CN117636137A - GPU bare metal computing power resource allocation scheduling method, device and storage medium - Google Patents
GPU bare metal computing power resource allocation scheduling method, device and storage medium Download PDFInfo
- Publication number
- CN117636137A CN117636137A CN202410108130.0A CN202410108130A CN117636137A CN 117636137 A CN117636137 A CN 117636137A CN 202410108130 A CN202410108130 A CN 202410108130A CN 117636137 A CN117636137 A CN 117636137A
- Authority
- CN
- China
- Prior art keywords
- sub
- image
- virtual machine
- computing power
- frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 239000002184 metal Substances 0.000 title claims abstract description 46
- 238000013468 resource allocation Methods 0.000 title claims description 20
- 230000011218 segmentation Effects 0.000 claims abstract description 79
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 238000000638 solvent extraction Methods 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 6
- 230000015654 memory Effects 0.000 description 17
- 238000004590 computer program Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention relates to the technical field of GPU resource management, and particularly discloses a method, a device and a storage medium for distributing and scheduling bare metal computing power resources of a GPU, wherein the method comprises the steps of carrying out gradient segmentation on the computing power resources to construct sub virtual units; determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency, and determining a reference difficulty based on the image recognition frequency; and selecting the sub-virtual machine according to the reference difficulty, transmitting the image to the sub-virtual machine for identification, and updating the image identification frequency according to the identification result. According to the invention, the computing power resources are subjected to gradient segmentation to obtain a plurality of sub virtual machines, when the images are received, the change conditions in the images are analyzed, and then, which sub virtual machine is used for processing the subsequent images is determined, so that different images are processed by equipment with different capabilities, and the resource utilization rate is greatly improved.
Description
Technical Field
The invention relates to the technical field of GPU resource management, in particular to a method and a device for distributing and scheduling bare metal computing power resources of a GPU and a storage medium.
Background
Bare metal computing power is a high performance computing capability provided by physical servers. The method directly utilizes hardware resources of a physical server, such as a CPU, a GPU, a memory and the like, and provides efficient, stable and reliable computing service for users; the GPU bare metal server has limited computational resources, the data facing the GPU bare metal server is graphics data, the graphics data has single content, the graphics data has complex content, the processing difficulties are different, and the GPU bare metal server may be contained in the same video, so that a single GPU can face graphics data with different processing difficulties.
In order to ensure the processing effect, the performance of the GPU must take the most complex graphics data as a reference, that is, the GPU must be able to identify the most complex graphics data, so that when the GPU faces simple graphics data, a part of resources are not utilized, the resource utilization rate is not high, and how to improve the resource utilization rate when the GPU processes video data is a technical problem to be solved by the technical scheme of the present invention.
Disclosure of Invention
The invention aims to provide a method and a device for distributing and scheduling bare metal computing power resources of a GPU and a storage medium, so as to solve the problems in the background art.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a GPU bare metal computing power resource allocation scheduling method, the method comprising:
acquiring computing power resources of a GPU bare metal server, and carrying out gradient segmentation on the computing power resources to construct a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency, and determining a reference difficulty based on the image recognition frequency;
selecting a sub-virtual machine according to the reference difficulty, transmitting an image to the sub-virtual machine for identification, and updating the image identification frequency according to the identification result;
acquiring the image processing quantity of each sub-virtual machine in real time, and updating the computing power resource quantity occupied by the sub-virtual machines according to the image processing quantity;
the step of obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource and constructing a sub-virtual machine set comprises the following steps:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub virtual machine with the smallest accumulated result.
As a further scheme of the invention: the step of determining the image reading frequency and the image recognition frequency according to the computing power resource, reading the image based on the image reading frequency, and determining the reference difficulty based on the image recognition frequency comprises the following steps:
acquiring the type of the video to be processed, and inquiring the corresponding sub-virtual machine set according to the type;
reading the total segmentation number corresponding to the sub-virtual machine set, and determining the image reading frequency according to the total segmentation number;
reading the segmentation step length corresponding to the sub-virtual machine set, and determining the image recognition frequency according to the segmentation step length;
extracting images from the video to be processed based on the image reading frequency, and synchronously determining the reference difficulty according to the image recognition frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
As a further scheme of the invention: the step of selecting the sub-virtual machine according to the reference difficulty, sending the image to the sub-virtual machine for identification, and updating the image identification frequency according to the identification result comprises the following steps:
selecting a sub-virtual machine according to the reference difficulty, and transmitting the extracted image to the selected sub-virtual machine;
identifying the image based on the sub-virtual machine, partitioning pixel points in the image, and determining each region;
matching the areas of the adjacent images, and determining the corresponding relation of the areas;
and calculating the change speed of each region based on the corresponding relation, and updating the image recognition frequency according to the change speed.
As a further scheme of the invention: the sub-virtual machine is used for identifying the image, partitioning the pixel points in the image and determining each region, wherein the step of determining each region comprises the following steps:
inquiring a space conversion flow of the image, and extracting a single-value layer in the space conversion flow;
traversing the single-value layer, and calculating the mean value and the mode value;
dividing the value range of the single-value layer according to the mean value and the mode value to obtain a sub-range;
and marking pixel points in the single-value image layer by the sub-range, and determining the area according to the distribution condition of the marked pixel points.
As a further scheme of the invention: the step of marking the pixel points in the single-value graph layer by the sub-range and determining the area according to the distribution condition of the marked pixel points comprises the following steps:
sequentially reading the sub-ranges, traversing pixel points in the single-value image layer, and judging whether the numerical value of the pixel points is contained in the sub-ranges;
if a certain value is included in the sub-range, marking the corresponding pixel point;
calculating the number of pixels with the distance smaller than a preset value by taking each pixel as a center, and connecting the two pixels when the distance is smaller than the preset value;
and circularly executing to obtain the region.
As a further scheme of the invention: the step of calculating the change speed of each region based on the corresponding relation and updating the image recognition frequency according to the change speed comprises the following steps:
calculating the change speed of each region based on the corresponding relation;
calculating the maximum value of the change speed, and comparing the maximum value with a preset speed threshold;
when the maximum value reaches a preset speed threshold value, updating the image recognition frequency according to the difference value between the maximum value and the speed threshold value;
wherein the image recognition frequency is proportional to the difference.
As a further scheme of the invention: the step of acquiring the image processing quantity of each sub-virtual machine in real time and updating the computing power resource quantity occupied by the sub-virtual machine according to the image processing quantity comprises the following steps:
acquiring the processed number of each sub-virtual machine in a preset time period and the number to be processed in the preset time period in real time;
determining the correction proportion of each sub virtual machine according to the processed quantity and the quantity to be processed;
wherein the correction ratio is proportional to the number to be processed and proportional to the number processed.
The technical scheme of the invention also provides a device for distributing and scheduling the bare metal computing power resources of the GPU, which comprises the following components:
the resource segmentation module is used for acquiring the computational power resources of the GPU bare metal server, carrying out gradient segmentation on the computational power resources and constructing a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
the parameter setting module is used for determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency and determining a reference difficulty based on the image recognition frequency;
the parameter updating module is used for selecting the sub-virtual machine according to the reference difficulty, sending the image to the sub-virtual machine for recognition, and updating the image recognition frequency according to the recognition result;
the resource updating module is used for acquiring the image processing quantity of each sub-virtual machine in real time and updating the computing power resource quantity occupied by the sub-virtual machines according to the image processing quantity;
the method for obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource, and constructing the content of the sub-virtual machine set comprises the following steps:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub virtual machine with the smallest accumulated result.
As a further scheme of the invention: the parameter setting module comprises:
the inquiring unit is used for acquiring the type of the video to be processed and inquiring the corresponding sub-virtual machine set according to the type;
the first frequency determining unit is used for reading the total segmentation number corresponding to the sub-virtual machine set and determining the image reading frequency according to the total segmentation number;
the second frequency determining unit is used for reading the segmentation step length corresponding to the sub-virtual machine set and determining the image recognition frequency according to the segmentation step length;
the frequency application unit is used for extracting images from the video to be processed based on the image reading frequency and synchronously determining the reference difficulty according to the image identification frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
The technical scheme of the invention also provides a storage medium, at least one program code is stored in the storage medium, and when the program code is loaded and executed by a processor, the method for distributing and scheduling the bare metal computing power resources of the GPU is realized.
Compared with the prior art, the invention has the beneficial effects that: according to the invention, the computing power resources are subjected to gradient segmentation to obtain a plurality of sub virtual machines, when the images are received, the change conditions in the images are analyzed, and then, which sub virtual machine is used for processing the subsequent images is determined, so that different images are processed by equipment with different capabilities, and the resource utilization rate is greatly improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following description will briefly introduce the drawings that are needed in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present invention.
FIG. 1 is a block flow diagram of a method for scheduling allocation of bare metal computing resources for a GPU.
Fig. 2 is a block diagram of the composition and structure of a GPU bare metal computing power resource allocation and scheduling device.
Detailed Description
In order to make the technical problems, technical schemes and beneficial effects to be solved more clear, the invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flow chart of a GPU bare metal computing power resource allocation scheduling method, and in an embodiment of the present invention, a GPU bare metal computing power resource allocation scheduling method is provided, and the method includes:
step S100: acquiring computing power resources of a GPU bare metal server, and carrying out gradient segmentation on the computing power resources to construct a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
bare metal computing power is a high performance computing capability provided by physical servers. The method directly utilizes hardware resources of a physical server, such as a CPU, a GPU, a memory and the like, and provides efficient, stable and reliable computing service for users; the GPU bare metal server has limited computational resources, the faced data are graphics data, the content of the graphics data is single, the content of the graphics data is complex, the processing difficulties are different, and the graphics data are possibly contained in the same video, so that the single GPU can face graphics data with different processing difficulties, in order to ensure the processing effect, the GPU performance must take the most complex graphics data as a reference, that is, the GPU must be capable of identifying the most complex graphics data, so that a part of resources are not utilized when the GPU faces simple graphics data, therefore, an allocation scheme appears in the prior art, that is, the computational resources of the GPU bare metal server are allocated, different image data are processed, the processing process is optimized, and the resource utilization rate is improved.
Step S200: determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency, and determining a reference difficulty based on the image recognition frequency;
analyzing the computing power resource of the GPU bare metal server, and determining an image reading frequency and an image recognition frequency, wherein the image reading frequency is the frequency of the GPU for reading images, reflects how many images are read for recognition in one second, is the inherent data of the GPU, and is higher in performance; the image recognition frequency determines how long to recognize an image once, for example, twice a second, which is smaller than the image reading frequency, and its actual meaning is how long to recognize a read image once.
And reading the image based on the image reading frequency, and then identifying the image based on the image identification frequency, wherein the image identification frequency affects another parameter, namely the reference difficulty, and determining which sub-virtual machine the image is delivered to for processing by the reference difficulty.
Step S300: selecting a sub-virtual machine according to the reference difficulty, transmitting an image to the sub-virtual machine for identification, and updating the image identification frequency according to the identification result;
selecting a sub-virtual machine according to the reference difficulty, and after the sub-virtual machine is selected, transmitting the subsequently read image to the corresponding sub-virtual machine and then identifying; on the basis, the recognition result is obtained in real time, if the image is detected to be complex, the image recognition frequency is increased, the corresponding reference difficulty is also increased, and when the image is distributed, the sub-virtual machines with more resources are required to be queried.
Step S400: acquiring the image processing quantity of each sub-virtual machine in real time, and updating the computing power resource quantity occupied by the sub-virtual machines according to the image processing quantity;
when each sub-virtual machine processes the image, the image processing number is acquired, and the amount of computing power resources occupied by each sub-virtual machine is updated based on the image processing number, which is essentially the allocation process of the adjustment step S100.
The image recognition frequency and the working process of the sub-virtual machine are in a recursive relation, the image recognition frequency influences the selection process of the sub-virtual machine, the selected working process of the sub-virtual machine feeds back and adjusts the image recognition frequency, the image recognition frequency is related to the image change condition, the image recognition frequency is in direct proportion to the change speed of the area with the largest change in the image, and the follow-up content is specifically referred to.
As a preferred embodiment of the technical scheme of the present invention, the step of obtaining the computing power resource of the GPU bare metal server, performing gradient segmentation on the computing power resource, and constructing the sub-virtual machine set includes:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub virtual machine with the smallest accumulated result.
In an example of the technical scheme of the invention, describing the creation process of the sub-virtual machine, constructing an arithmetic series, wherein the number of terms and the step length of the arithmetic series are randomly determined, and the limiting condition is that the sum of the values of each term is one, and at the moment, the value of each term represents the ratio of the value of each term in the total resource quantity; different term numbers and step sizes are randomly determined, a plurality of arithmetic series can be obtained, each arithmetic series corresponds to a segmentation mode, and a sub-virtual machine set is further obtained.
Then, various types of video data are queried based on a big data technology, the video data are existing data, the video data are used as test data in the technical scheme of the application, the test data are input into each segmentation mode for testing, and the segmentation mode which is most suitable for more types can be determined according to the total identification duration, namely, sub-virtual units of various types of video data are determined.
As a preferred embodiment of the present invention, the step of determining the image reading frequency and the image recognition frequency according to the computing power resource, reading the image based on the image reading frequency, and determining the reference difficulty based on the image recognition frequency includes:
acquiring the type of the video to be processed, and inquiring the corresponding sub-virtual machine set according to the type;
reading the total segmentation number corresponding to the sub-virtual machine set, and determining the image reading frequency according to the total segmentation number;
reading the segmentation step length corresponding to the sub-virtual machine set, and determining the image recognition frequency according to the segmentation step length;
extracting images from the video to be processed based on the image reading frequency, and synchronously determining the reference difficulty according to the image recognition frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
The above-mentioned content defines step S200, which is an application process, when the video to be processed is faced, the type of the video to be processed is obtained, according to the type, sub-virtual units can be queried, each sub-virtual unit corresponds to a segmentation total number and a segmentation step length, the more the segmentation total number is, the more the related sub-virtual machines are, the corresponding image reading frequency can be higher, the more the segmentation step length is, the smaller the difference between adjacent sub-virtual machines is, at this time, the higher image recognition frequency is required, the images are continuously recognized, and a more accurate image distribution mode (which sub-virtual machine processes the images) is determined.
As a preferred embodiment of the technical scheme of the present invention, the step of selecting the sub-virtual machine according to the reference difficulty, sending the image to the sub-virtual machine for recognition, and updating the image recognition frequency according to the recognition result comprises:
selecting a sub-virtual machine according to the reference difficulty, and transmitting the extracted image to the selected sub-virtual machine;
identifying the image based on the sub-virtual machine, partitioning pixel points in the image, and determining each region;
matching the areas of the adjacent images, and determining the corresponding relation of the areas;
and calculating the change speed of each region based on the corresponding relation, and updating the image recognition frequency according to the change speed.
In one example of the technical scheme of the invention, the scheme of identifying the image by the sub-virtual machine is that each pixel point in the image is traversed, and the pixel points are partitioned according to the values of the pixel points, so that different areas are determined; in a continuous video, almost no large change occurs between adjacent images (if the situation of complete non-correspondence occurs, the video is indicated to have transition in the corresponding frame, at the moment, the change speed is the maximum value), the corresponding relation of each region can be determined by comparing the adjacent images, namely, the states of the same region in two images, the displacement of the center point of the same region is obtained, the change speeds of each region can be calculated, all the change speeds are counted, and the image recognition frequency is updated by the change speeds; wherein, the higher the change speed, the higher the image recognition frequency.
As a preferred embodiment of the present invention, the step of identifying the image based on the sub-virtual machine, partitioning the pixels in the image, and determining each region includes:
inquiring a space conversion flow of the image, and extracting a single-value layer in the space conversion flow;
traversing the single-value layer, and calculating the mean value and the mode value;
dividing the value range of the single-value layer according to the mean value and the mode value to obtain a sub-range;
and marking pixel points in the single-value image layer by the sub-range, and determining the area according to the distribution condition of the marked pixel points.
The above description specifically describes the image partitioning process, firstly, when the GPU processes the image, the GPU performs spatial conversion on the image, for example, converts the RGB image into the HSV image, which is an essential preprocessing process, reads the preprocessing result of the GPU on the image, reads a certain layer, and uses the value of a pixel point in the certain layer as a subsequent analysis target, so that the pixel point is referred to as a single-value layer.
Traversing the single-value layer, calculating a mean value and a mode value, and primarily estimating the distribution uniformity of the values of the pixel points according to the mean value and the mode value, wherein the closer the mean value and the mode value are, the more the data are, the larger the sub-range can be selected at the moment; the meaning of the sub-range is that the single value range is [0, 255], the sub-range can be [0, 10), [10, 20), etc., and can also be [0, 20), [20, 40), etc., which are related to the difference between the mean value and the mode, that is, the span of the sub-range is inversely proportional to the difference.
In an example of the present invention, the step of marking the pixel points in the single-value layer by the sub-range and determining the area according to the distribution condition of the marked pixel points includes:
sequentially reading the sub-ranges, traversing pixel points in the single-value image layer, and judging whether the numerical value of the pixel points is contained in the sub-ranges;
if a certain value is included in the sub-range, marking the corresponding pixel point;
calculating the number of pixels with the distance smaller than a preset value by taking each pixel as a center, and connecting the two pixels when the distance is smaller than the preset value;
and circularly executing to obtain the region.
As a preferred embodiment of the present invention, the step of calculating a change speed of each region based on the correspondence, and updating the image recognition frequency according to the change speed includes:
calculating the change speed of each region based on the corresponding relation;
calculating the maximum value of the change speed, and comparing the maximum value with a preset speed threshold;
when the maximum value reaches a preset speed threshold value, updating the image recognition frequency according to the difference value between the maximum value and the speed threshold value;
wherein the image recognition frequency is proportional to the difference.
The above description describes the updating process of the image recognition frequency, the parameter affecting the image recognition frequency is the maximum change speed, the larger the maximum change speed is, the larger the difference between the maximum change speed and the preset speed threshold value is, at this time, the description shows that the image contains the region with faster change, the image recognition frequency needs to be improved for the region with faster change, the image recognition frequency is increased, the reference difficulty is increased, and the image is processed by the sub-virtual machine with more resources. Meanwhile, the higher the image recognition frequency is, the more the number of times of judging the change condition of each region is increased, when no region with quicker change exists, the execution main body of the method can rapidly detect the change, and then the image is sent to the sub-virtual machine with less resource amount, so that a manner for rapidly and accurately distributing the image is built.
In an example of the technical solution of the present invention, the step of acquiring the image processing number of each sub-virtual machine in real time and updating the amount of computing power resources occupied by the sub-virtual machine according to the image processing number includes:
acquiring the processed number of each sub-virtual machine in a preset time period and the number to be processed in the preset time period in real time;
determining the correction proportion of each sub virtual machine according to the processed quantity and the quantity to be processed;
wherein the correction ratio is proportional to the number to be processed and proportional to the number processed.
In general, the more the number of to-be-processed is, the more the resource amount of the sub-virtual machine is required, and the application provides a correction scheme, and on the original distribution scheme, the distribution scheme is adjusted according to the number of to-be-processed; meanwhile, the application also considers the processed quantity, and the more the processed quantity is, the longer the actual working time of the sub virtual machine in the working is, the larger the working pressure is, so that the resource quantity is also required to be enlarged.
Fig. 2 is a block diagram of a composition structure of a GPU bare metal computing power resource allocation and scheduling device, and the technical scheme of the present invention further provides a GPU bare metal computing power resource allocation and scheduling device, the device 10 includes:
the resource segmentation module 11 is used for obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource and constructing a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
a parameter setting module 12, configured to determine an image reading frequency and an image recognition frequency according to the computing power resource, read an image based on the image reading frequency, and determine a reference difficulty based on the image recognition frequency;
the parameter updating module 13 is used for selecting a sub-virtual machine according to the reference difficulty, sending the image to the sub-virtual machine for recognition, and updating the image recognition frequency according to the recognition result;
the resource updating module 14 is configured to acquire the image processing number of each sub-virtual machine in real time, and update the amount of computing power resources occupied by the sub-virtual machines according to the image processing number;
the step of obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource and constructing a sub-virtual machine set comprises the following steps:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub virtual machine with the smallest accumulated result.
Further, the parameter setting module 12 includes:
the inquiring unit is used for acquiring the type of the video to be processed and inquiring the corresponding sub-virtual machine set according to the type;
the first frequency determining unit is used for reading the total segmentation number corresponding to the sub-virtual machine set and determining the image reading frequency according to the total segmentation number;
the second frequency determining unit is used for reading the segmentation step length corresponding to the sub-virtual machine set and determining the image recognition frequency according to the segmentation step length;
the frequency application unit is used for extracting images from the video to be processed based on the image reading frequency and synchronously determining the reference difficulty according to the image identification frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
The functions which can be realized by the GPU bare metal computing power resource allocation scheduling method are all completed by computer equipment, the computer equipment comprises one or more processors and one or more memories, at least one program code is stored in the one or more memories, and the program code is loaded and executed by the one or more processors to realize the GPU bare metal computing power resource allocation scheduling method.
The processor takes out instructions from the memory one by one, analyzes the instructions, then completes corresponding operation according to the instruction requirement, generates a series of control commands, enables all parts of the computer to automatically, continuously and cooperatively act to form an organic whole, realizes the input of programs, the input of data, the operation and the output of results, and the arithmetic operation or the logic operation generated in the process is completed by the arithmetic unit; the Memory comprises a Read-Only Memory (ROM) for storing a computer program, and a protection device is arranged outside the Memory.
For example, a computer program may be split into one or more modules, one or more modules stored in memory and executed by a processor to perform the present invention. One or more of the modules may be a series of computer program instruction segments capable of performing specific functions for describing the execution of the computer program in the terminal device.
It will be appreciated by those skilled in the art that the foregoing description of the service device is merely an example and is not meant to be limiting, and may include more or fewer components than the foregoing description, or may combine certain components, or different components, such as may include input-output devices, network access devices, buses, etc.
The processor may be a central processing unit (Central Processing Unit, CPU), other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, which is the control center of the above-described terminal device, and which connects the various parts of the entire worker terminal using various interfaces and lines.
The memory may be used for storing computer programs and/or modules, and the processor may implement various functions of the terminal device by running or executing the computer programs and/or modules stored in the memory and invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as an information acquisition template display function, a product information release function, etc.), and the like; the storage data area may store data created according to the use of the berth status display system (e.g., product information acquisition templates corresponding to different product types, product information required to be released by different product providers, etc.), and so on. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card), at least one disk storage device, flash memory device, or other volatile solid-state storage device.
The modules/units integrated in the terminal device may be stored in a computer readable medium if implemented in the form of software functional units and sold or used as separate products. Based on this understanding, the present invention may implement all or part of the modules/units in the system of the above-described embodiments, or may be implemented by instructing the relevant hardware by a computer program, which may be stored in a computer-readable medium, and which, when executed by a processor, may implement the functions of the respective system embodiments described above. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, executable files or in some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (10)
1. A GPU bare metal computing power resource allocation scheduling method, the method comprising:
acquiring computing power resources of a GPU bare metal server, and carrying out gradient segmentation on the computing power resources to construct a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency, and determining a reference difficulty based on the image recognition frequency;
selecting a sub-virtual machine according to the reference difficulty, transmitting an image to the sub-virtual machine for identification, and updating the image identification frequency according to the identification result;
acquiring the image processing quantity of each sub-virtual machine in real time, and updating the computing power resource quantity occupied by the sub-virtual machines according to the image processing quantity;
the step of obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource and constructing a sub-virtual machine set comprises the following steps:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub-virtual machine set with the smallest accumulated result.
2. The method for scheduling GPU bare metal computing power resource allocation according to claim 1, wherein the step of determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency, and determining a reference difficulty based on the image recognition frequency comprises:
acquiring the type of the video to be processed, and inquiring the corresponding sub-virtual machine set according to the type;
reading the total segmentation number corresponding to the sub-virtual machine set, and determining the image reading frequency according to the total segmentation number;
reading the segmentation step length corresponding to the sub-virtual machine set, and determining the image recognition frequency according to the segmentation step length;
extracting images from the video to be processed based on the image reading frequency, and synchronously determining the reference difficulty according to the image recognition frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
3. The method for scheduling GPU bare metal computing power resource allocation according to claim 2, wherein the step of selecting the sub-virtual machine according to the reference difficulty, transmitting the image to the sub-virtual machine for recognition, and updating the image recognition frequency according to the recognition result comprises:
selecting a sub-virtual machine according to the reference difficulty, and transmitting the extracted image to the selected sub-virtual machine;
identifying the image based on the sub-virtual machine, partitioning pixel points in the image, and determining each region;
matching the areas of the adjacent images, and determining the corresponding relation of the areas;
and calculating the change speed of each region based on the corresponding relation, and updating the image recognition frequency according to the change speed.
4. A method for scheduling GPU bare metal computing power resource allocation according to claim 3, wherein the step of identifying the image, partitioning pixels in the image, and determining each region comprises:
inquiring a space conversion flow of the image, and extracting a single-value layer in the space conversion flow;
traversing the single-value layer, and calculating the mean value and the mode value;
dividing the value range of the single-value layer according to the mean value and the mode value to obtain a sub-range;
and marking pixel points in the single-value image layer by the sub-range, and determining the area according to the distribution condition of the marked pixel points.
5. The method for scheduling GPU bare metal computing power resource allocation according to claim 4, wherein the step of marking pixels in a single-valued layer by the sub-range and determining an area according to the distribution of the marked pixels comprises:
sequentially reading the sub-ranges, traversing pixel points in the single-value image layer, and judging whether the numerical value of the pixel points is contained in the sub-ranges;
if a certain value is included in the sub-range, marking the corresponding pixel point;
calculating the number of pixels with the distance smaller than a preset value by taking each pixel as a center, and connecting the two pixels when the distance is smaller than the preset value;
and circularly executing to obtain the region.
6. A method for scheduling GPU bare metal computing power resource allocation according to claim 3, wherein the step of calculating a change speed of each region based on the correspondence relation, and updating the image recognition frequency according to the change speed comprises:
calculating the change speed of each region based on the corresponding relation;
calculating the maximum value of the change speed, and comparing the maximum value with a preset speed threshold;
when the maximum value reaches a preset speed threshold value, updating the image recognition frequency according to the difference value between the maximum value and the speed threshold value;
wherein the image recognition frequency is proportional to the difference.
7. The method for scheduling GPU bare metal computing power resource allocation according to claim 5, wherein the step of acquiring the image processing number of each sub-virtual machine in real time and updating the computing power resource amount occupied by the sub-virtual machine according to the image processing number comprises:
acquiring the processed number of each sub-virtual machine in a preset time period and the number to be processed in the preset time period in real time;
determining the correction proportion of each sub virtual machine according to the processed quantity and the quantity to be processed;
wherein the correction ratio is proportional to the number to be processed and proportional to the number processed.
8. A GPU bare metal computing power resource allocation scheduling device, the device comprising:
the resource segmentation module is used for acquiring the computational power resources of the GPU bare metal server, carrying out gradient segmentation on the computational power resources and constructing a sub virtual machine set; wherein, the computing power resource of the obtained sub virtual machine set accords with an arithmetic progression;
the parameter setting module is used for determining an image reading frequency and an image recognition frequency according to the computing power resource, reading an image based on the image reading frequency and determining a reference difficulty based on the image recognition frequency;
the parameter updating module is used for selecting the sub-virtual machine according to the reference difficulty, sending the image to the sub-virtual machine for recognition, and updating the image recognition frequency according to the recognition result;
the resource updating module is used for acquiring the image processing quantity of each sub-virtual machine in real time and updating the computing power resource quantity occupied by the sub-virtual machines according to the image processing quantity;
the method for obtaining the computing power resource of the GPU bare metal server, carrying out gradient segmentation on the computing power resource, and constructing the content of the sub-virtual machine set comprises the following steps:
randomly determining the total segmentation number and the segmentation step length, and carrying out gradient segmentation on the computing power resource according to the total segmentation number and the segmentation step length to construct a plurality of sub-virtual units;
obtaining video data of each type, converting the video data into an image sequence, adopting a random transmission mode to transmit the image to a sub-virtual machine in the sub-virtual machine set, and accumulating processing time;
and comparing the accumulated processing time length, and selecting and outputting the sub virtual machine with the smallest accumulated result.
9. The GPU bare metal computing power resource allocation scheduling device of claim 8, wherein the parameter setting module comprises:
the inquiring unit is used for acquiring the type of the video to be processed and inquiring the corresponding sub-virtual machine set according to the type;
the first frequency determining unit is used for reading the total segmentation number corresponding to the sub-virtual machine set and determining the image reading frequency according to the total segmentation number;
the second frequency determining unit is used for reading the segmentation step length corresponding to the sub-virtual machine set and determining the image recognition frequency according to the segmentation step length;
the frequency application unit is used for extracting images from the video to be processed based on the image reading frequency and synchronously determining the reference difficulty according to the image identification frequency; the relation between the image recognition frequency and the reference difficulty is preset by a worker.
10. A storage medium having stored therein at least one program code which, when loaded and executed by a processor, implements a GPU bare metal computing resource allocation scheduling method according to any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410108130.0A CN117636137B (en) | 2024-01-26 | 2024-01-26 | GPU bare metal computing power resource allocation scheduling method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410108130.0A CN117636137B (en) | 2024-01-26 | 2024-01-26 | GPU bare metal computing power resource allocation scheduling method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117636137A true CN117636137A (en) | 2024-03-01 |
| CN117636137B CN117636137B (en) | 2024-04-02 |
Family
ID=90035923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410108130.0A Active CN117636137B (en) | 2024-01-26 | 2024-01-26 | GPU bare metal computing power resource allocation scheduling method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117636137B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110764901A (en) * | 2019-09-17 | 2020-02-07 | 阿里巴巴集团控股有限公司 | Data processing method based on GPU (graphics processing Unit) resources, electronic equipment and system |
| WO2022001086A1 (en) * | 2020-06-29 | 2022-01-06 | 苏州浪潮智能科技有限公司 | Efficient gpu resource allocation optimization method and system |
| CN115454636A (en) * | 2022-09-16 | 2022-12-09 | 城云科技（中国）有限公司 | Container cloud platform GPU resource scheduling method, device and application |
| CN115564635A (en) * | 2022-09-30 | 2023-01-03 | 杭州海康威视系统技术有限公司 | GPU resource scheduling method and device, electronic equipment and storage medium |
| CN115617364A (en) * | 2022-12-20 | 2023-01-17 | 中化现代农业有限公司 | GPU virtualization deployment method, system, computer equipment and storage medium |
| CN115951999A (en) * | 2023-01-05 | 2023-04-11 | 湖南大学 | Multi-display-card-based fragmentation virtualization load balancing scheduling in virtual domain |
| CN116185554A (en) * | 2021-11-29 | 2023-05-30 | 华为技术有限公司 | Configuration device, scheduling device, configuration method and scheduling method |
| CN116757915A (en) * | 2023-08-16 | 2023-09-15 | 北京蓝耘科技股份有限公司 | Cluster GPU resource scheduling method |
-
2024
- 2024-01-26 CN CN202410108130.0A patent/CN117636137B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110764901A (en) * | 2019-09-17 | 2020-02-07 | 阿里巴巴集团控股有限公司 | Data processing method based on GPU (graphics processing Unit) resources, electronic equipment and system |
| WO2022001086A1 (en) * | 2020-06-29 | 2022-01-06 | 苏州浪潮智能科技有限公司 | Efficient gpu resource allocation optimization method and system |
| CN116185554A (en) * | 2021-11-29 | 2023-05-30 | 华为技术有限公司 | Configuration device, scheduling device, configuration method and scheduling method |
| CN115454636A (en) * | 2022-09-16 | 2022-12-09 | 城云科技（中国）有限公司 | Container cloud platform GPU resource scheduling method, device and application |
| CN115564635A (en) * | 2022-09-30 | 2023-01-03 | 杭州海康威视系统技术有限公司 | GPU resource scheduling method and device, electronic equipment and storage medium |
| CN115617364A (en) * | 2022-12-20 | 2023-01-17 | 中化现代农业有限公司 | GPU virtualization deployment method, system, computer equipment and storage medium |
| CN115951999A (en) * | 2023-01-05 | 2023-04-11 | 湖南大学 | Multi-display-card-based fragmentation virtualization load balancing scheduling in virtual domain |
| CN116757915A (en) * | 2023-08-16 | 2023-09-15 | 北京蓝耘科技股份有限公司 | Cluster GPU resource scheduling method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117636137B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chin et al. | Guaranteed outlier removal with mixed integer linear programs | |
| CN108510499B (en) | Image threshold segmentation method and device based on fuzzy set and Otsu | |
| CN115037766B (en) | Industrial equipment Internet of things data acquisition method and device | |
| CN110490204B (en) | Image processing method, image processing device and terminal | |
| CN111882565B (en) | Image binarization method, device, equipment and storage medium | |
| CN115730605B (en) | Data analysis method based on multidimensional information | |
| CN114862929A (en) | Three-dimensional target detection method and device, computer readable storage medium and robot | |
| CN114359048A (en) | Image data enhancement method and device, terminal equipment and storage medium | |
| CN112966687B (en) | Image segmentation model training method and device and communication equipment | |
| CN117636137B (en) | GPU bare metal computing power resource allocation scheduling method, device and storage medium | |
| CN111199228B (en) | License plate positioning method and device | |
| CN115100541B (en) | Satellite remote sensing data processing method, system and cloud platform | |
| CN115601564B (en) | Colloid contour detection method and system based on image recognition | |
| CN116260970A (en) | Network communication method and system based on data mass transmission | |
| CN115620317A (en) | Method and system for verifying authenticity of electronic engineering document | |
| CN116208465A (en) | Intelligent monitoring and early warning method and system for electric power information communication condition | |
| CN112200004B (en) | Training method and device for image detection model and terminal equipment | |
| CN113344208A (en) | Data reasoning method, device and system | |
| CN114998172A (en) | Image processing method and related system | |
| CN111797922A (en) | Text image classification method and device | |
| CN112347899B (en) | Moving object image extraction method, device, equipment and storage medium | |
| CN117880857B (en) | Cloud computing method, system and storage medium based on mobile network resource management | |
| CN116600105B (en) | Color label extraction method, device, equipment and medium for video material | |
| CN115633321B (en) | Wireless communication network monitoring method and system | |
| CN111768410B (en) | Image processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |