CN115994023A - Visual intelligent system resource information scheduling method, device, terminal and medium - Google Patents
Visual intelligent system resource information scheduling method, device, terminal and medium Download PDFInfo
- Publication number
- CN115994023A CN115994023A CN202310290470.5A CN202310290470A CN115994023A CN 115994023 A CN115994023 A CN 115994023A CN 202310290470 A CN202310290470 A CN 202310290470A CN 115994023 A CN115994023 A CN 115994023A
- Authority
- CN
- China
- Prior art keywords
- information
- resource
- task
- video
- resource information
- 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 65
- 230000000007 visual effect Effects 0.000 title claims abstract description 60
- 238000012545 processing Methods 0.000 claims abstract description 133
- 238000013139 quantization Methods 0.000 claims abstract description 60
- 238000004422 calculation algorithm Methods 0.000 claims description 86
- 238000004364 calculation method Methods 0.000 claims description 25
- 238000004458 analytical method Methods 0.000 abstract description 17
- 238000007726 management method Methods 0.000 description 17
- 230000008569 process Effects 0.000 description 11
- 238000004590 computer program Methods 0.000 description 6
- 230000001133 acceleration Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 102100030148 Integrator complex subunit 8 Human genes 0.000 description 4
- 101710092891 Integrator complex subunit 8 Proteins 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000000630 rising effect Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007430 reference method Methods 0.000 description 2
- 240000008574 Capsicum frutescens Species 0.000 description 1
- 101710092887 Integrator complex subunit 4 Proteins 0.000 description 1
- 101100317378 Mus musculus Wnt3 gene Proteins 0.000 description 1
- 102100037075 Proto-oncogene Wnt-3 Human genes 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 239000003999 initiator Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
Images
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
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a resource information scheduling method, a device, a terminal and a medium of a visual intelligent system, which comprise the following steps: acquiring equipment resource information of video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool; carrying out structuring and quantization processing on the acquired task information, and carrying out task scheduling according to the task information after structuring and quantization processing and equipment resource information in a resource pool; updating the resource information of each video computing device according to the task running condition, and updating the resource pool and the information of the hollow operation slot in the resource pool; the invention carries out structuring and quantization processing on the resource demand information in the video analysis task and the related information of the computing resource in the visual intelligent system, reduces the complexity of the system on task and resource management and scheduling, and can be suitable for the scheduling task of the visual intelligent system with high computing performance and heterogeneous computing requirements.
Description
Technical Field
The invention relates to the technical field of video and image processing, in particular to a method, a device, a terminal and a medium for scheduling resource information of a visual intelligent system.
Background
Visual intelligence systems typically involve three parts, a video source, a video processing unit, and an algorithm computing unit when processing video analysis tasks. The video source typically includes raw video data collected by an image sensor, or real-time compressed video collected by a camera, or an offline compressed video file stored in a storage server. The video processing unit is provided with a hardware acceleration module for video decoding or encoding, is mainly responsible for encoding and decoding of video, and can also perform the works of scaling, format conversion and the like on the video. The algorithm calculating unit is provided with a calculation acceleration module and is mainly responsible for loading and running of the algorithm. A typical video analysis task is generally implemented by processing collected video data by a video processing unit, and sending the processed video data to an algorithm computing unit through a high-speed channel inside the device for intelligent analysis. In a vision-based intelligent system of a certain scale, a wide variety of devices may be involved. Some devices may contain both a video source, a video processing unit and an algorithm computing unit, such as a smart camera with an AI acceleration unit, while some devices may contain only one or two parts of them, such as a conventional IPC camera that can only be used as a video source, a commercially available intel box of the intel @ Jetson contains a video processing unit and an algorithm computing unit.
The visual intelligence system also typically includes at least one scheduling unit to effect scheduling of the plurality of tasks among the plurality of video computing devices for resource allocation in a manner that is as rational as possible.
At present, the computing resource management in the market mainly uses a virtualization clustering method, but has a certain limitation on a visual intelligent system with higher performance requirements. Mainly has the following aspects:
(1) The hardware acceleration cores of the different devices in the system may be heterogeneous due to business or cost requirements, such as a GPU server in a system that may include inflight, and a rising AI server in a system. However, algorithms that can be run on an inflight GPU typically run directly on the AI platform of huawa, which presents a certain difficulty. There is a limit to managing heterogeneous resources using virtualization clustering techniques.
(2) For some embedded computing devices, some cannot effectively use virtualized clustered resource management techniques such as containers due to device resources and the like.
(3) Most of the resource management schemes currently on the market divide computing resources into static job slots with standard amounts of resources. In video tasks, the resources required for video processing and the resources required for algorithm computation are not in a fixed proportion, for example, 1080p decoding and 10T computing power are required for some tasks, and 4K decoding and 1T computing power are required for some tasks, so it is difficult to divide the job slot using a static standard.
(4) In a video analysis task with higher performance requirements, video data needs to be transmitted from a video processing unit to an algorithm computing unit through a high-speed channel inside the device, and a certain coupling relationship exists between the video processing unit and the algorithm computing unit, if a video processing part is distributed to the device 1 and an algorithm computing part is distributed to the device 2, the performance of the video analysis task can be greatly reduced. This is to be avoided by the system.
(5) The accumulation of resources in number does not necessarily meet the task requirement, for example, one video analysis task needs to decode one 4K video, but the maximum decoding capability of each video processing unit in the system is 1080P, and even thousands of such processing units cannot meet the requirement of decoding 4K video. This requires consideration of the maximum processing power per video processing unit.
Typically, visual intelligence systems typically encounter the following problems when scheduling task resources:
the visual intelligent system resource mainly comprises a video processing resource and an algorithm computing resource, and the video analysis task information mainly comprises three factor information of video source information, algorithm information and task configuration information. The main resources consumed in the task are a video processing part and an algorithm calculating part. Resource scheduling is ultimately the coordination of tasks and resources.
The relationship between tasks and resources has a certain complexity, and specifically has the following aspects:
(1) The video source information, the algorithm information and the task configuration information have inherent properties which are not coupled with each other, and once a video source is selected by a task, only video specification options provided by the video source can be selected, but the specification properties of the video source cannot be changed. Similarly, after the algorithm is selected, the task cannot change the running environment, resource consumption and other information of the algorithm, the task configuration information is configured by a user according to the requirement, and the user does not care about the specification attribute of the video source and the algorithm resource consumption information. So the three information have certain independence and are combined together to form the main part of the task information.
(2) The video source is provided by a video acquisition device or a video file, and the output video specification parameters are original video parameters, such as original resolution and original frame rate. The amount of resources consumed by video processing is in direct proportion to the resolution and frame rate of the video, and the higher the resolution or the higher the frame rate, the larger the amount of resources occupied by video processing, and conversely, the smaller the amount of resources occupied by video processing. Video capture devices, such as surveillance cameras, typically provide a variety of video formats, resolutions, frame rates for selection, and business users can choose by configuring task parameters. If the video file is a video file, the video format, resolution and frame rate are fixed, the video duration may be different, and the service user may process the packet number and sequence of the video file in batch through task parameter configuration.
(3) Algorithms are provided by algorithm developers and distributed in the system for use by thousands of business users. The operation of the algorithm requires a greater amount of computing power, the greater the number of frames processed per second. So the simple algorithm information is the amount of computational power resources that it consumes at run-time cannot be determined; algorithms run in a system in the form of instances, and an algorithm may run in a system with multiple instances of the algorithm. And how many frames of video images are processed per second by the algorithm instance, this parameter is typically determined by the task, provided in the task configuration parameters, i.e., the task frame rate. Typically different from the original frame rate of the video source. For example, the entrance guard bayonet face recognition task has low requirements on processing video frame rate, and the task frame rate is generally 1FPS or 0.5 FPS. The analysis tasks for monitoring pedestrians and vehicles in real time have higher requirements on the processing frame rate, and the task frame rate usually reaches 5-20 FPS and even requires frame-by-frame processing; on the other hand, the algorithm's demand for video memory is typically relatively fixed, independent of the frame rate.
(4) The task configuration parameters are generated by the service user according to the service scene requirements, and include the task frame rate as described in (3). The video source information, the algorithm information and the task configuration parameter information can not independently complete the expression of the task resource demand information, and only the three information are combined together.
(5) The video sources and algorithms may not be in one-to-one correspondence, and may be one video as a data source of multiple algorithms, where the amount of resources consumed is one video processing resource+multiple algorithm computing resources. It is also possible that multiple video is used as a data source for an algorithm, where the amount of resources consumed by the task is multiple video processing resources+an algorithm computing resource; therefore, when the system manages the resources, the video processing resources and the algorithm computing resources cannot be completely coupled, and certain flexibility needs to be maintained.
(6) The resource information description method may vary from device to device (particularly heterogeneous device); video codec performance is described in specifications such as haisi 3559A as follows:
video decoding:
● Up to H264/H265 8K@30fps or H.264/H.265 4 K@120 fps;
video coding:
● H.264/h.265 multi-stream real-time coding capability:
-7680*4320@30fps+1080P@30fps+ 7680*4320@2fps；
and (5) taking a snapshot.
Whereas the video codec is described in Atlas200 specification as follows:
video decoding capability:
● Support H.264/H.265 Decoder hardware decoding, 20-way 1080P 25FPS,YUV420;
● Support H.264/H.265 Decoder hardware decoding, 16-way 1080P 30FPS,YUV420;
● Support H.264/H.265 Decoder hardware decoding, 2-way 4K 60FPS, YUV420;
● Support H.264/H.265 Encoder hardware coding, 1-way 1080P 30FPS,YUV420;
● JPEG decoding capability 1080P 256FPS, encoding capability 1080P 64FPS, maximum resolution: 8192 4320;
● PNG decoding capability 1080p 24FPS, maximum resolution: 4096 x 2160.
In the two description modes, one is a method description using the maximum coding and decoding resolution, and the other is a method description of how many paths of videos with certain specification are coded and decoded; in addition, the description of the computing power has similar problems, wherein the operating performance under the floating point operation is specified by using FLOPS in the specification of some computing platforms, the computing performance under the INT8 is described by using TOPS (INT 8) for some platforms, and the computing performance under the INT4 is described by using TOPS (INT 4) for some platforms; the non-unification of the description adds a certain complexity to the management and scheduling.
Based on the above points, a reasonable task resource requirement information and resource information expression mode are needed to embody the above relation.
Therefore, the existing virtualization technology is not suitable for resource management and scheduling of a system with high requirements on video analysis performance and multiple heterogeneous computing devices.
Disclosure of Invention
The invention aims to solve the technical problems that the prior art has defects, and provides a method, a device, a terminal and a medium for scheduling resource information of a visual intelligent system, so as to solve the technical problems that the existing scheduling method of the visual intelligent system cannot meet the requirements of large-scale high-performance video analysis and heterogeneous computation.
The technical scheme adopted for solving the technical problems is as follows:
in a first aspect, the present invention provides a method for scheduling resource information of a visual intelligent system, including:
acquiring equipment resource information of video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool;
acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and equipment resource information in the resource pool;
and updating the resource information of each video computing device according to the task running condition, and updating the resource pool and the information of the hollow job slot in the resource pool.
In one implementation, the obtaining device resource information of the video computing device, performing structuring and quantization processing on the device resource information, obtaining resource information of at least one video computing device after structuring and quantization processing, and creating a resource pool includes:
acquiring equipment resource information of video computing equipment, and carrying out structuring and quantization processing on the equipment resource information;
Acquiring resource information after structuring and quantifying of at least one video computing device;
heterogeneous equipment classification is carried out according to the equipment platform information in the equipment resource information;
traversing, for each type of video computing device, resource information for each device;
creating an idle job slot according to the available resource information of the equipment;
and forming idle job slots of all devices in the same class into an idle resource pool of the current class of devices.
In one implementation manner, the forming the idle job slots of all devices in the same class into the idle resource pool of the devices in the current class includes:
the resource pool is managed through the scheduling unit, and the resource information of each idle job slot is dynamically updated and maintained;
and the scheduling unit and the corresponding video computing equipment timely update and maintain the available resource information of the equipment.
In one implementation, the device resource information includes: equipment platform information, algorithm computing resource information, video processing resource information, and available storage information;
the resource information of the idle job slot includes: the available algorithms calculate resource information, available video processing resource information, and available storage information.
In one implementation, the task scheduling according to the task information after the structuring and quantifying processing and the device resource information in the resource pool includes:
determining task resource demand information according to the structured and quantized task information, and matching according to the task resource demand information to obtain equipment types corresponding to the tasks;
video processing capacity, algorithm computing capacity, and storage capacity required for computing tasks;
judging whether an idle job slot meeting the video processing capability, the algorithm computing capability and the storage capability exists in the current class equipment resource pool;
if the idle job slots are met, determining a scheduling strategy, and issuing corresponding tasks to the video computing equipment which is correspondingly met according to the scheduling strategy;
subtracting task resources from available resources in equipment resource information according to task operation information fed back by the corresponding video computing equipment, updating the equipment resource information, and updating the information of the resource pool and the idle job slot.
In one implementation, the determining whether there is a free job slot in the current class of device resource pool that satisfies the video processing capability, the algorithmic computing capability, and the storage capability, then includes:
If no free job slot is satisfied, suspending the task until the free job slot satisfies the task resource requirement;
or comparing the task resource demand information with the resource information of all the idle job slots in the resource pool, modifying task parameters and updating task information according to the comparison result, and returning to the scheduling unit to perform task scheduling again.
In one implementation, the task resource requirement information includes: platform information, video source information, algorithm resource demand information and task configuration information are applicable;
wherein the algorithm resource requirement information includes: calculating power demand information and memory demand information;
the task configuration information includes: task processing frame rate, algorithm operating parameters, and storage requirement information.
In one implementation, the calculation formula of the video processing capability is:
wherein ,
video processing capability;
the calculation formula of the video processing capability is used for calculating the equipment resource information, the task resource demand information and the resource information of the idle job slot;
The calculation formula of the algorithm calculation capability required by the task is as follows:
wherein ,
computing power for an algorithm;
In one implementation, the determining whether there is a free job slot in the current class of device resource pool that satisfies the video processing capability, the algorithm computing capability, and the storage capability includes:
if a plurality of tasks with the same resource requirements are scheduled, calculating the number of paths of the tasks which can be processed by each idle job slot;
according to the number of paths of the tasks which can be processed by each idle job slot, the maximum number of paths of the tasks which can be processed in parallel by the system is estimated by calculation;
and judging whether a plurality of tasks with the same resource requirements exist in the resource pool of the equipment in the current category according to the number of paths of the processable tasks of each idle job slot and the number of paths of the maximally parallel processable tasks.
In one implementation, the updating the resource information of each video computing device according to the task running condition includes:
and updating the available computing capacity values, the available computing memory size, the available storage size, the available decoding capacity values and the available encoding capacity values of the resource pool and the idle job slots.
In a second aspect, the present invention provides a resource information scheduling device of a visual intelligent system, including:
the resource pool creation module is used for acquiring equipment resource information of the video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool;
the task scheduling module is used for acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and the equipment resource information in the resource pool;
and the resource updating module is used for updating the resource information of each video computing device according to the task running condition and updating the resource pool and the information of the hollow job slot in the resource pool.
In a third aspect, the present invention provides a terminal comprising: the system comprises a processor and a memory, wherein the memory stores a visual intelligent system resource information scheduling program which is used for realizing the operation of the visual intelligent system resource information scheduling method according to the first aspect when being executed by the processor.
In a fourth aspect, the present invention also provides a medium, where the medium is a computer readable storage medium, where a visual intelligent system resource information scheduler is stored, where the visual intelligent system resource information scheduler is configured to implement the operations of the visual intelligent system resource information scheduling method according to the first aspect when executed by a processor.
The technical scheme adopted by the invention has the following effects:
the invention can establish a resource pool in the scheduling unit according to the equipment resource information by acquiring the equipment resource information of at least one video computing equipment, so as to perform task scheduling according to the task information and the equipment resource information in the resource pool, and update the information of each video computing equipment resource information and the empty job slot in the resource pool according to the scheduling task; the invention carries out structuring and quantization processing on the resource demand information in the video analysis task and the related information of the computing resource in the visual intelligent system, reduces the complexity of the system on task and resource management and scheduling, and can be suitable for the scheduling task of the visual intelligent system with high computing performance and heterogeneous computing requirements.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to the structures shown in these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a visual intelligent system resource information scheduling method in one implementation of the present invention.
FIG. 2 is a scheduling diagram of a scheduling unit in one implementation of the invention.
FIG. 3 is a schematic diagram of scheduling of different device classes in one implementation of the invention.
FIG. 4 is a flow chart of a scheduling unit creating a resource pool in one implementation of the invention.
FIG. 5 is a flow chart of task scheduling in one implementation of the invention.
FIG. 6 is a flow chart of the inspection of a single job slot in one implementation of the present invention.
FIG. 7 is a flow chart of calculating the number of ways each free job slot runs tasks in one implementation of the invention.
Fig. 8 is a functional schematic of a terminal in one implementation of the invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clear and clear, the present invention will be further described in detail below with reference to the accompanying drawings and examples. 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.
Exemplary method
At present, a visual intelligent system with a certain scale on the market mainly adopts a virtualization clustering technology to carry out resource management and scheduling. However, the use of the virtualized clustering technology to manage heterogeneous resources has a certain limitation, and for some embedded computing devices, due to the reasons of device resources and the like, some virtualized clustered resource management technologies such as containers and the like cannot be effectively used, a static standard is difficult to use to partition a job slot, and in a video analysis task with higher performance requirements, the performance of the video analysis task is greatly reduced by a complex scheduling mode, so that the task requirements cannot be met.
In view of the above technical problems, in this embodiment, a method for scheduling resource information of a visual intelligent system is provided, in this embodiment, resource demand information in a video analysis task and related information of computing resources in the visual intelligent system are structured and quantized, so that complexity of the system in task and resource management and scheduling is reduced, and the method is applicable to scheduling tasks of the visual intelligent system with high computing performance and heterogeneous computing requirements.
As shown in fig. 1, an embodiment of the present invention provides a method for scheduling resource information of a visual intelligent system, including the following steps:
Step S100, obtaining equipment resource information of the video computing equipment, carrying out structuring and quantization processing on the equipment resource information, obtaining resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool.
In this embodiment, the method for scheduling resource information of the visual intelligent system is applied to a terminal, where the terminal includes but is not limited to: a computer, etc.
The method provided by the embodiment can effectively solve or improve the problem that the visual intelligent system is scheduling task resources.
The main resources of the visual intelligent system include: video processing resources and algorithm computing resources, video processing capabilities are mainly related to video formats, resolutions and frame rates, and have a certain relative multiple relationship. The algorithm computing power is related to the number of operations per second and the accuracy. The relation provides basis for quantifying video processing resources and algorithm computing resources; in this embodiment, device resource information of the video computing device is structured.
In the embodiment, the video processing resources and the algorithm computing resources are quantized, so that the video processing resources and the algorithm computing resources can be decomposed or combined within a certain range, and the resource fine scheduling is achieved. The quantized resource information, such as total resource quantity, available resource quantity, task resource demand quantity and the like, can be subjected to numerical operation and logic operation, so that the flow and logic judgment in resource scheduling are simplified, and the complexity of resource scheduling is greatly reduced. For example, in the context of intelligent transcoding tasks, it is generally required to process hundreds or thousands of paths of video, such as the video that needs to be transcoded 10000 paths 1080p@30fps, if the system has unified quantization processing on the resource information, the number of paths of video that can be maximally processed in parallel by the system can be obtained according to the available resource amount, for example, the video that can be maximally transcoded 1000 paths 1080p@30fps by the currently available video processing unit resource can be maximally processed in parallel, and the scheduling unit can quickly obtain a scheduling scheme, for example, divide 10000 paths of video into 10 groups, each 1000 paths. If the system does not uniformly quantize the computing resource information, certain complexity exists in scheduling the tasks; moreover, as the complexity of management and scheduling is greatly reduced, the system can conveniently realize dynamic management and scheduling of resources.
In this embodiment, in the process of performing quantization processing on video processing resources and algorithm computing resources, the method includes the following parts:
first, a video processing capability reference standard and a computing capability reference standard are introduced, and the task demand for computing capability, as well as the video processing capability and computing capability of the device, are subjected to quantization processing.
For a quantized description of the video processing capability, for example, taking 1080p@1fps as a video capability reference, the above-mentioned haisi 3559A video H264 decoding capability value is 8k@30fps/1080p@1fps= 480.0, and the above-mentioned Atlas200 video H264 decoding capability value is 16×1080p@30fps/1080p@1fps= 480.0. It can be seen that the H264 decoding capability of the two platforms described above is the same, although in a different manner of description.
For the quantitative description of the computing power, the current industry computing platform generally supports the calculation of INT8 precision, and in this embodiment, 1TOPS (INT 8) is used as a reference index to estimate the computing power value of the computing platform.
And then, describing the information related to the resource scheduling from four parts of equipment, a video source, an algorithm and a task respectively, describing the information by using a unified structuring method, and quantizing the parameters corresponding to the video processing capacity and the computing capacity by using a quantization method.
Finally, after the resources are quantized, the resources can be decomposed or combined within the internal range of the equipment, so that the resource fine management and scheduling are realized.
In this embodiment, in the process of resource fine management and scheduling, a dynamically updated resource pool needs to be created and maintained.
Specifically, in one implementation of the present embodiment, step S100 includes the steps of:
step S101, obtaining equipment resource information of video computing equipment, and carrying out structuring and quantization processing on the equipment resource information;
step S102, obtaining resource information after structuring and quantifying processing of at least one video computing device;
step S103, heterogeneous equipment classification is carried out according to the equipment platform information in the equipment resource information;
step S104, traversing the resource information of each device for each type of video computing device;
step S105, creating an idle job slot according to the available resource information of the equipment;
and S106, forming idle job slots of all the devices of the same class into an idle resource pool of the device of the current class.
In this embodiment, all the same class of video computing devices need to be assembled into a resource pool during the creation of the resource pool. Specifically, after structuring and quantizing the device resource information, obtaining the structured and quantized resource information of at least one video computing device, i.e. the obtained processed resource information may be the structured and quantized resource information of one video computing device, or the structured and quantized resource information of a plurality of video computing devices, or even the structured and quantized resource information of all video computing devices; and then classifying according to the equipment platform information in the equipment resource information and forming a resource pool.
As shown in fig. 3, the resource pool includes: a free resource pool and a used resource pool; the free resource pool is composed of available resources of devices, and the available resources of each device serve as a free job slot, and one job slot can complete one or more tasks. The used resource pool is made up of used resources of the device. Since the available resources and used resources of the device change as the running task increases or decreases, the size of the resources of the free job slot also changes.
Specifically, in one implementation of the present embodiment, step S106 includes the following steps:
step S107, the resource pool is managed by the scheduling unit, and the resource information of each idle job slot is dynamically updated and maintained;
and step S108, the scheduling unit and the corresponding video computing equipment timely update and maintain the available resource information of the equipment.
In this embodiment, the resource pool and the idle job slots therein are uniformly managed mainly by the scheduling unit, and their information is updated and maintained dynamically in time. Meanwhile, the scheduling unit and the equipment also need to update and maintain the available resource information of the equipment in time; the scheduling manner of the scheduling unit in this embodiment is shown in fig. 2.
The following describes the resource quantization of the present embodiment through an application scenario:
quantization method F for defining video coding and decoding performance first V The reference method is as follows:
selecting a certain resolution W b xH b @F b fps as a benchmark index, W b 、H b 、F b The width, height, and frame rate of the resolution as a quantization reference for video processing performance, respectively. The maximum performance of a device for encoding and decoding in a certain video format is thatn dev Road w dev xh dev @f dev fps, the device then encodes the capability value p in the video format dev The calculation formula is as follows:
P dev =n dev *w dev *h dev *f dev /(W b *H b *F b )
secondly, defining a quantization method F of computing performance of the computing platform A Different types of computing platform quantization methods F A Perhaps different, INT8 precision computing is currently commonly supported by industry computing platforms. Taking Hua as a rising (HUAWEI Assend) 310 as an example, its calculation force specification is half-precision (FP 16) calculation force c fp16 =8tops, integer precision (INT 8) force is c int8 =16 TOPS，F A The reference method is as follows:
uniformly using 1TOPS (INT 8 precision) as a reference index to convert the computing capability value c of the computing platform dev The specific conversion relation is as follows:
if the calculated force value c under INT8 precision is used int8 Conversion is carried out, c dev =c int8 1TOPS, then Hua is c of the computing platform of the Living (HUAWEI Assnd) 310 dev =16TOPS/1TOPS=16；
If the calculated force value c under the FP16 precision is used fp16 Conversion is carried out, c dev =c fp16 *2/1TOPS, then Hua is c of the computing platform of the Living (HUAWII Assend) 310 dev =8TOPS*2/1TOPS=16；
As shown in fig. 2, a visual intelligent system generally includes at least one scheduling unit, which is responsible for the management and scheduling of resources. The resource management process is as follows:
firstly, the scheduling unit needs to create a resource pool, as shown in fig. 4:
step S1.1: acquiring device resource information A of all video computing devices in a system;
step S1.2: classifying heterogeneous equipment according to the equipment platform information in the equipment resource information A;
step S1.3: all similar video computing devices are formed into a resource pool; traversing all resource information of the similar equipment, and creating an idle resource pool, an idle job slot and an used resource pool according to the available resource information and the used resource information of the equipment. And obtaining the resource information J of each idle job slot.
Step S1.4: the scheduling unit uniformly manages the resource pools and dynamically updates and maintains the information of the resource pools in time, including the resource information J of each idle job slot. Meanwhile, the scheduling unit and the equipment update and maintain the available resource information of the equipment in time.
In the above step, the device resource information a in step S1.1 includes: the device platform information, algorithm computing resource information, video processing resource information and available storage information are shown in the following table 1:
TABLE 1
And carrying out structuring and quantization processing on the device resource information A according to the information description in the device resource information A to obtain the device resource information A after structuring and quantization processing.
The sub information in table 1 contains the following contents:
a) Device platform information; including operating system information used by the device (e.g., ubuntu18.04, windows10, etc.), computing acceleration platform information (e.g., england GPU, england Jetson, warrior rising, chilly NPU, etc.), necessary driver version information (e.g., jetpack5.0, CUDA7@ 440.33.01, etc.), etc., for distinguishing between different system environments.
b) The algorithm calculates resource information as shown in table 2 below:
TABLE 2
c) Video processing resource information, as shown in table 3 below:
TABLE 3 Table 3
d) The available storage information is shown in table 4 below:
TABLE 4 Table 4
The idle job slot resource information J in step S1.3 includes: the available algorithm calculates resource information, available video processing resource information, and available storage information, as shown in table 5 below:
TABLE 5
In table 5, the sub information contains the following:
a) The available algorithms calculate resource information, as shown in table 6 below:
TABLE 6
b) The available video processing resource information is shown in table 7 below:
TABLE 7
c) The available storage information is shown in table 8 below:
TABLE 8
In the embodiment, the related information in the resource scheduling is decomposed into a plurality of parts of equipment, video, algorithm and task, and each part is mutually decoupled and provided by an independent entity; the device information can be provided by a device manufacturer, the video information can be provided by a video stream or a video file selected by a task, the algorithm resource requirement information can be provided by an algorithm selected by the task, and the task configuration information can be provided by a task initiator according to task requirements.
In the embodiment, different task demands can be flexibly processed by carrying out quantization processing on video processing resources and algorithm computing resources. For example, one video as a data source for multiple algorithms, or multiple videos as a data source for one algorithm, or one task requires 1080p decoding and 10T computing power, while one task requires 4K decoding and 1T computing power. And meanwhile, the resource scheduling of large-scale batch tasks can be greatly simplified.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the method for scheduling resource information of a visual intelligent system further includes the following steps:
step 200, task information is obtained, structuring and quantifying is carried out on the task information, and task scheduling is carried out according to the task information after structuring and quantifying and the equipment resource information in the resource pool.
In this embodiment, after receiving a task, the scheduling unit identifies a device class suitable for running the task according to the task information. The scheduling unit can determine whether enough resources are available to complete the task by comparing the task resource demand and the resource information of the idle job slots of the class; if the resources are enough, the scheduling unit matches the job slots with proper resources according to the task resource demand and issues the tasks. The task resources are occupied, the scheduling unit updates the information of the idle job slots, and the scheduling unit and the equipment update the available resource information; if the resources are not satisfied, the scheduling unit can adjust task parameters according to the task resource demand and the resource information of the idle job slot, so that the task resource demand can be matched with the resource of the idle job slot, and the task can be completed.
Specifically, in one implementation of the present embodiment, step S200 includes the steps of:
step S201, task resource demand information is determined according to the structured and quantized task information, and equipment categories corresponding to tasks are obtained by matching according to the task resource demand information;
step S202, calculating video processing capacity, algorithm computing capacity and storage capacity required by a task;
Step S203, determining whether there is an idle job slot in the current class of device resource pool that satisfies the video processing capability, the algorithm computing capability, and the storage capability.
In one implementation of the present embodiment, step S203 includes the steps of:
step S203a, if a plurality of tasks with the same resource requirement are scheduled, calculating the number of the tasks that can be processed by each idle job slot;
step 203b, according to the number of paths of the tasks that can be processed by each idle job slot, estimating the maximum number of paths of the tasks that can be processed in parallel by the system through calculation;
step 203c, judging whether a plurality of tasks with the same resource requirement exist in the current class equipment resource pool according to the number of paths of the tasks which can be processed by each idle job slot and the number of paths of the tasks which can be processed by the maximum parallel.
Specifically, in one implementation of the present embodiment, step S200 further includes the following steps:
step S204, if a satisfied idle job slot exists, determining a scheduling strategy, and issuing a corresponding task to a video computing device which is correspondingly satisfied according to the scheduling strategy;
step S205, subtracting task resources from available resources in equipment resource information according to task operation information fed back by the corresponding video computing equipment, updating the equipment resource information, and updating the information of the resource pool and the idle job slot.
Step S206, if no free job slot is satisfied, suspending the task until the free job slot satisfies the task resource requirement; or comparing the task resource demand information with the resource information of all the idle job slots in the resource pool, modifying task parameters and updating task information according to the comparison result, and returning to the scheduling unit to perform task scheduling again.
In this embodiment, after receiving a task, the scheduling unit performs structuring and quantization processing on the task information; and searching idle job slots meeting the video processing capacity, algorithm computing capacity and storage capacity of the task from the established resource pool according to the task resource demand information, and then issuing the task to the corresponding executable equipment according to the formulated scheduling scheme to complete the whole scheduling process.
The task scheduling process of the present embodiment is described below by an application scenario, as shown in fig. 5, including:
step S2.1: the system sends task information to a scheduling unit, and the scheduling unit carries out structuring and quantization processing on the task information to obtain task resource demand information B;
step S2.2: matching the equipment category W suitable for running the task according to the applicable platform information B1 in the task resource demand information B; if the matching is successful, executing the step S2.3; if the matching fails, returning failure to the system;
Step S2.3: according to the task resource demand information B, calculating the video processing capacity V required by the task task Algorithm computing power C tesk Storage capacity S task ；
Step S2.4: checking whether all free job slots in the resource pool W have video processing capability V satisfying the calculation in S2.3 task Algorithm computing power C tesk Storage capacity S task The method comprises the steps of carrying out a first treatment on the surface of the If the idle operation slot is satisfied, executing the step S2.5; if no free job slot is satisfied, executing an optional scheduling scheme;
step S2.5: making a scheduling scheme, issuing tasks to equipment, and checking the running state of the equipment; if the operation is successful, executing the step S2.6; if the operation fails, returning failure to the system, and prompting that the resources are unoccupied;
step S2.6: the equipment node returns success to the dispatching node, subtracts the task resource B from the available resource in the equipment resource information A, and updates the equipment resource information A; the scheduling unit also correspondingly updates the information J of the resource pool and the idle job slot.
In the execution process of step S2.1, the task resource requirement information B obtained by the scheduling unit includes: the applicable platform information, video source information, algorithm resource requirement information and task configuration information are shown in the following table 9:
TABLE 9
And carrying out structuring and quantization processing according to the information description in the task resource demand information B to obtain the task resource demand information B after structuring and quantization processing.
In table 9, the sub information contains the following contents:
a) The applicable platform information describes the platform information capable of running the task and corresponds to the 'A1 equipment platform information' in the equipment resource information A.
b) Video data source information, as shown in table 10 below:
table 10
c) The algorithm resource requirement information is shown in the following table 11:
TABLE 11
d) Task configuration information, specifically as shown in table 12 below:
table 12
In the execution of step S2.3, the video processing capability V required by the task is calculated according to the task resource requirement information B task Algorithm computing power C task 。
The calculation formula of the video processing capability is as follows:
wherein ,
video processing capability;
The calculation formula of the video processing capability is used for calculating the equipment resource information, the task resource demand information and the resource information of the idle job slot;
the calculation formula of the algorithm calculation capability required by the task is as follows:
wherein ,
computing power for an algorithm;
In the execution of step S2.4, it is checked whether all the free job slots in the resource pool W have video processing capability V satisfying the task requirements task Algorithm computing power C task The algorithm calculates the memory size vAlg_Cmpmem, the Storage requirement vTask_storage.
As shown in fig. 6, the inspection process for a single job slot is as follows:
step S2.4.1: checking whether available video processing capability information J2 of the idle job slot satisfies video format sVideo FmtName and video processing capability V required by the task task The method comprises the steps of carrying out a first treatment on the surface of the If yes, go to step S2.4.2; if not, execute step S2.4.4;
step S2.4.2: checking whether the available algorithm computing capability information J1 of the idle job slot meets the computing capability requirement information C of the task task And calculating memory demand information vAlg - CmpMem; if yes, go to step S2.4.3; if not, execute step S2.4.4;
step S2.4.3: checking whether available storage information of idle job slots meets storage requirement vTask - Storage; if yes, executing step S2.5; if not, execute step S2.4.4;
step S2.4.4: the prompt resource can not meet the task requirement, and the task scheduling fails.
By traversing all the free job slots in the resource pool W, until a suitable job slot is found to meet the resource requirements of the task.
Steps S2.4.1 to S2.4.3 are directed to a scenario of a single scheduling task, if a plurality of tasks with the same resource requirements are scheduled, i.e. the task resource requirement information B is the same, the number of paths of each idle job slot operation task is calculated first, and then, whether there are idle job slots meeting the video processing capability, the algorithm computing capability and the storage capability of a plurality of tasks with the same resource requirements in the current class of equipment resource pool is judged according to the number of paths of each idle job slot operation task.
As shown in fig. 7, the number of paths of each idle job slot running task is specifically calculated as follows:
from video processingCapability angle, calculating the number n of paths of the same type of task which can be operated by the operation groove v ：
wherein ,
for the remaining available decoding capability values.
From the algorithm computing capability perspective, the number n of paths of the same type of task which can be operated by the job slot is calculated A ：
wherein ,
to be a usable computing power value. />
From the algorithm calculation memory angle, calculating the number n of paths of the same type of task which can be operated by the operation slot M ：
wherein ,
calculating a memory size for the available memory; />
To calculate the memory requirement information.
From the storage perspective, calculating the number n of paths of the job slot capable of running similar tasks S ：
wherein ,
to an available storage size; />
To store the demand information.
Combining the calculation results of the three parts to obtain the road number CH of the same type of task which can be operated on the operation groove task ：
In the executing process of the step S2.4, if there is no satisfied idle job slot, executing an optional scheduling scheme; wherein, the optional scheduling scheme is as follows:
the method comprises the steps that 1, a system modifies task parameters, updates task information and reschedules according to numerical comparison of S2.4;
and (3) selecting a scheduling scheme 2, and suspending the task until an idle job slot meets the task resource requirement.
In the execution process of the step S2.6, the device node returns success to the scheduling node, subtracts the task resource B from the available resource in the device resource information a, and updates the device resource information a. The scheduling unit also correspondingly updates the information J of the resource pool and the idle job slot.
The updated information in the device resource information a includes: the available computing power value vdev_availcmppower, the available computing memory size vdev_availcmpmem, the available memory size vdev_availstorage, the available decoding power value vdavailperf_dec, and the available encoding power value vdavailperf_enc.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the method for scheduling resource information of a visual intelligent system further includes the following steps:
and step S300, updating the resource information of each video computing device according to the task running condition, and updating the resource pool and the information of the hollow job slots in the resource pool.
In this embodiment, after the task is finished, the system recovers the task resource, and the device updates the device resource information a. The scheduling unit also correspondingly updates the information J of the resource pool and the idle job slot.
According to the method, the device and the system, the video capability reference standard and the calculation accuracy reference standard are introduced, the resource demand information of the task and the equipment resource information are quantized, and the available resource information and the used resource information can be obtained through simple calculation, so that the complexity of resource scheduling is greatly reduced. When the information of the available resources of the system can not meet the demands of task resources, the task resource consumption can be reduced by adjusting the task configuration parameters, so that the system can execute the tasks; in addition, the method in the embodiment creates a dynamic resource pool and a dynamic job slot, the resources are decomposed and combined within a certain range, and efficient scheduling is carried out on the end equipment or the side equipment which cannot be clustered and managed.
The following technical effects are achieved through the technical scheme:
according to the embodiment, by acquiring the equipment resource information of at least one video computing equipment, a resource pool can be created in a scheduling unit according to the equipment resource information, so that task scheduling is performed according to the task information and the equipment resource information in the resource pool, and the information of each video computing equipment resource information and the information of the empty job slot in the resource pool are updated according to the scheduling task; the embodiment carries out structuring and quantization processing on the resource demand information in the video analysis task and the computing resource related information in the visual intelligent system, reduces the complexity of the system on task and resource management and scheduling, and can be suitable for the scheduling task of the visual intelligent system with high computing performance and heterogeneous computing requirements.
Exemplary apparatus
Based on the above embodiment, the present invention further provides a device for scheduling resource information of a visual intelligent system, including:
the resource pool creation module is used for acquiring equipment resource information of the video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool;
The task scheduling module is used for acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and the equipment resource information in the resource pool;
and the resource updating module is used for updating the resource information of each video computing device according to the task running condition and updating the resource pool and the information of the hollow job slot in the resource pool.
Based on the above embodiment, the present invention also provides a terminal, and a functional block diagram thereof may be shown in fig. 8.
The terminal comprises: the system comprises a processor, a memory, an interface, a display screen and a communication module which are connected through a system bus; wherein the processor of the terminal is configured to provide computing and control capabilities; the memory of the terminal comprises a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and computer programs in the storage medium; the interface is used for connecting external equipment such as mobile terminals, computers and other equipment; the display screen is used for displaying corresponding information; the communication module is used for communicating with a cloud server or a mobile terminal.
The computer program is configured to perform the operations of a visual intelligent system resource information scheduling method when executed by a processor.
It will be appreciated by those skilled in the art that the functional block diagram shown in fig. 8 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the terminal to which the present inventive arrangements may be applied, and that a particular terminal may include more or less components than those shown, or may combine some of the components, or have a different arrangement of components.
In one embodiment, a terminal is provided, including: the system comprises a processor and a memory, wherein the memory stores a visual intelligent system resource information scheduling program which is used for realizing the operation of the visual intelligent system resource information scheduling method when being executed by the processor.
In one embodiment, a storage medium is provided, wherein the storage medium stores a visual intelligent system resource information scheduler, which when executed by a processor is operable to implement the operations of the visual intelligent system resource information scheduling method as above.
Those skilled in the art will appreciate that implementing all or part of the above-described methods may be accomplished by way of a computer program comprising instructions for the relevant hardware, the computer program being stored on a non-volatile storage medium, the computer program when executed comprising the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory.
In summary, the invention provides a method, a device, a terminal and a medium for scheduling resource information of a visual intelligent system, wherein the method comprises the following steps: acquiring equipment resource information of video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool; acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and equipment resource information in a resource pool; updating the resource information of each video computing device according to the task running condition, and updating the resource pool and the information of the hollow operation slot in the resource pool; the invention carries out structuring and quantization processing on the resource demand information in the video analysis task and the related information of the computing resource in the visual intelligent system, reduces the complexity of the system on task and resource management and scheduling, and can be suitable for the scheduling task of the visual intelligent system with high computing performance and heterogeneous computing requirements.
It is to be understood that the invention is not limited in its application to the examples described above, but is capable of modification and variation in light of the above teachings by those skilled in the art, and that all such modifications and variations are intended to be included within the scope of the appended claims.
Claims (13)
1. The resource information scheduling method of the visual intelligent system is characterized by comprising the following steps of:
acquiring equipment resource information of video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool;
acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and equipment resource information in the resource pool;
and updating the resource information of each video computing device according to the task running condition, and updating the resource pool and the information of the hollow job slot in the resource pool.
2. The method for scheduling resource information of a visual intelligent system according to claim 1, wherein the steps of obtaining device resource information of a video computing device, performing structuring and quantization processing on the device resource information, obtaining resource information of at least one video computing device after structuring and quantization processing, and creating a resource pool include:
acquiring equipment resource information of video computing equipment, and carrying out structuring and quantization processing on the equipment resource information;
Acquiring resource information after structuring and quantifying of at least one video computing device;
heterogeneous equipment classification is carried out according to the equipment platform information in the equipment resource information;
traversing, for each type of video computing device, resource information for each device;
creating an idle job slot according to the available resource information of the equipment;
and forming idle job slots of all devices in the same class into an idle resource pool of the current class of devices.
3. The method for scheduling resource information of a visual intelligent system according to claim 2, wherein the step of forming the free job slots of all devices of the same class into the free resource pool of the devices of the current class comprises the steps of:
the resource pool is managed through the scheduling unit, and the resource information of each idle job slot is dynamically updated and maintained;
and the scheduling unit and the corresponding video computing equipment timely update and maintain the available resource information of the equipment.
4. A visual intelligent system resource information scheduling method according to claim 3, wherein the device resource information comprises: equipment platform information, algorithm computing resource information, video processing resource information, and available storage information;
The resource information of the idle job slot includes: the available algorithms calculate resource information, available video processing resource information, and available storage information.
5. The method for scheduling resource information of a visual intelligent system according to claim 1, wherein the task scheduling according to the structured and quantized task information and the device resource information in the resource pool comprises:
determining task resource demand information according to the structured and quantized task information, and matching according to the task resource demand information to obtain equipment types corresponding to the tasks;
video processing capacity, algorithm computing capacity, and storage capacity required for computing tasks;
judging whether an idle job slot meeting the video processing capability, the algorithm computing capability and the storage capability exists in the current class equipment resource pool;
if the idle job slots are met, determining a scheduling strategy, and issuing corresponding tasks to the video computing equipment which is correspondingly met according to the scheduling strategy;
subtracting task resources from available resources in equipment resource information according to task operation information fed back by the corresponding video computing equipment, updating the equipment resource information, and updating the information of the resource pool and the idle job slot.
6. The visual intelligent system resource information scheduling method according to claim 5, wherein the determining whether there is a free job slot in the current class of device resource pool that satisfies the video processing capability, the algorithmic computing capability, and the storage capability comprises:
if no free job slot is satisfied, suspending the task until the free job slot satisfies the task resource requirement;
or comparing the task resource demand information with the resource information of all the idle job slots in the resource pool, modifying task parameters and updating task information according to the comparison result, and returning to the scheduling unit to perform task scheduling again.
7. The visual intelligent system resource information scheduling method according to claim 5, wherein the task resource requirement information comprises: platform information, video source information, algorithm resource demand information and task configuration information are applicable;
wherein the algorithm resource requirement information includes: calculating power demand information and memory demand information;
the task configuration information includes: task processing frame rate, algorithm operating parameters, and storage requirement information.
8. The visual intelligent system resource information scheduling method according to claim 5, wherein the calculation formula of the video processing capability is:
wherein ,
video processing capability;
the calculation formula of the video processing capability is used for calculating the equipment resource information, the task resource demand information and the resource information of the idle job slot;
the calculation formula of the algorithm calculation capability required by the task is as follows:
wherein ,
computing power for an algorithm;
9. The method for scheduling visual intelligent system resource information according to claim 5, wherein determining whether there is a free job slot in the current class of device resource pool that satisfies the video processing capability, the algorithmic computing capability, and the storage capability comprises:
if a plurality of tasks with the same resource requirements are scheduled, calculating the number of paths of the tasks which can be processed by each idle job slot;
according to the number of paths of the tasks which can be processed by each idle job slot, the maximum number of paths of the tasks which can be processed in parallel by the system is estimated by calculation;
and judging whether a plurality of tasks with the same resource requirements exist in the resource pool of the equipment in the current category according to the number of paths of the processable tasks of each idle job slot and the number of paths of the maximally parallel processable tasks.
10. The method for scheduling resource information of a visual intelligent system according to claim 5, wherein the updating the resource information of each video computing device according to the task operation condition comprises:
and updating the available computing capacity values, the available computing memory size, the available storage size, the available decoding capacity values and the available encoding capacity values of the resource pool and the idle job slots.
11. A visual intelligent system resource information scheduling apparatus, comprising:
the resource pool creation module is used for acquiring equipment resource information of the video computing equipment, carrying out structuring and quantization processing on the equipment resource information, acquiring resource information of at least one video computing equipment after structuring and quantization processing, and creating a resource pool;
the task scheduling module is used for acquiring task information, carrying out structuring and quantization processing on the task information, and carrying out task scheduling according to the task information after structuring and quantization processing and the equipment resource information in the resource pool;
and the resource updating module is used for updating the resource information of each video computing device according to the task running condition and updating the resource pool and the information of the hollow job slot in the resource pool.
12. A terminal, comprising: a processor and a memory storing a visual intelligent system resource information scheduler, which when executed by the processor is operative to implement the visual intelligent system resource information scheduling method of any one of claims 1-10.
13. A medium, characterized in that the medium is a computer readable storage medium, the medium storing a visual intelligent system resource information scheduler, which when executed by a processor is adapted to carry out the operations of the visual intelligent system resource information scheduling method according to any one of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310290470.5A CN115994023B (en) | 2023-03-23 | 2023-03-23 | Visual intelligent system resource information scheduling method, device, terminal and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310290470.5A CN115994023B (en) | 2023-03-23 | 2023-03-23 | Visual intelligent system resource information scheduling method, device, terminal and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115994023A true CN115994023A (en) | 2023-04-21 |
| CN115994023B CN115994023B (en) | 2023-06-06 |
Family
ID=85995269
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310290470.5A Active CN115994023B (en) | 2023-03-23 | 2023-03-23 | Visual intelligent system resource information scheduling method, device, terminal and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115994023B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103338228A (en) * | 2013-05-30 | 2013-10-02 | 江苏大学 | Cloud calculating load balancing scheduling algorithm based on double-weighted least-connection algorithm |
| US9479382B1 (en) * | 2012-03-27 | 2016-10-25 | Amazon Technologies, Inc. | Execution plan generation and scheduling for network-accessible resources |
| CN107291547A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | A kind of task scheduling processing method, apparatus and system |
| CN110928649A (en) * | 2018-09-19 | 2020-03-27 | 北京国双科技有限公司 | Resource scheduling method and device |
| CN111144830A (en) * | 2019-11-20 | 2020-05-12 | 上海泛云信息科技有限公司 | Enterprise-level computing resource management method, system and computer equipment |
| CN111767134A (en) * | 2020-05-18 | 2020-10-13 | 鹏城实验室 | Multitask dynamic resource scheduling method |
| CN115168027A (en) * | 2022-06-15 | 2022-10-11 | 中国科学院沈阳自动化研究所 | Calculation power resource measurement method based on deep reinforcement learning |
-
2023
- 2023-03-23 CN CN202310290470.5A patent/CN115994023B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9479382B1 (en) * | 2012-03-27 | 2016-10-25 | Amazon Technologies, Inc. | Execution plan generation and scheduling for network-accessible resources |
| CN103338228A (en) * | 2013-05-30 | 2013-10-02 | 江苏大学 | Cloud calculating load balancing scheduling algorithm based on double-weighted least-connection algorithm |
| CN107291547A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | A kind of task scheduling processing method, apparatus and system |
| CN110928649A (en) * | 2018-09-19 | 2020-03-27 | 北京国双科技有限公司 | Resource scheduling method and device |
| CN111144830A (en) * | 2019-11-20 | 2020-05-12 | 上海泛云信息科技有限公司 | Enterprise-level computing resource management method, system and computer equipment |
| CN111767134A (en) * | 2020-05-18 | 2020-10-13 | 鹏城实验室 | Multitask dynamic resource scheduling method |
| CN115168027A (en) * | 2022-06-15 | 2022-10-11 | 中国科学院沈阳自动化研究所 | Calculation power resource measurement method based on deep reinforcement learning |
Non-Patent Citations (2)
| Title |
|---|
| A. AZIZ 等: "Grid resource allocation and task scheduling for resource intensive applications", 《2006 INTERNATIONAL CONFERENCE ON PARALLEL PROCESSING WORKSHOPS (ICPPW\'06)》 * |
| 孙振河 等: "网格计算环境下基于任务池的任务调度方法", 《黑龙江大学自然科学学报》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115994023B (en) | 2023-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104885455B (en) | A kind of computer implemented method and device for Video coding | |
| JP5420254B2 (en) | Accelerated video coding | |
| US20220414940A1 (en) | Information processing apparatus and method | |
| US20070220523A1 (en) | Method for operating a real-time multimedia terminal in a qos manner | |
| US10708627B2 (en) | Volumetric video compression with motion history | |
| WO2020237647A1 (en) | Point cloud data encoding method, encoding device, decoding method and decoding device | |
| CN114035937A (en) | Distributed training and reasoning method, system, equipment and readable storage medium based on artificial intelligence | |
| CN108363713A (en) | Video image information resolver, system and method | |
| WO2022121648A1 (en) | Point cloud data encoding method, point cloud data decoding method, device, medium, and program product | |
| KR20230046310A (en) | Signaling of feature map data | |
| Ainala et al. | An improved enhancement layer for octree based point cloud compression with plane projection approximation | |
| CN105592314A (en) | Parallel decoding method and device | |
| CN106162174A (en) | A kind of video multi-resolution encoding method and apparatus | |
| CN115994023B (en) | Visual intelligent system resource information scheduling method, device, terminal and medium | |
| CN115908672A (en) | Three-dimensional scene rendering acceleration method, system, medium, device and terminal | |
| Li et al. | Motion estimation and coding structure for inter-prediction of LiDAR point cloud geometry | |
| CN115984675B (en) | System and method for realizing multipath video decoding and AI intelligent analysis | |
| Gao et al. | IVP: an intelligent video processing architecture for video streaming | |
| Batra | Modeling and efficient optimization for object-based scalability and some related problems | |
| He et al. | MPEG-4 based interactive video using parallel processing | |
| Radicke et al. | Many-core HEVC encoding based on wavefront parallel processing and GPU-accelerated motion estimation | |
| US20230401755A1 (en) | Mesh Compression Using Coding Units with Different Encoding Parameters | |
| EP1296243A2 (en) | A method for operating a real-time multimedia terminal in a QoS manner | |
| WO2023066345A1 (en) | Method, apparatus and medium for point cloud coding | |
| WO2023036316A1 (en) | Method, apparatus, and medium for point cloud coding |
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 |