to: obtain an imbalance degree of computing node allocation based on the mapping relationship, and evaluate whether the mapping relationship is reasonable based on the imbalance degree;

[0043] The node dynamic allocation module is configured to dynamically allocate computing nodes based on the mapping relationship if the mapping relationship is reasonable, so as to achieve three-phase power load balancing during daily operation.

[0044] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the three-phase power load balancing method based on the HPC adaptive scheduling strategy as described in the first aspect.

[0045] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the three-phase power load balancing method based on the HPC adaptive scheduling strategy as described in the first aspect are implemented.

[0046] One or more of the above technical solutions have the following beneficial effects:

[0047] This invention adaptively adjusts the HPC job scheduling strategy based on the mapping relationship between computing nodes and three-phase power sources to achieve three-phase power load balancing. Specifically, the three-phase power load change is obtained by using the three-phase power load information before and after the data center runs a job. Based on this three-phase power load change, a mapping relationship between computing nodes and three-phase power load is established. The rationality of the resulting mapping relationship is evaluated, and the HPC job scheduling strategy is adaptively adjusted based on the rational mapping relationship.

[0048] The present invention can adapt to changes in the data center's workload in real time by adaptively adjusting the job scheduling strategy. With minimal changes to the power plug, it can achieve dynamic balancing of the three-phase power load, thereby achieving the following benefits: Improved UPS efficiency and reduced energy consumption: Reducing additional energy consumption such as reactive power loss and line loss caused by load imbalance significantly improves UPS operating efficiency, directly reduces data center energy consumption, and saves electricity bills. Extended UPS life and improved power supply security: Three-phase power load balancing can reduce UPS overload and unbalanced operating time, extend the UPS's service life, and reduce the risk of high temperature and insulation corrosion on cables and contact points, thereby improving the security of data center power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0050] Figure 1 It is a flow chart of a three-phase power load balancing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0054] HPC job scheduling systems are key components for managing job execution in high-performance computing (HPC) clusters. They are responsible for allocating computing resources, monitoring job status, optimizing resource utilization, and ensuring that jobs are executed on demand.

[0055] Job scheduling strategy is a method of allocating computing node resources to HPC jobs.

[0056] Scheduling HPC jobs with three-phase power load balancing requires access to power distribution system monitoring data, which goes beyond the scope of traditional HPC job scheduling systems, which typically only consider compute node resources. Considering three-phase power load balancing in HPC job scheduling requires first obtaining power distribution monitoring data for the cabinets housing the compute nodes, splitting it into three phases, A, B, and C, and then performing joint calculations in a timely manner according to HPC job scheduling requirements. This process is difficult to implement with existing technologies.

[0057] Example 1

[0058] like Figure 1 As shown, this embodiment discloses a three-phase power load balancing method based on the HPC adaptive scheduling strategy, which includes the following steps:

[0059] To simplify the description, this embodiment assumes an IT cabinet filled with compute nodes with the same hardware configuration. The physical environment within the cabinet is balanced, the compute nodes use multiple single-phase power modules, all compute nodes have the same network link hop count and performance, all compute nodes are assigned to the same HPC job queue, and the running scale of each job is based on the compute node.

[0060] S1: Obtain computing node information and three-phase power load information when the data center is idle;

[0061] In this embodiment, it is defined that N={N1, N2, ..., N p} is a node set, where p is the number of nodes. Define L0 = {L0 a ,L0 b ,L0 c} is the three-phase power load value when all computing nodes are idle, the unit is generally kW, L0 a Indicates the load value of phase A when all computing nodes are idle, L0 b Indicates the load value of phase B when all computing nodes are idle, L0 c Indicates the load value of phase C when all computing nodes are idle.

[0062] S101: Extract all computing nodes in the cabinet through the HPC job scheduling system to form a node set N.

[0063] S102: Extracting the three-phase power load information L0 of the cabinet when the computing node is idle through the power distribution monitoring system.

[0064] S2: Design a test job and submit it to the data center for execution, obtaining three-phase power load information during operation;

[0065] S201: Design a computationally intensive job J, such as LINPACK, to test the performance of a high-performance computer system using the LINPACK linear systems software package. The run should utilize all cores of a computing node and last at least T minutes (typically T ≥ 3).

[0066] S202: For each computing node N in the node set N j , submit job J respectively, and record the three-phase power load information L through the power distribution monitoring system after the job starts running j ={Lj a ,Lj b ,Lj c}, and then switch to the next node after the job is completed.

[0067] S3: extracting a three-phase power load change from the three-phase power load information during the operation;

[0068] S301: Computing job J on computing node N j Three-phase power load change ΔL during operation j ΔN j ={ΔNj a , ΔNj b ,...,ΔNj c}, indicating that job J is running on computing node N j The change in three-phase power load during operation, where ΔLj k =Lj k -L0 k , k=a or k=b or k=c.

[0069] S302: Determine computing node N j The phase set p that causes the load increase in phases A, B, or C j ={a,b,c|ΔLj k >0,k=a,b,c}, where P j Indicates computing node N j The phase set that causes the load to increase in phases A, B, or C. a, b, and c represent phases A, B, and C respectively. ΔLj k Indicates computing node N j The resulting change in the power load of phase k.

[0070] S4: Based on the three-phase power load change of each computing node, a mapping relationship between the computing node and the three-phase power load is established;

[0071] S401: Select a computing node set N={N1, N2, ..., N p}.

[0072] Calculate the three-phase power load change of each computing node and obtain the total three-phase power load change ΔL={ΔLj a , ΔLj b , ΔLj c , j = 1, 2, ..., p}.

[0073] S402: Clean the total three-phase power load change ΔL to eliminate the load increase caused by external factors such as instrument measurement errors. The phases with insignificant impact are set to 0, and the cleaned data is still recorded as L. The load change threshold θ is set in advance, and θ can be 5%*max{L0 k ,k=a,b,c}. Such as ΔLj k <θ, then it is considered that the computing node N j The effect on phase k is not significant, so it is set to 0, i.e. ΔLj k =0.

[0074] S403: Based on the cleaned data L, re-establish the corresponding relationship between the calculation node and the phase, that is, the mapping relationship P according to S302 j The mapping relationship is a mapping relationship between the node number and the three-phase power phase where the computing node is located.

[0075] S5: Based on the mapping relationship, obtain the imbalance degree of computing node allocation, and evaluate whether the mapping relationship is reasonable based on the imbalance degree. Specifically, based on the mapping relationship, obtain the imbalance degree of computing node allocation according to the following steps, and evaluate whether the three-phase power load is balanced when all computing nodes under the current mapping run HPC jobs simultaneously based on the imbalance degree.

[0076] S501: Obtain a mapping relationship set B based on the computing nodes and the corresponding relationship, B=(N j ,P j ,j=1,2,...,p), and determine the computing node N according to this set j With which phase P j There is an influence relationship.

[0077] S502: Based on the mapping relationship set B, calculate the imbalance degree of the three-phase quantity allocated to all computing nodes in the IT cabinet, which is expressed as:

[0078]

[0079] Among them, X a is the number of computing nodes plugged into phase A, X b is the number of computing nodes plugged into phase B, X c is the number of computing nodes plugged into phase C, i.e. X k =|{N j |k∈Pj , j = 1, 2, ..., p}|, where p represents the number of all computing nodes. The smaller the imbalance degree, i.e., the value of Y, the better.

[0080] S503: When the imbalance Y is greater than the threshold γ, where γ represents a reasonable threshold, it is considered that the three-phase power load is unbalanced and the following steps need to be repeated for adjustment:

[0081] (1) Assuming that phase k has the most nodes and phase j has the least nodes, replace the power module of a computing node in phase k with that in phase j, and then calculate the new imbalance Y′;

[0082] (2) If Y′≤γ, the mapping relationship is reasonable, the three-phase power load is balanced, and the adjustment is completed.

[0083] (3) Otherwise, execute step (1).

[0084] S6: If the mapping relationship is reasonable, the computing nodes are dynamically allocated based on the mapping relationship to achieve three-phase power load balancing during daily operation;

[0085] This step enables continuous detection and dynamic allocation of computing nodes during daily operations of the data center.

[0086] S601: Submit the n=1th job J n Before the job scheduling system allocates the cabinet, it is in an idle state. Based on the above steps, the three-phase power load is balanced at this time.

[0087] S602: Count the current time t n All the files used in running jobs include d n A set of N nodes n and the actual three-phase power load L n .remember L n ={m n，a , m n，b , m n，c}.

[0088] in, The value 1 indicates node N p,i Mapping in phase k (ie k∈P p,i ), a value of 0 indicates that it is not mapped in phase k; L i,k Represents node N p,i The electrical load distributed on phase k.

[0089] S603: Using a greedy algorithm, select resources for the submitted nth job:

[0090] Homework J n The number of computing nodes required is cn .

[0091] (1) Get the set of idle computing nodes based on the set of all computing nodes running jobs. If N\N n (i.e. the remaining idle computing nodes, i.e. N to N n The difference of ) is empty or contains less than c computing nodes n , then wait for a period of time Δt (i.e., release some nodes after other jobs are completed), and then go to step 602 for execution; otherwise, go to the next step (2) for execution.

[0092] (2) Take j=1.

[0093] (3) Determine the phase k with the minimum current load j , as follows:

[0094]

[0095] Among them, N max Indicates the maximum number of nodes that can be allocated in each phase, m n,k Indicates the number of nodes used in the current phase k.

[0096] (4) From N\N n Select the phase k j The computational node with the smallest number that has a mapping is denoted as node Right now

[0097] (5) Get a new one And recalculate the new corresponding three-phase power load L according to step 602 n .

[0098] (6) Take j = j + 1, if j ≤ c n , then go to step (3) for execution; otherwise go to step 604 for execution.

[0099] S604: For newly submitted job J n Run the allocated compute nodes Thus, an adaptive scheduling strategy is implemented. At this time, the number of computing nodes used to run the job is d n =d n +c n .

[0100] S605: For the newly submitted n=n+1th job, execute steps 602-604 in a loop until there are no more newly submitted jobs.

[0101] Through the above technical solution, compared with the traditional HPC job scheduling strategy of directly allocating jobs to the computing node with the smallest number, the dynamic allocation of this application adopts an adaptive adjustment job scheduling strategy. First, the phase with the smallest load is determined by step (3), and then the computing node with the smallest number that has a mapping relationship with this phase is selected. As the number of computing nodes put into operation increases, the phase with the smallest load changes dynamically.

[0102] S7: When the hardware configuration and network connection of the computing nodes in the data center change, step S1 is repeated, and the mapping relationship is updated in time, and then the dynamic allocation of S6 is performed.

[0103] The hazards caused by the unbalanced three-phase load in the data center power supply and distribution system are generally as follows:

[0104] (1) Increased energy consumption. Unbalanced three-phase power supply can lead to abnormal three-phase voltage fluctuations, which in turn increases power line losses. Currently, IT power consumption in data centers is often recorded in local control boxes. Therefore, an increase in line losses is equivalent to an increase in supporting energy consumption, which directly affects the power usage efficiency (PUE) of the data center.

[0105] (2) Equipment damage. Severe three-phase unbalanced power supply has a serious impact on the service life of dry-type transformers frequently used in data centers. The collection of forced air cooling data for dry-type transformers is based on the average temperature of the three-phase windings. In order to achieve higher energy efficiency standards, some transformers often set the gap between the transformer's forced air cooling start temperature and the melting point of the insulating varnish. When the transformer's three-phase average temperature triggers forced air cooling, the temperature of the phase that causes the load imbalance is often very close to the melting point of the insulating varnish. If the insulation is damaged, it will cause a short circuit, directly leading to transformer damage.

[0106] (3) Temperature changes on the high-voltage incoming line side. Three-phase imbalance will cause the temperature of the high-voltage incoming line side with large current to be higher than that on the two sides with small current, resulting in temperature imbalance in the incoming line box and rust on the metal contact surface. The generation of rust will bring greater hidden fault risks. This hidden danger is relatively hidden, with only a slight oxide layer on the incoming line contact surface. It is difficult to detect during maintenance and is easily overlooked, which in turn causes abnormal tripping of the low-voltage side equipment for unknown reasons, affecting the power supply security of the IT load at the end of the data center.

[0107] The present invention provides a three-phase power load balancing method based on an HPC adaptive scheduling strategy, aiming to improve the three-phase power load balancing degree, ensure stable and reliable operation of equipment, and enhance power supply and distribution efficiency. By allocating computing node resources used by jobs through the mapping relationship between computing nodes and three-phase power, dynamic balancing of the three-phase power load is achieved, effectively alleviating the above-mentioned hazards, thereby improving UPS efficiency, reducing energy consumption, and ultimately enhancing the power supply security of the data center.

[0108] Example 2

[0109] This embodiment discloses a three-phase power load balancing system based on an HPC adaptive scheduling strategy, including:

[0110] A data acquisition module is configured to: acquire computing node information and three-phase power load information when the data center is idle;

[0111] An operation data acquisition module is configured to: design a test job and submit it to the data center for operation, and obtain three-phase power load information during the operation;

[0112] a mapping relationship building module, configured to: extract a three-phase power load variation from the three-phase power load information during the operation, and establish a mapping relationship between the computing node and the three-phase power load based on the three-phase power load variation;

[0113] a mapping relationship evaluation module configured to: obtain an imbalance degree of computing node allocation based on the mapping relationship, and evaluate whether the mapping relationship is reasonable based on the imbalance degree;

[0114] The node dynamic allocation module is configured to dynamically allocate computing nodes based on the mapping relationship if the mapping relationship is reasonable, so as to achieve three-phase power load balancing during daily operation.

[0115] Example 3

[0116] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0117] Example 4

[0118] The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, which performs the steps of the method of embodiment 1 when executed by a processor.

[0119] The steps involved in the apparatuses of Examples 3 and 4 above correspond to those of Method Example 1. For detailed implementation, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any of the methods of the present invention.

[0120] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0121] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0122] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A three-phase power load balancing method based on HPC adaptive scheduling strategy, characterized in that: include: Obtain computing node information and three-phase power load information when the data center is idle; Design test jobs and submit them to the data center for execution, obtaining three-phase power load information during operation; Extracting a three-phase power load variation from the three-phase power load information during the operation, and establishing a mapping relationship between the computing node and the three-phase power load based on the three-phase power load variation of each computing node; Obtaining an imbalance degree of computing node allocation based on the mapping relationship, and evaluating whether the mapping relationship is reasonable based on the imbalance degree; If the mapping relationship is reasonable, the computing nodes are dynamically allocated based on the mapping relationship to achieve three-phase power load balancing during daily operation.

2. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 1, characterized in that: The specific extraction of three-phase power load changes is: Calculate the three-phase power load change during the operation of job J on computing node Nj; A set of phases in which the calculation of the node Nj causes a load increase among the three phases is determined.

3. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 1, characterized in that: The mapping relationship between computing nodes and three-phase power loads is established as follows: Obtaining a total three-phase power load variation based on the three-phase power load variation of each computing node; Cleaning the total three-phase power load change based on a load change threshold to obtain cleaned data; A mapping relationship between computing nodes and phases is established based on the cleaned data.

4. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 3, characterized in that: The mapping relationship is expressed as: Pj={a,b,c|ΔLjk>0,k=a,b,c} Wherein, Pj represents the phase set in which the load of phase A, B, or C increases due to the calculation node Nj, a, b, and c represent phases A, B, and C respectively, and ΔLjk represents the change in the power load of phase k caused by the calculation node Nj.

5. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 1, characterized in that: The evaluation of whether the mapping relationship is reasonable based on the imbalance is specifically as follows: The imbalance degree of computing node allocation is calculated based on the mapping relationship; Comparing and evaluating the imbalance degree with a reasonable threshold; When the imbalance is greater than a reasonable threshold, the number of computing nodes assigned to the three phases is adjusted and a new imbalance is calculated. When the imbalance is less than or equal to a reasonable threshold, the mapping relationship is evaluated to be reasonable.

6. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 1, characterized in that: If the mapping relationship is reasonable, the computing nodes are dynamically allocated based on the mapping relationship as follows: Submit the new job Jn to the data center; Count the set of all computing nodes running jobs at the current moment and the actual three-phase power load; Use greedy algorithm to select resources for job Jn; An adaptive scheduling strategy is used to allocate computing nodes for the execution of job Jn.

7. The three-phase power load balancing method based on HPC adaptive scheduling strategy according to claim 6, characterized in that: Using the greedy algorithm, the resource selection for job Jn is as follows: Get the set of idle computing nodes based on the set of all computing nodes running jobs; If the set of idle computing nodes is empty or contains fewer computing nodes than those required by job Jn, wait until other jobs are completed and recalculate the set of all computing nodes currently running jobs and the actual three-phase power load. If the number of computing nodes included in the idle computing node set is greater than or equal to the number of computing nodes required by job Jn, the phase kj with the smallest current load is determined, and the computing node with the smallest number that has a mapping relationship with phase kj is selected from the idle computing node set. Combine the idle compute node set with the compute node Take the difference set to get a new set of idle computing nodes and recalculate the actual three-phase power load; Repeat the previous two steps until all required computing nodes are allocated to job Jn.

8. A three-phase power load balancing system based on HPC adaptive scheduling strategy, characterized in that: include: A data acquisition module is configured to: acquire computing node information and three-phase power load information when the data center is idle; An operation data acquisition module is configured to: design a test job and submit it to the data center for operation, and obtain three-phase power load information during the operation; a mapping relationship building module configured to: extract a three-phase power load change from the three-phase power load information during the operation, and establish a mapping relationship between the computing node and the three-phase power load based on the three-phase power load change of each computing node; a mapping relationship evaluation module configured to: obtain an imbalance degree of computing node allocation based on the mapping relationship, and evaluate whether the mapping relationship is reasonable based on the imbalance degree; The node dynamic allocation module is configured to dynamically allocate computing nodes based on the mapping relationship if the mapping relationship is reasonable, so as to achieve three-phase power load balancing during daily operation.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the three-phase power load balancing method based on the HPC adaptive scheduling strategy as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the three-phase power load balancing method based on the HPC adaptive scheduling strategy according to any one of claims 1 to 7 are implemented.

Images