Method and Device for Measuring Computing Power Utilization Rate of Intelligent Computing Center for Inclusive Computing Power

Abstract

The present invention provides a method and device for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power. The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure. The method includes: Step S1: Obtain the historical data of the utilization rate of the computing power cluster; Step S2: Calculate the second scheduling weight of the computing power resource nodes of the multiple nodes based on the calculation result of the computing power resource demand within the future time period and the first scheduling weight of the computing power resource nodes of the multiple nodes; Step S3: Calculate the computing power resource utilization rate calculation information of the target GPU within the future time period based on the second scheduling weight of the computing power resource nodes and the characteristic information of the target graphics processing unit (GPU). The present invention can greatly improve the calculation accuracy of the computing power resource utilization rate of the target GPU, avoid a large amount of waste of computing power resources, greatly reduce the economic cost, and is beneficial to the wide application of inclusive computing power.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers and computing power infrastructure, and particularly relates to a method and device for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power. Background Art

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "intelligent computing centers" have emerged as the times require.

[0003] An "intelligent computing center" refers to a facility that uses large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, and mainly provides the required computing power, data and algorithms for artificial intelligence applications (such as scenarios of artificial intelligence deep learning model development, model training and model inference, etc.). The intelligent computing center covers facilities, hardware and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0004] The "intelligent computing center" includes but is not limited to the "intelligent computing center".

[0005] An "intelligent computing center", that is, an artificial intelligence computing center, is a type of computing power infrastructure that is based on artificial intelligence theory, adopts an artificial intelligence computing architecture, and provides computing power services, data services and algorithm services required for artificial intelligence applications.

[0006] "Computing power" is the core of "intelligent computing centers" and "intelligent computing centers". It is the ability of computer devices or computing / data centers to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of target results by processing information data, and a new type of productive force that integrates information computing power, network carrying capacity and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0007] With the continuous development of artificial intelligence technology, Graphics Processing Unit (GPU) has played an increasingly important role in people's lives. However, the computing power resource utilization rates of current different GPUs are different. Therefore, before using a GPU, it is necessary to calculate the computing power resource utilization rate of the GPU. Currently, it is usually necessary to manually calculate the computing power resource utilization rate of the GPU. It can be seen that the current calculation accuracy of the computing power resource utilization rate of the GPU is very poor, resulting in a large waste of computing power resources and a very high economic cost, which is not conducive to the wide application of inclusive computing power. Summary of the Invention

[0008] The present invention provides a method and device for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power, which is used to solve the problem that the calculation accuracy of the computing power resource utilization rate of the GPU is very poor, resulting in a large waste of computing power resources, a very high economic cost, and being not conducive to the wide application of inclusive computing power.

[0009] To solve the above technical problems, the present invention is implemented as follows:

[0010] In a first aspect, the present invention provides a method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power, including:

[0011] Step S1: Obtain the historical utilization data of the computing power cluster, where the historical utilization data is the historical data of the computing power cluster using computing power resources;

[0012] Step S2: Based on the calculation result of the computing power resource demand in a future time period and the first scheduling weights of the computing power resource nodes of multiple nodes, calculate the second scheduling weights of the computing power resource nodes of the multiple nodes. The multiple nodes are the nodes within the computing power cluster. The first scheduling weight of the computing power resource node is used to represent the historical scheduling weight of the computing power resources among the multiple nodes, and the second scheduling weight of the computing power resource node is used to represent the scheduling weight of the computing power resources among the multiple nodes in the future time period. The calculation result of the computing power resource demand is obtained based on the historical utilization data of the computing power cluster;

[0013] Step S3: Based on the second scheduling weight of the computing power resource node and the characteristic information of the target GPU, calculate the computing power resource utilization rate calculation information of the target GPU in the future time period. The target GPU is the GPU corresponding to at least some of the multiple nodes.

[0014] Optionally, step S1 includes:

[0015] Step S11: Obtain the historical utilization data of the computing power cluster, and input the historical utilization data into a Long Short-Term Memory (LSTM) model for calculating the computing power resource demand, and output the calculation result of the computing power resource demand of the computing power cluster in the future time period. The LSTM model applies an Attention mechanism.

[0016] Optionally, step S2 includes:

[0017] Step S21: Input the first scheduling weight of the computing power resource node and the calculation result of the computing power resource demand into a deep learning model for weight calculation, and output the second scheduling weights of the computing power resource nodes of the multiple nodes. The deep learning model applies a multi-head attention mechanism.

[0018] Optionally, step S21 includes:

[0019] Step S211: Construct a node competition graph corresponding to the first scheduling weight of the computing power resource node, where the node competition graph is used to calculate the scheduling weight of the computing power resources among the multiple nodes;

[0020] Step S212: Input the node competition graph and the calculation result of the computing power resource demand into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node.

[0021] Optionally, S3 includes:

[0022] Step S31: Input the second scheduling weight of the computing power resource node and the feature information into a tree model to calculate the utilization rate of the computing power resources in the future time period, and output the calculation information of the computing power resource utilization rate.

[0023] Optionally, the feature information includes at least one of the following: static feature information, dynamic feature information, and context feature information of the target GPU;

[0024] Among them, the static feature information is used to represent the fixed parameters of the target GPU, the dynamic feature information is used to represent the dynamic change parameters during the use of the target GPU, and the context feature information is used to represent the parameters of other GPUs connected to the same node as the target GPU.

[0025] Optionally, it further includes:

[0026] Step S4: Calculate whether the calculation information of the computing power resource utilization rate is less than a preset threshold;

[0027] Step S5: When the calculation information of the computing power resource utilization rate is less than the preset threshold, control the scheduling of the computing power resources of the target GPU according to the calculation information of the computing power resource utilization rate.

[0028] Optionally, it further includes:

[0029] Step S6: When the calculation information of the computing power resource utilization rate is greater than or equal to the preset threshold, output a warning signal, where the warning signal is used to indicate that there is a deviation in the calculation information of the computing power resource utilization rate.

[0030] In a second aspect, the present invention provides an intelligent computing center computing power utilization rate measurement device for inclusive computing power, including:

[0031] An acquisition module, configured to acquire historical utilization rate data of a computing power cluster, where the historical utilization rate data is historical data of the computing power cluster using computing power resources;

[0032] A first calculation module, configured to calculate a second scheduling weight of computing power resources of the multiple nodes based on a calculation result of computing power resource requirements within a future time period and a first scheduling weight of computing power resources of the multiple nodes. The multiple nodes are nodes within the computing power cluster. The first scheduling weight of computing power resources is used to represent the historical scheduling weight of computing power resources among the multiple nodes. The second scheduling weight of computing power resources is used to represent the scheduling weight of computing power resources among the multiple nodes within the future time period. The calculation result of computing power resource requirements is obtained based on historical utilization data of the computing power cluster.

[0033] A second calculation module, configured to calculate computing power resource utilization calculation information of the target GPU within the future time period based on the second scheduling weight of computing power resources of the nodes and the characteristic information of the target GPU. The target GPU is a GPU corresponding to at least some of the multiple nodes.

[0034] In a third aspect, the present invention provides an electronic device, including: a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, the steps of the method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power as described in the first aspect above are implemented.

[0035] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power as described in the first aspect above are implemented.

[0036] In a fifth aspect, the present invention provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power as described in the first aspect above are implemented.

[0037] In the present invention, historical utilization data of a computing power cluster is obtained, and the historical utilization data is historical data of the computing power cluster using computing power resources; based on the calculation result of the computing power resource demand within a future time period and the first scheduling weights of the computing power resource nodes of multiple nodes, the second scheduling weights of the computing power resource nodes of the multiple nodes are calculated. The multiple nodes are nodes within the computing power cluster, and the first scheduling weight of the computing power resource node is used to represent the historical scheduling weight of the computing power resources among the multiple nodes. The second scheduling weight of the computing power resource node is used to represent the scheduling weight of the computing power resources among the multiple nodes within the future time period. The calculation result of the computing power resource demand is obtained based on the historical utilization data of the computing power cluster; based on the second scheduling weight of the computing power resource node and the characteristic information of the target graphics processing unit (GPU), the computing power resource utilization calculation information of the target GPU within the future time period is calculated. The target GPU is the GPU corresponding to at least some of the multiple nodes.

[0038] In this way, first, the calculation result of the computing power resource demand is obtained through the historical utilization data of the computing power cluster, and then the second scheduling weights of the computing power resource nodes of the multiple nodes are calculated through the calculation result of the computing power resource demand and the first scheduling weights of the computing power resource nodes of the multiple nodes. Finally, based on the second scheduling weight of the computing power resource node and the characteristic information of the target GPU, the computing power resource utilization calculation information of the target GPU within the future time period is calculated. In this way, when calculating the computing power resource utilization calculation information of the target GPU within the future time period, the historical utilization data of the computing power cluster and the first scheduling weights of the computing power resource nodes of the multiple nodes are considered. Moreover, the above historical utilization data of the computing power cluster and the first scheduling weights of the computing power resource nodes of the multiple nodes have a strong correlation with the target GPU, thereby improving the accuracy of the finally calculated computing power resource utilization calculation information of the target GPU within the future time period, avoiding a large amount of waste of computing power resources, significantly reducing economic costs, and being conducive to the wide application of inclusive computing power. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0040] Figure 1 is one of the flow diagrams of the method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power provided by the present invention;

[0041] Figure 2It is the second flowchart diagram of the method for measuring the computing power utilization rate of the intelligent computing center for inclusive computing power provided by the present invention;

[0042] Figure 3 It is the structural schematic diagram of the device for measuring the computing power utilization rate of the intelligent computing center for inclusive computing power provided by the present invention;

[0043] Figure 4 It is the structural schematic diagram of the electronic device provided by the present invention. Detailed implementation manners

[0044] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the accompanying drawings in the present invention. Obviously, the described content is part of the content of the present invention, rather than all of the content. Based on the content in the present invention, all other content obtained by those of ordinary skill in the art without creative efforts belongs to the scope of protection of the present invention.

[0045] First, the technical terms related to the present invention will be briefly described below.

[0046] The "computing power" referred to in the present invention means: It is the ability of a computer device or a computing / data center to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of the target result by processing information data, and a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0047] The "computational power" (Computational Power, CP) referred to in the present invention means: It is the ability of a data center server to process data and achieve the output of the result, a comprehensive index for measuring the computing ability of a data center, including general computing ability, supercomputing ability, and intelligent computing ability. The commonly used measurement unit is the number of floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), and the larger the value, the stronger the comprehensive computing ability. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe 2A or 500,000 mainstream server CPUs or 2 million mainstream laptops. The calculation formula is: CP = CP_general + CP_intelligent + CP_super.

[0048] The "carrying capacity" (Network Power, NP) referred to in the present invention means: It is the performance of the data transmission ability of the computing power facility, a comprehensive ability including network architecture, network bandwidth, transmission delay, intelligent management and scheduling, etc., involving network transmission inside and between data centers, and is a comprehensive index for measuring network transmission scheduling ability.

[0049] The "Storage Power" (SP) described in the present invention refers to: the comprehensive ability of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon. It is a comprehensive indicator to measure the data storage capacity of a data center, including external storage devices such as storage arrays and built-in storage devices of servers. The common measurement unit of storage capacity is exabyte (EB, 1EB = 2^60 bytes), the common measurement unit of performance is the number of read and write operations per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB), and the disaster recovery ratio is an important manifestation of security and reliability.

[0050] The "computing power infrastructure" described in the present invention refers to: a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage power, which can realize the centralized computing, storage, transmission, and application of information, presenting characteristics such as multi-element ubiquitous, intelligent and agile, secure and reliable, green and low-carbon, etc., and is of great significance for boosting industrial transformation and upgrading, empowering China's scientific and technological innovation, meeting people's better life, and realizing high-efficiency social governance.

[0051] The "new type of information infrastructure" described in the present invention mainly refers to: network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, and satellite Internet, computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, and supercomputing centers, and new technology facilities such as artificial intelligence, blockchain, and quantum computing. With the emergence and popularization of new general technologies, the form of the new type of information infrastructure will be more rich and diverse.

[0052] The "computing power" described in the present invention includes: general computing power, intelligent computing power, and super computing power.

[0053] The "general computing power" described in the present invention refers to: the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0054] The "intelligent computing power" described in the present invention refers to: for various artificial intelligence innovation applications, a computing platform based on the large-scale deployment of dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit), such as natural language processing, machine vision, etc.

[0055] The "super computing power" described in the present invention refers to: mainly the computing power provided by high-performance computing power clusters such as supercomputers. It uses the centralized computing resources of multiple computer systems working in parallel and processes extremely complex or data-intensive problems through a dedicated operating system. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, gene analysis, etc.

[0056] The "intelligent computing center" described in the present invention refers to: a facility that mainly provides the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios like artificial intelligence deep learning model development, model training, and model inference) by using large-scale heterogeneous computing power resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0057] The "intelligent computing center" described in the present invention includes but is not limited to the intelligent computing center.

[0058] The "intelligent computing center" described in the present invention, that is, the artificial intelligence computing center, is a type of computing power infrastructure based on artificial intelligence theory, adopting an artificial intelligence computing architecture, and providing computing power services, data services, and algorithm services required for artificial intelligence applications.

[0059] The "computing power center" described in the present invention refers to: a facility mainly composed of infrastructure such as wind, fire, water, and electricity and IT software and hardware devices, with computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0060] The "supercomputing center" described in the present invention refers to: that is, the supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters and can provide functions such as large-scale computing, storage, and network services, and is widely used in application scenarios such as aerospace, national defense, oil exploration, climate modeling, and genome sequencing.

[0061] The "computing power resources" described in the present invention refers to: technologies and facilities required for the development of the digital society with information computing, transmission, storage, and application capabilities, including but not limited to computing resources such as CPU and GPU, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and support and guarantee resources such as wind, fire, water, and electricity.

[0062] The "inclusive computing power" described in the present invention refers to providing appropriate and effective computing power services for all social strata and groups with computing power service needs at an affordable cost based on the requirements of equal opportunity and the principle of commercial sustainability.

[0063] Please refer to Figure 1 , Figure 1It is a flowchart of a method for measuring the computing power utilization rate of an intelligent computing center for inclusive computing power provided by the present invention. As Figure 1 shown, it includes the following steps:

[0064] Step S1: Obtain the historical data of the utilization rate of the computing power cluster, where the historical data of the utilization rate is the historical data of the computing power cluster using computing power resources;

[0065] It should be noted that the present invention can be applied to an electronic device, and the electronic device can include a resource aggregation layer, or the electronic device is connected to the resource aggregation layer. The historical data of the utilization rate of the computing power cluster can be stored on the resource aggregation layer. In this way, the electronic device can obtain the above historical data of the utilization rate from the resource aggregation layer, or the above historical data of the utilization rate can be stored on other electronic devices. The other electronic device can be a server, and the electronic device in the present invention can obtain the above historical data of the utilization rate from other electronic devices through the resource aggregation layer.

[0066] In addition, the triggering conditions of step S1 can include at least one of the following: every preset period, and when the load fluctuation of the computing power cluster exceeds the load threshold. The specific values of the above preset period and load threshold are not limited here.

[0067] Among them, the specific types of the historical data of the utilization rate are not limited here. Optionally, the historical data of the utilization rate can include the historical data of the utilization rate of the GPUs in the computing power cluster. That is, the computing power cluster can include multiple nodes, and each node can include multiple GPUs. The historical data of the utilization rate of the GPUs in the computing power cluster can refer to the sum of the historical data of the utilization rate of at least some of the GPUs included in the above computing power cluster.

[0068] Optionally again, the historical data of the utilization rate can also include at least one of the following: the historical data of the utilization rate of the central processing unit (CPU) of the computing power cluster, the historical data of the utilization rate of the memory of the computing power cluster, and the historical data of the utilization rate of the input / output (IO) interface. It should be noted that the above historical data of the utilization rate of the CPU, the historical data of the utilization rate of the memory, and the historical data of the utilization rate of the IO interface are all related to the historical data of the utilization rate of the GPUs in the computing power cluster. For example: the higher the utilization rate of the GPUs in the computing power cluster, the higher the utilization rate of the CPU, the utilization rate of the memory, and the utilization rate of the IO interface. In this way, the richer the content included in the historical data of the utilization rate, the higher the accuracy of the calculated information on the utilization rate of the computing power resources of the target GPU in the future time period.

[0069] It should be noted that the time period corresponding to the historical data of the utilization rate of the computing power cluster is not limited here. Optionally, the time period corresponding to the historical data of the utilization rate of the computing power cluster can be the first time period before the current moment. The start time and end time of the first time period can be the first moment and the second moment respectively. The interval duration between the second moment and the current moment can be a preset duration, and the specific value of the preset duration is not limited here. Optionally, the preset duration can be 0, that is, the second moment is the current moment; alternatively, the preset duration can be greater than 0, that is, the second moment is not the current moment.

[0070] Step S2: Based on the calculation result of the computing power resource demand in the future time period and the first scheduling weights of the computing power resource nodes of multiple nodes, calculate the second scheduling weights of the computing power resource nodes of the multiple nodes. The multiple nodes are the nodes in the computing power cluster. The first scheduling weights of the computing power resource nodes are used to represent the historical scheduling weights of the multiple nodes for the computing power resources. The second scheduling weights of the computing power resource nodes are used to represent the scheduling weights of the multiple nodes for the computing power resources in the future time period. The calculation result of the computing power resource demand is obtained based on the historical data of the utilization rate of the computing power cluster.

[0071] Among them, the specific value of the future time period is not limited here. Optionally, the future time period can be the first cycle with a duration of 30 minutes starting from the current moment. Alternatively, the future time period can also be the second cycle after the above first cycle. The durations of the first cycle and the second cycle can be equal, and there can be a time interval between the first cycle and the second cycle. That is, the future time period can be understood as the future time period after the current moment.

[0072] Among them, the calculation result of the computing power resource demand can also be referred to as the overall resource demand of the computing power cluster in the future time period.

[0073] Among them, the first scheduling weights of the computing power resource nodes can be understood as being used to represent the scheduling weights of the multiple nodes for the computing power resources. If the scheduling weight is larger, the utilization rate of the computing power resources of this node is higher. For example, the first scheduling weights of the computing power resource nodes are used to represent the scheduling weights of different nodes for the CPU, or the first scheduling weights of the computing power resource nodes are used to represent the scheduling weights of different nodes for the memory bandwidth.

[0074] The storage form of the first scheduling weights of the computing power resource nodes is not limited here. Optionally, the first scheduling weights of the computing power resource nodes can be stored in a matrix or a curve graph. When the first scheduling weights of the computing power resource nodes are stored in the matrix form, the above matrix can be referred to as the computing power resource interaction matrix; alternatively, when the first scheduling weights of the computing power resource nodes are stored in the curve graph, the above curve graph can be referred to as the computing power resource interaction curve graph.

[0075] Among them, the first scheduling weight of the computing power resource node can be understood as the scheduling weight of the computing power resources among multiple nodes before the current moment, while the second scheduling weight of the computing power resource node can be understood as the scheduling weight of the computing power resources among multiple corresponding nodes after the current moment, that is, the second scheduling weight of the computing power resource node can be understood as the scheduling weight of the computing power resources among multiple nodes within a future time period.

[0076] Step S3: Based on the second scheduling weight of the computing power resource node and the characteristic information of the target GPU, calculate the computing power resource utilization rate calculation information of the target GPU within the future time period, where the target GPU is the GPU corresponding to at least some of the multiple nodes.

[0077] Among them, the characteristic information of the target GPU can also be referred to as the hardware characteristic information or feature engineering information of the target GPU, and the above-mentioned feature engineering information can be abbreviated as feature engineering, and the characteristic information of the target GPU can be calculated based on the parameters of the target GPU and / or the parameters of other GPUs connected to the same node as the target GPU.

[0078] Optionally, the target GPU can be one of the multiple GPUs corresponding to a certain node among the multiple nodes; alternatively, the target GPU can be several of the multiple GPUs corresponding to a certain node among the multiple nodes, and specific details are not limited here.

[0079] In the present invention, first, the computing power resource demand calculation result is calculated through the historical utilization rate data of the computing power cluster, then the second scheduling weight of the computing power resource node of multiple nodes is calculated through the computing power resource demand calculation result and the first scheduling weight of the computing power resource node of multiple nodes, and finally, based on the second scheduling weight of the computing power resource node and the characteristic information of the target GPU, the computing power resource utilization rate calculation information of the target GPU within the future time period is calculated. In this way, when calculating the computing power resource utilization rate calculation information of the target GPU within the future time period, the historical utilization rate data of the computing power cluster and the first scheduling weight of the computing power resource node of multiple nodes are considered, and the above-mentioned historical utilization rate data of the computing power cluster and the first scheduling weight of the computing power resource node of multiple nodes have a strong correlation with the target GPU, so as to improve the accuracy of the finally calculated computing power resource utilization rate calculation information of the target GPU within the future time period, avoid a large amount of waste of computing power resources, greatly reduce the economic cost, and be conducive to the wide application of inclusive computing power.

[0080] Optionally, step S1 includes:

[0081] Step S11: Obtain the historical utilization data of the computing power cluster, input the historical utilization data into the LSTM model for computing power resource demand calculation, and output the calculation result of the computing power resource demand of the computing power cluster in the future time period. The Attention mechanism is applied in the LSTM model.

[0082] Among them, since the time period corresponding to the historical utilization data can be the first time period before the current moment, that is, if the first time period is different, the historical utilization data may also be different, that is, the historical utilization data is related to time. Therefore, the historical utilization data can be referred to as time series data.

[0083] Optionally, step S11 includes: when the change value of the computing power resource demand calculation result is within a preset range or less than a preset change threshold, input the historical utilization data into the LSTM model for computing power resource demand calculation, and output the calculation result of the computing power resource demand of the computing power cluster in the future time period. The change value of the above computing power resource demand calculation result being within a preset range or less than a preset change threshold can also be referred to as the stability of the computing power resource demand calculation result.

[0084] In addition, when the change value of the computing power resource demand calculation result is not within the preset range or is greater than or equal to the preset change threshold, it can be referred to as the instability of the computing power resource demand calculation result. At this time, the LSTM model can be retrained.

[0085] In the present invention, since the LSTM model is good at processing time series data and can learn the periodic law of the change of computing power resources through the historical utilization data, and the Attention mechanism can identify key time points and dynamically adjust the calculation weights of various input data, thereby improving the accuracy of the finally calculated computing power resource demand calculation result. The above various input data can refer to different data in the historical utilization data of the computing power cluster. For example, it can refer to data at different times.

[0086] Among them, the specific types of the periodic law and the key time points are not limited here. Optionally, the periodic law can refer to the differential law of the computing power load during the morning and evening peaks, and the key time point can refer to the sudden traffic period, such as the period when the traffic suddenly increases or the period when the traffic suddenly decreases.

[0087] It should be noted that since the consumption of the computing power resources of the entire computing power cluster has a large correlation with the key fluctuation period, that is, the configuration of the computing power resources of the entire computing power cluster needs to ensure that the needs of the key fluctuation period are met. For example: the configuration of the computing power resources of the entire computing power cluster needs to ensure that the needs of key fluctuation periods such as the period when the traffic suddenly increases and the period when the traffic suddenly decreases are met. In this way, if the configuration of the overall resources of the computing power cluster is always configured according to the needs of the period when the traffic suddenly increases, it is easy to cause over-allocation of computing power resources in other periods, resulting in waste of resources. And the present invention calculates the calculation result of the computing power resource demand of the computing power cluster in the future time period, so as to accurately calculate the demand for computing power resources at different times in the future time period, and flexibly configure the computing power resources at different times according to this demand, thereby avoiding waste of computing power resources.

[0088] For example: The application scenario of the present invention is: the demand for computing power resources in a period can be calculated according to the period of a major e-commerce promotion, and the resource pool of the computing power cluster in this period can be expanded in advance before the major e-commerce promotion according to the demand in this period.

[0089] Optionally, step S2 includes:

[0090] Step S21: Input the first scheduling weight of the computing power resource node and the calculation result of the computing power resource demand into a deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node of the multiple nodes. The deep learning model applies a multi-head attention mechanism.

[0091] Among them, the specific type of the deep learning model is not limited here. Optionally, the deep learning model can be a Transformer model.

[0092] Among them, the second scheduling weight of the computing power resource node can be the scheduling weight of different nodes for the computing power resources in the future time period.

[0093] In the present invention, a multi-head attention mechanism is applied in the deep learning model, and the deep learning model can separate resource competition modes in different dimensions through the multi-head attention mechanism. For example: separating CPU-intensive tasks and IO-intensive tasks, and outputting the second scheduling weight of the computing power resource node of each corresponding node according to different resource competition modes. In this way, the calculated second scheduling weight of the computing power resource node is more accurate, and the calculation method of the second scheduling weight of the computing power resource node is more flexible, optimizing the calculation method of the second scheduling weight of the computing power resource node.

[0094] It should be noted that, the vector corresponding to the first scheduling weight of the computing power resource node and the vector corresponding to the calculation result of the computing power resource requirement can be extracted first, and then the vector corresponding to the first scheduling weight of the computing power resource node and the vector corresponding to the calculation result of the computing power resource requirement are input into the deep learning model for weight calculation, and the second scheduling weight of the computing power resource node of multiple nodes is output.

[0095] For example: when the vector corresponding to the calculation result of the computing power resource requirement can be used as the reference vector for position encoding, and then the vector corresponding to the first scheduling weight of the computing power resource node and the above reference vector are input into the deep learning model for weight calculation, so as to output the second scheduling weight of the computing power resource node of multiple nodes.

[0096] Optionally, the step S21 includes:

[0097] Step S211: Construct a node competition graph corresponding to the first scheduling weight of the computing power resource node, and the node competition graph is used to calculate the scheduling weight of the computing power resources among the multiple nodes;

[0098] Step S212: Input the node competition graph and the calculation result of the computing power resource requirement into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node.

[0099] Among them, by constructing a node competition graph, and the node competition graph is used to calculate the scheduling weight of the computing power resources among multiple nodes, and the scheduling weights of different nodes for the computing power resources are different. In this way, the cross-node computing power resource scheduling relationship among multiple nodes can be intuitively reflected through the node competition graph.

[0100] In the present invention, by constructing a node competition graph, the scheduling weight of the computing power resources among multiple nodes can be calculated, avoiding the appearance of "starved nodes", that is, making the computing power resource scheduling of multiple nodes more balanced, and optimizing the second scheduling weight of the computing power resource node. At the same time, through the deep learning model for weight calculation, the second scheduling weight of the computing power resource node of multiple nodes is output. In this way, the calculation efficiency of the second scheduling weight of the computing power resource node can be improved, and the calculation accuracy of the second scheduling weight of the computing power resource node can be improved.

[0101] It should be noted that, the above starved nodes can be understood as nodes with less computing power resource scheduling.

[0102] The application scenario of the present invention can be: when the computing power cluster is an open-source container orchestration and management platform computing power cluster, and optionally, the computing power cluster is a Kubernetes computing power cluster, during the scheduling process of the Kubernetes computing power cluster, the second scheduling weight of the computing power resource node of multiple nodes can be output, so that the load of the nodes corresponding to the GPU and the nodes not corresponding to the GPU can be dynamically balanced.

[0103] Optionally, S3 includes:

[0104] Step S31: Input the second scheduling weight of the computing power resource node and the feature information into a tree model to calculate the utilization rate of the computing power resources in the future time period, and output the computing power resource utilization rate calculation information.

[0105] Wherein, the specific type of the tree model is not limited herein. Optionally, the tree model may include at least one of a Light Gradient Boosting Machine (LightGBM) model and an eXtreme Gradient Boosting (XGBoost) model.

[0106] In the present invention, the utilization rate of the computing power resources in the future time period is calculated through a tree model, so as to output the computing power resource utilization rate calculation information. The second scheduling weight of the computing power resource node and the feature information may include non-linear features, and the tree model is good at processing the interaction of non-linear features. In this way, the calculation efficiency of the computing power resource utilization rate calculation information can be improved; at the same time, the tree model can break through the key bottleneck of feature importance analysis. For example, it can reduce the impact of the fragmented index information included in the feature information on the inference delay, thereby further improving the calculation efficiency of the computing power resource utilization rate calculation information, and adjusting or guiding the video memory pre-scheduling strategy between the target GPU and other GPUs through the output computing power resource utilization rate calculation information. The above video memory can be understood as the computing power resources of the GPU.

[0107] Wherein, when the feature information includes fragmented index information and batch size parameters, the tree model can be used to process the correlation between the fragmented index information and the batch size parameters.

[0108] It should be noted that the application scenario of the present invention may include: the tree model can be applied in deep learning training, so as to optimize the video memory utilization rate and reduce the occurrence of Out Of Memory (OOM) error phenomena. And the present invention can calculate fine-grained computing power resources such as the computing power resource utilization rate calculation information through the tree model and high-dimensional heterogeneous data (such as fragmented index information).

[0109] Optionally, the feature information includes at least one of the following: static feature information, dynamic feature information, and context feature information of the target GPU;

[0110] Among them, the static feature information is used to represent the fixed parameters of the target GPU, the dynamic feature information is used to represent the dynamically changing parameters during the use of the target GPU, and the context feature information is used to represent the parameters of other GPUs connected to the same node as the target GPU.

[0111] Among them, the static feature information may include at least one of the following parameters of the target GPU: GPU model, video memory capacity, Peripheral Component Interconnect Express (PCIe) bandwidth, etc.

[0112] Among them, the dynamic feature information may include the video memory fragmentation index of the target GPU, and the video memory fragmentation index may also be referred to as the video memory fragmentation metric, fragmentation index information, or video memory fragmentation degree parameter.

[0113] It should be noted that the acquisition method of the above video memory fragmentation index is not limited herein. Optionally, the above video memory fragmentation index can be obtained by real-time sampling through a GPU performance monitoring tool.

[0114] Among them, the context feature information may include at least one of the following: various parameters of other GPUs connected to the same node as the target GPU (such as the second scheduling weight of the computing power resource node) and the process-level resource lock status information.

[0115] In the present invention, the feature information includes at least one of the following: the static feature information, dynamic feature information, and context feature information of the target GPU. In this way, the diversity of the types of feature information is increased. When the types of feature information are more, the accuracy of the computing power resource utilization rate calculation information of the target GPU calculated finally in the future time period is also higher.

[0116] Optionally, it further includes:

[0117] Step S4: Calculate whether the computing power resource utilization rate calculation information is less than a preset threshold;

[0118] Step S5: When the computing power resource utilization rate calculation information is less than the preset threshold, control the computing power resource scheduling of the target GPU according to the computing power resource utilization rate calculation information.

[0119] Among them, the specific value of the preset threshold is not limited herein. Optionally, the preset threshold may be an empirical value obtained through multiple statistics during the training process. Optionally, the preset threshold may be a value downloaded from other electronic devices.

[0120] In the present invention, when the calculated computing power resource utilization rate information is less than a preset threshold, it indicates that the accuracy of the calculated computing power resource utilization rate information is relatively high. Therefore, the computing power resource scheduling of the target GPU can be controlled based on the computing power resource utilization rate information, ensuring the accuracy of the computing power resource scheduling of the target GPU.

[0121] Optionally, it further includes:

[0122] Step S6: When the calculated computing power resource utilization rate information is greater than or equal to the preset threshold, an early warning signal is output, and the early warning signal is used to indicate that there is a deviation in the calculated computing power resource utilization rate information.

[0123] In the present invention, when the calculated computing power resource utilization rate information is greater than or equal to the preset threshold, it indicates that there is a deviation in the calculated computing power resource utilization rate information and there is a phenomenon of video memory leakage. Therefore, an early warning signal can be output to enhance the early warning effect.

[0124] It should be noted that when the early warning signal is output, it indicates that there is a deviation in the calculation process of the calculated computing power resource utilization rate information at this time. At this time, the model used in the calculation process of the calculated computing power resource utilization rate information can be calibrated, thereby improving the calculation accuracy of the calculated computing power resource utilization rate information. The above process can also be referred to as a deviation backpropagation mechanism or an error backpropagation mechanism.

[0125] For example: when the calculated computing power resource utilization rate information is greater than or equal to the preset threshold, the calibration of the calculation model of the node corresponding to the target GPU can be triggered. When the cumulative error of the calculation model corresponding to the node reaches the target value, the attention mechanism weight of the calculation model of the computing power cluster corresponding to the node can be updated. In this way, the end-to-end connection of the hierarchical calculation of the target GPU, node, and computing power cluster can be realized, ensuring the consistency of the calculation results. Optionally, the calculation model of the above computing power cluster can include an LSTM model, and the calculation model corresponding to the node can include a deep learning model.

[0126] See Figure 2 , Figure 2 which is a specific illustration provided by the present invention to fully illustrate the above content. The technical features in Figure 2 can refer to the above relevant descriptions and have the same beneficial technical effects, which will not be elaborated here specifically. Optionally, Figure 2The Transformer encoder in it can be understood as the above-mentioned Transformer model, the node resource weight can be understood as the second scheduling weight of the computing power resources of multiple nodes above, the GPU load prediction can be understood as the calculation process of the computing power resource utilization rate calculation information of the target GPU in the future time period, and the anomaly detection can be understood as the process of calculating whether the computing power resource utilization rate calculation information is less than a preset threshold.

[0127] See Figure 3 , Figure 3 FIG. is a schematic structural diagram of a computing power utilization rate measurement device for an intelligent computing center for inclusive computing power provided by the present invention. As Figure 3 shown, the computing power utilization rate measurement device 300 for an intelligent computing center for inclusive computing power includes:

[0128] An acquisition module 301, configured to acquire historical utilization rate data of a computing power cluster, where the historical utilization rate data is historical data of the computing power cluster using computing power resources;

[0129] A first calculation module 302, configured to calculate a second scheduling weight of the computing power resource nodes of the multiple nodes based on a calculation result of the computing power resource demand in a future time period and a first scheduling weight of the computing power resource nodes of the multiple nodes. The multiple nodes are nodes in the computing power cluster. The first scheduling weight of the computing power resource nodes is used to represent the historical scheduling weight of the computing power resources among the multiple nodes, and the second scheduling weight of the computing power resource nodes is used to represent the scheduling weight of the computing power resources among the multiple nodes in the future time period. The calculation result of the computing power resource demand is calculated based on the historical utilization rate data of the computing power cluster;

[0130] A second calculation module 303, configured to calculate computing power resource utilization rate calculation information of a target GPU in the future time period based on the second scheduling weight of the computing power resource nodes and feature information of the target GPU. The target GPU is a GPU corresponding to at least some of the multiple nodes.

[0131] Optionally, the acquisition module 301 is further configured to acquire the historical utilization rate data of the computing power cluster, input the historical utilization rate data into a long short-term memory network (LSTM) model for calculating the computing power resource demand, and output the calculation result of the computing power resource demand of the computing power cluster in the future time period. The attention mechanism is applied in the LSTM model.

[0132] Optionally, the first calculation module 302 is further configured to input the first scheduling weight of the computing power resource node and the calculation result of the computing power resource requirement into a deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node of the multiple nodes. The deep learning model applies a multi-head attention mechanism.

[0133] Optionally, the first calculation module 302 includes:

[0134] A construction sub-module, configured to construct a node competition graph corresponding to the first scheduling weight of the computing power resource node, where the node competition graph is used to calculate the scheduling weight of the computing power resource between the multiple nodes;

[0135] A calculation sub-module, configured to input the node competition graph and the calculation result of the computing power resource requirement into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node.

[0136] Optionally, the second calculation module 303 is further configured to input the second scheduling weight of the computing power resource node and the feature information into a tree model to calculate the utilization rate of the computing power resource in the future time period, and output the computing power resource utilization rate calculation information.

[0137] Optionally, the feature information includes at least one of the following: static feature information, dynamic feature information, and context feature information of the target GPU;

[0138] Wherein, the static feature information is used to represent the fixed parameters of the target GPU, the dynamic feature information is used to represent the dynamic change parameters during the use of the target GPU, and the context feature information is used to represent the parameters of other GPUs connected to the same node as the target GPU.

[0139] Optionally, the intelligent computing center computing power utilization rate measurement device 300 for inclusive computing power further includes:

[0140] A third calculation module, configured to calculate whether the computing power resource utilization rate calculation information is less than a preset threshold;

[0141] A control module, configured to control the scheduling of the computing power resource of the target GPU according to the computing power resource utilization rate calculation information when the computing power resource utilization rate calculation information is less than the preset threshold.

[0142] Optionally, the intelligent computing center computing power utilization rate measurement device 300 for inclusive computing power further includes:

[0143] An output module, configured to output a warning signal when the computing power resource utilization rate calculation information is greater than or equal to the preset threshold, where the warning signal is used to indicate that there is a deviation in the computing power resource utilization rate calculation information.

[0144] The computing power utilization rate measuring device 300 for the inclusive computing power-oriented intelligent computing center provided by the present invention can execute each step in the above-mentioned method for measuring the computing power utilization rate of the inclusive computing power-oriented intelligent computing center, and thus has the same beneficial technical effects as the above-mentioned method for measuring the computing power utilization rate of the inclusive computing power-oriented intelligent computing center, and will not be elaborated herein for the sake of brevity.

[0145] Please refer to Figure 4 , the present invention also provides an electronic device 40, including a processor 41, a memory 42, and a computer program stored on the memory 42 and executable on the processor 41. When the computer program is executed by the processor 41, it realizes each process shown in the above-mentioned method for measuring the computing power utilization rate of the inclusive computing power-oriented intelligent computing center, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0146] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes each process shown in the above-mentioned method for measuring the computing power utilization rate of the inclusive computing power-oriented intelligent computing center, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0147] The present invention also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they realize each process shown in the above-mentioned Figure 1 method for measuring the computing power utilization rate of the inclusive computing power-oriented intelligent computing center shown, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0148] It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without more limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article, or device including that element.

[0149] Through the description of the above embodiments, those skilled in the art can clearly understand that the method provided by the above invention can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to enable a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the various methods provided by the present invention.

[0150] The present invention has been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them belong to the protection scope of the present invention.

Claims

1. A method for measuring computing power utilization rate of intelligent computing centers for universal computing power, characterized in that: include: Step S1: Obtaining historical usage data of a computing power cluster, where the historical usage data is historical data of computing power resources used by the computing power cluster; Step S2: based on the calculation result of the computing power resource demand in the future time period and the first scheduling weights of the computing power resource nodes of the multiple nodes, calculate the second scheduling weights of the computing power resource nodes of the multiple nodes, the multiple nodes are nodes in the computing power cluster, the first scheduling weights of the computing power resource nodes are used to represent the historical scheduling weights of the computing power resources between the multiple nodes, the second scheduling weights of the computing power resource nodes are used to represent the scheduling weights of the computing power resources between the multiple nodes in the future time period, and the computing power resource demand calculation result is calculated based on the historical usage data of the computing power cluster; Step S3: Calculating computing resource usage information of the target GPU in the future time period based on the second scheduling weight of the computing resource node and the characteristic information of the target graphics processor GPU, wherein the target GPU is a GPU corresponding to at least some of the multiple nodes; The step S1 comprises: Step S11: Obtaining the usage history data of the computing power cluster, and inputting the usage history data into a long short-term memory network (LSTM) model to calculate the computing power resource demand, and outputting the calculation result of the computing power resource demand of the computing power cluster in the future time period, wherein an attention mechanism is applied in the LSTM model; The step S2 comprises: Step S21: Input the first scheduling weight of the computing power resource node and the calculation result of the computing power resource demand into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node of the multiple nodes. A multi-head attention mechanism is applied in the deep learning model.

2. The method according to claim 1, characterized in that The step S21 comprises: Step S211: constructing a node competition graph corresponding to the first scheduling weight of the computing power resource node, wherein the node competition graph is used to calculate the scheduling weights of the computing power resources between the multiple nodes; Step S212: Input the node competition graph and the computing power resource demand calculation result into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node.

3. The method according to claim 1, characterized in that The S3 includes: Step S31: input the second scheduling weight of the computing resource node and the characteristic information into the tree model to calculate the utilization rate of the computing resource in the future time period, and output the computing resource utilization rate calculation information.

4. The method according to claim 3, characterized in that: The feature information includes at least one of the following: static feature information, dynamic feature information, and context feature information of the target GPU; The static feature information is used to represent the fixed parameters of the target GPU, the dynamic feature information is used to represent the dynamically changing parameters of the target GPU during use, and the context feature information is used to represent the parameters of other GPUs connected to the same node as the target GPU.

5. The method according to any one of claims 1 to 4, characterized in that Also includes: Step S4: Calculate whether the computing power resource utilization calculation information is less than a preset threshold; Step S5: When the computing power resource utilization calculation information is less than the preset threshold, the computing power resource scheduling of the target GPU is controlled according to the computing power resource utilization calculation information.

6. The method according to claim 5, characterized in that Also includes: Step S6: When the computing power resource utilization calculation information is greater than or equal to the preset threshold, an early warning signal is output, wherein the early warning signal is used to indicate that there is a deviation in the computing power resource utilization calculation information.

7. A computing power utilization rate metering device for intelligent computing centers oriented to universal computing power, characterized in that: include: An acquisition module is used to acquire historical usage data of a computing power cluster, where the historical usage data is historical data of computing power resources used by the computing power cluster; A first calculation module is used to calculate the second scheduling weights of the computing resource nodes of the multiple nodes based on the computing resource demand calculation result in the future time period and the computing resource node first scheduling weights of the multiple nodes, the multiple nodes are nodes in the computing power cluster, the computing resource node first scheduling weight is used to represent the historical scheduling weights of the computing resources between the multiple nodes, the computing resource node second scheduling weight is used to represent the scheduling weights of the computing resources between the multiple nodes in the future time period, and the computing resource demand calculation result is calculated based on the usage history data of the computing power cluster; A second calculation module is used to calculate the computing resource usage calculation information of the target GPU in the future time period based on the second scheduling weight of the computing resource node and the characteristic information of the target GPU, wherein the target GPU is a GPU corresponding to at least some of the multiple nodes; The acquisition module is further used to acquire the usage history data of the computing power cluster, input the usage history data into the long short-term memory network LSTM model to calculate the computing power resource demand, and output the calculation result of the computing power resource demand of the computing power cluster in the future time period, and the LSTM model is applied with the attention mechanism; The first calculation module is also used to input the first scheduling weight of the computing power resource node and the calculation result of the computing power resource demand into the deep learning model for weight calculation, and output the second scheduling weight of the computing power resource node of the multiple nodes. A multi-head attention mechanism is applied in the deep learning model.

8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for measuring computing power usage rate of an intelligent computing center for inclusive computing power as described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for measuring computing power utilization rate of an intelligent computing center for inclusive computing power as described in any one of claims 1 to 6.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of the method for measuring computing power utilization rate of an intelligent computing center for inclusive computing power as described in any one of claims 1 to 6.

Images