CN116627631A - Resource scheduling method, system, electronic equipment and storage medium - Google Patents
Resource scheduling method, system, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN116627631A CN116627631A CN202310421301.0A CN202310421301A CN116627631A CN 116627631 A CN116627631 A CN 116627631A CN 202310421301 A CN202310421301 A CN 202310421301A CN 116627631 A CN116627631 A CN 116627631A
- Authority
- CN
- China
- Prior art keywords
- node
- target
- information
- resource
- video
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000003860 storage Methods 0.000 title claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 119
- 238000004458 analytical method Methods 0.000 claims abstract description 61
- 238000005457 optimization Methods 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims description 22
- 238000012544 monitoring process Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 description 31
- 230000006870 function Effects 0.000 description 21
- 238000012545 processing Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 238000007726 management method Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000002360 preparation method Methods 0.000 description 4
- 230000008602 contraction Effects 0.000 description 3
- 239000002071 nanotube Substances 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 210000001503 joint Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000004141 dimensional analysis Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
-
- 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
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of computing power networks, and provides a resource scheduling method, a resource scheduling system, electronic equipment and a storage medium, wherein the resource scheduling method comprises the following steps: acquiring node information of each computing power node; analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table; and scheduling the target video based on the resource recommendation table. According to the method, the resource recommendation table is determined by comprehensively considering the video analysis and the self demand of the power network through the multi-target optimization algorithm, and then the target video is scheduled according to the resource recommendation table so as to analyze the target video, so that the efficiency of video analysis by using the power network is improved.
Description
Technical Field
The present invention relates to the field of computing networks, and in particular, to a resource scheduling method, system, electronic device, and storage medium.
Background
Video plays an important role in various industries. Under various factors, requirements for analysis, identification, AI analysis and the like of videos are increasingly raised in terms of timeliness, real-time performance, computing power resource consumption and the like of various scenes.
The power network nano-tube is used for managing a large amount of power resources, network resources, storage resources and the like, realizing interconnection and intercommunication of various heterogeneous power, providing fine scheduling aiming at task issuing so as to provide low-delay, large-bandwidth and short-distance resource allocation service, reducing task delay and improving service efficiency.
It is desirable to find a way to combine video analysis with a power network to increase the efficiency of video analysis with the power network.
Disclosure of Invention
The invention provides a resource scheduling method, a system, electronic equipment and a storage medium, which are used for solving the problem of low video analysis efficiency, comprehensively considering video analysis tasks and the self demands of a computing power network to determine a resource recommendation table through a multi-target optimization algorithm, and then scheduling target videos according to the resource recommendation table so as to analyze the target videos, thereby improving the efficiency of video analysis by utilizing the computing power network.
The invention provides a resource scheduling method, which comprises the following steps:
acquiring node information of each computing power node;
analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table;
and scheduling the target video based on the resource recommendation table.
In one embodiment, the analyzing the video stream information, the user demand information and the node information of each computing power node by adopting the multi-objective optimization algorithm to determine the resource recommendation table includes:
determining the video stream information, the user demand information and node information of each computing power node, and respectively corresponding weight values;
the video stream information, the user demand information and the node information of each computing power node are subjected to weighted analysis by adopting the multi-objective optimization algorithm, the weight value and the historical recommendation information so as to determine the resource recommendation table;
the resource recommendation table comprises an algorithm power node recommendation table and an AI algorithm recommendation table.
In one embodiment, the scheduling the target video based on the resource recommendation table includes:
determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video;
determining at least one target AI algorithm based on the AI algorithm recommendation table;
and dispatching the target video to the first target computing power node, wherein the first target computing power node performs AI analysis on the target video based on the target AI algorithm.
In one embodiment, the determining at least one first target computing node based on the computing node recommendation table includes:
if the analysis task of the target video is a video training task, determining at least one central cloud node as the first target computing node based on the computing node recommendation table;
and if the analysis task of the target video is a video reasoning task, determining at least one edge cloud node as the first target computing node based on the computing node recommendation table.
In one embodiment, the determining at least one target AI algorithm based on the AI algorithm recommendation table includes:
acquiring an AI service capability registry of each first target computing node;
determining the association relation between the AI service capability registry and the analysis task;
and determining at least one target AI algorithm based on the association relationship and the AI algorithm recommendation table.
In one embodiment, after determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video, the method further includes:
if the capacity of the first target computing node is smaller than a set value, determining at least one second target computing node;
scheduling the target video to the first target computing node and the second target computing node;
wherein the first target computing node is of a different node type than the second target computing node.
In one embodiment, after the multi-objective optimization algorithm is adopted to analyze the video stream information, the user demand information and the node information of each computing power node to determine the resource recommendation table, the method further includes:
monitoring information of each optimization index of each computing node, and displaying information of each optimization index;
and adjusting the scheduling information of the target video based on the information of each optimization index.
The invention also provides a resource scheduling system, which comprises: a resource layer, an algorithm layer and a display layer;
the resource layer is used for collecting node information of each computing power node and sending the node information to the algorithm layer;
the algorithm layer is used for receiving the node information sent by the resource layer, analyzing the video stream information, the user demand information and the node information of each computing node by adopting a multi-objective optimization algorithm to determine a resource recommendation table, and sending the resource recommendation table to the display layer; scheduling a target video based on the resource recommendation table;
the display layer is used for receiving the resource recommendation table and displaying the resource recommendation table.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the resource scheduling method as described in any one of the above when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a resource scheduling method as described in any of the above.
The resource scheduling method, the system, the electronic equipment and the storage medium provided by the invention are characterized in that node information of each computing node is obtained; analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table; and scheduling the target video based on the resource recommendation table. According to the method, the resource recommendation table is determined by comprehensively considering the video analysis task and the self demand of the power network through the multi-target optimization algorithm, and then the target video is scheduled according to the resource recommendation table so as to analyze the target video, so that the efficiency of video analysis by using the power network is improved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a resource scheduling method provided by the present invention;
FIG. 2 is a second flow chart of a resource scheduling method according to the present invention;
FIG. 3 is a schematic diagram of a resource scheduling system according to the present invention;
fig. 4 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The resource scheduling method, system, electronic device and storage medium of the present invention are described below in conjunction with fig. 1-4.
Specifically, the present invention provides a resource scheduling method, and referring to fig. 1, fig. 1 is one of the flow diagrams of the resource scheduling method provided by the present invention.
The resource scheduling method provided by the embodiment of the invention comprises the following steps:
s100, acquiring node information of each computing node;
the computing power node refers to a node capable of providing computing power network resources, and comprises a center cloud, an edge server and the like.
The node information comprises calculation power information, storage information, network information, historical recommendation conditions, utilization rate, load rate, time delay and the like of the calculation power node, and is basic data for resource recommendation.
The execution subject of the embodiment is a resource scheduling system, which comprises an arrangement scheduling subsystem, a multi-cloud management platform and an edge cloud subsystem.
Each computing node is correspondingly provided with at least one interface, and the resource scheduling system acquires node information of the computing node of the current nano tube in real time through the interfaces. For example, the multi-cloud management platform obtains node information such as the type of the current computing node, the resource load and the like through interfaces of the central cloud and the edge cloud.
S200, analyzing video stream information, user demand information and node information of each computing power node by adopting a multi-objective optimization algorithm to determine a resource recommendation table;
it should be noted that the multi-objective optimization algorithm includes a video stream information analysis algorithm, an algorithm node selection algorithm, an AI selection algorithm, and the like.
The computing power node selection algorithm selects a proper computing power node through a user objective function and a system objective function, wherein the functions obtained according to the user side demand indexes such as video streaming delay requirements, computing power requirements, video streaming size and the like are the user objective functions; and according to the demand indexes of the computing power node side, such as the positions of the central cloud and the edge cloud nodes, the functions obtained by the utilization rate, the storage, the load, the time delay and the like are system objective functions.
The AI selection algorithm selects a proper AI algorithm through a video objective function and an AI objective function, wherein the function obtained according to video side requirement indexes such as video type, video industry, video function requirement and the like is the video objective function; and according to the algorithm side demand index, the function obtained by the AI capability and the like of the AI platform is an AI objective function.
And the multi-objective optimization algorithm performs multi-dimensional analysis according to the video data and the user data, screens the computing nodes layer by layer, and determines a resource recommendation table.
For example, the steps for determining the resource recommendation table based on the algorithm force node selection algorithm or the AI selection algorithm are as follows:
(1) Selecting an objective function; the computing power node selection algorithm adopts a user objective function and a system objective function as an objective function; the AI selection algorithm adopts a video objective function and an AI objective function as objective functions;
(2) Performing preliminary screening on the calculation force nodes or the AI algorithm through variation and intersection of the genetic algorithm;
(3) And (3) further precisely screening the computing power nodes or the AI algorithm preliminarily screened in the step (2) through non-dominant sorting to form a preliminary solution.
(4) And (3) sorting the primary solutions obtained in the step (3) in each dimension to form respective resource recommendation tables.
Video stream information including IP address of video acquisition, video access mode, video type, video industry, video function requirement, etc.
The user demand information includes the user's service level agreement (Service Level Agreement, SLA) requirements, artificial intelligence (Artificial Intelligence, AI) capabilities as needed, latency, cost estimation, etc. The user demand information also comprises user characteristics, wherein the user characteristics can be system settings or obtained by the system according to historical data statistics.
And the user submits the requirements on the unified service platform, video stream information and user requirement information, wherein the video stream information such as video IP addresses, video access modes and the like is submitted to the scheduling subsystem.
The resource recommendation table may prioritize the recommended resources according to at least one of video stream information, user demand information, and node information of each computing node.
And S300, scheduling the target video based on the resource recommendation table.
The resource scheduling system sends scheduling instructions to the scheduling subsystem, and the scheduling subsystem schedules the computing power node and the AI algorithm based on the received resource recommendation table after receiving the scheduling instructions.
The scheduling subsystem performs scheduling according to the resource recommendation table, and schedules the target video to the edge cloud node for selection by the scheduling subsystem.
The scheduling subsystem provides basic service capabilities such as power on and network on.
The resource scheduling method provided by the embodiment of the invention obtains the node information of each computing node; analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table; and scheduling the target video based on the resource recommendation table. According to the embodiment of the invention, the resource recommendation table is determined by comprehensively considering the video analysis task and the self demand of the power network through the multi-target optimization algorithm, and then the target video is scheduled according to the resource recommendation table so as to analyze the target video, thereby improving the efficiency of video analysis by using the power network.
Based on the above embodiment, a multi-objective optimization algorithm is adopted to analyze video stream information, user demand information and node information of each computing power node to determine a resource recommendation table, including:
s210, determining video stream information, user demand information and node information of each computing power node, and respectively corresponding weight values;
s220, weighting analysis is carried out on the video stream information, the user demand information and the node information of each computing power node by adopting a multi-objective optimization algorithm, a weight value and historical recommendation information so as to determine a resource recommendation table.
The resource recommendation table comprises an algorithm power node recommendation table and an AI algorithm recommendation table.
Distributing weight values according to the video stream information, the user demand information and the node information of each computing node; the weight value can be obtained according to the weight value distributed by the system statistics history, or the weight value can be preset by the system.
Through a multi-objective optimization algorithm, video reasoning and the self demand of the computing network are comprehensively considered, weighting analysis is carried out on video stream information, user demand information and node information of each computing node to form a resource recommendation table, interaction is carried out with a scheduling subsystem through an interface, and computing nodes are recommended.
The resource scheduling system collects basic data of user demand information, performs data analysis and summarization on video stream information, and performs analysis on video types, sizes, training reasoning modes and the like. And acquiring data of AI capacity of the video AI analysis platform to form a registry. The resource scheduling system analyzes and prefers the selection of the computing nodes and the selection of the AI capacity through a multi-objective optimization algorithm and by combining the historical recommendation condition of the computing nodes, and selects the training/reasoning nodes with low delay and high efficiency and the AI analysis capacity meeting the video requirement so as to form a corresponding computing node recommendation table and an AI algorithm recommendation table.
According to the embodiment of the invention, through adopting the multi-objective optimization algorithm, the weight value and the history recommendation information, the optimization of the computing power node and the optimization of the AI algorithm in the AI analysis process of the video are realized, the requirements of low time delay, low cost, high efficiency and the like of video processing can be met, meanwhile, the computing resources are saved, the existing resources are fully utilized, and the problem of idle computing power is avoided. The embodiment of the invention can improve the user recommendation satisfaction degree, such as by more than 30 percent, by combining the user demands; the video training reasoning efficiency can be improved, for example, the video training reasoning efficiency is improved by more than 40% through real-time scheduling of the video to a proper computing node and reasonable distribution of AI capacity; the average time delay can be reduced, for example, by more than 30 percent; by integrating and reasonably distributing the nano-tube calculation nodes, the resource utilization rate can be improved, for example, the resource utilization rate is improved by more than 50 percent.
Based on the above embodiment, scheduling the target video based on the resource recommendation table includes:
s310, determining at least one first target computing node based on the computing node recommendation table and an analysis task of the target video;
s320, determining at least one target AI algorithm based on the AI algorithm recommendation table;
s330, scheduling the target video to a first target computing node, wherein the first target computing node performs AI analysis on the target video based on the target AI algorithm.
It should be noted that, the analysis task of the target video may be an inference task or a training task.
Determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video; for example, by analyzing the target video, if the analysis task type of the target video is an inference task, one or more first target computing nodes matched with the inference task are preferentially selected in the computing node recommendation table.
Optionally, the orchestration and scheduling subsystem determines one or more first target computing nodes based on the computing node recommendation table, e.g., selects computing nodes with a high level according to a rank ordering of the computing nodes in the computing node recommendation table. The ranking of the power computing nodes in the power computing node recommendation table can rank the recommended power computing nodes in priority according to at least one of video stream information, user demand information and node information of the power computing nodes.
Determining one or more target AI algorithms based on the AI algorithm recommendation table; and calling one or more AI platform interfaces at corresponding positions to prepare resources according to the resource selection condition, wherein the preparation comprises the steps of carrying out service capacity expansion, calculation and resource storage on the bottom k8s platform, carrying out the preparation of related AI service capacity on the AI platform management capacity, and the like. And scheduling the target video to a first target power computing node, wherein the AI platform provides the service capability of the target AI algorithm, and is in butt joint with the target video access platform, and the first target power computing node processes video data based on the target AI algorithm.
The scheduling subsystem calls a network management module, and establishes network connection with the AI capability platform at the corresponding position according to video stream information such as video IP address, video access mode and the like submitted by the user.
According to the embodiment of the invention, the computing force node and the AI algorithm are preferentially selected based on the resource recommendation table, and the video is analyzed through the selected computing force node and AI algorithm, so that the matching time of the video analysis and the computing force node, namely the AI algorithm, is reduced, and the efficiency of video analysis by utilizing the computing force network is improved.
Based on the above embodiment, determining at least one first target computing node based on the computing node recommendation table, comprises:
s311, if the analysis task of the target video is a video training task, determining at least one central cloud node as a first target computing node based on a computing node recommendation table;
s312, if the analysis task of the target video is a video reasoning task, determining at least one edge cloud node as a first target computing node based on the computing node recommendation table.
The video training means that the model continuously optimizes own parameters through training data so as to improve the accuracy of the model. The video training task needs to perform model training based on massive samples, because the model volume and the data volume are large, the requirement on the calculation amount is high, the requirement on time delay and the like is low, large storage resources, multi-machine multi-card graphics processor (Graphics Processing Unit, GPU) calculation resources and the like are generally needed, and the calculation force node required by the video training task is usually a central cloud and is usually high in cost.
If the analysis task of the target video is a video training task, at least one central cloud node is determined in a computing node recommendation table according to storage resources, model types, cost budget and the like required by a user, and the key targets required to be met in the process are computing node satisfaction degree (central processing unit, GPU, storage and the like), user cost budget and the like.
Video reasoning is the model applying the ability to learn from training to work. The video reasoning task approaches to the user side, needs to provide low-delay and stable AI service, and therefore is usually deployed in an edge server, and the selection of a specific edge server needs to be selected in combination with delay, overall load of the server, user cost and the like, and meanwhile, the network type, network throughput and the like from the terminal equipment to the edge server need to be considered.
If the analysis task of the target video is a video reasoning task, at least one edge cloud node is determined in the computing power node recommendation table according to scene types, data throughput and the like, and key targets required to be met in the process comprise computing power node satisfaction degree, time delay, real-time network resource allocation and the like.
Further, for the problem of edge servers and service switching involved in the reasoning scene, the recommendation is mainly based on historical resource conditions of the computing nodes in the embodiment of the invention.
According to the embodiment of the invention, the analysis task of the target video is divided into the video training task and the video reasoning task, and the first target computing node is determined according to different video analysis tasks. According to the embodiment of the invention, according to the user demands of different industries and the difference of the user in the aspects of the demanded software environment, cost, time delay and the like, the user computing task is unloaded to various resources such as a central cloud, an edge cloud or an edge server and the like, so that the coordination of various computing resources is realized.
Based on the above embodiment, determining at least one target AI algorithm based on the AI algorithm recommendation table includes:
s321, acquiring an AI service capability registry of each first target computing node;
s322, determining the association relationship between the AI service capability registry and the analysis task;
s323, determining at least one target AI algorithm based on the association relationship and the AI algorithm recommendation table.
And acquiring data of the AI capacity of the first target computing node to form an AI service capacity registry. The AI-service capability registry refers to AI-service capabilities that the computing node has completed registering, where the computing node cannot use unregistered AI-service capabilities for video analysis. The multi-cloud management platform acquires node information of the computing nodes in real time, and obtains an AI service capability registry of each computing node.
Further, the resource scheduling system obtains the underlying hardware and application registration information of the computing node through the interface.
Determining an association relation between the AI service capability registry and the analysis task, determining at least one target AI algorithm based on the association relation and the AI algorithm recommendation table, for example, determining the association degree between the AI service capability registry and the analysis task, and selecting one or more target AI algorithms with the maximum association degree from the AI algorithm recommendation table according to the association degree, for example, if the analysis task is video reasoning, in the AI service capability registry, the AI algorithm associated with the video reasoning has 1, 2, 3 and 4, and determining one or more algorithms with the maximum association degree in 1, 2, 3 and 4 as the target AI algorithm.
According to the embodiment of the invention, the optimal AI algorithm is found through the association relation between the AI service capability registry and the analysis task, so that the user requirement of video processing can be met, and meanwhile, the calculation cost is saved.
Based on the above embodiment, after determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video, the method further includes:
s313, if the capacity of the first target computing node is smaller than the set value, determining at least one second target computing node;
s314, scheduling the target video to a first target computing node and a second target computing node;
wherein the first target computing node is of a different node type than the second target computing node.
The method comprises the steps of carrying out log presentation and monitoring on a calculation force node recommendation table, optimizing calculation force nodes according to the capacity of the calculation force nodes, and determining at least one second target calculation force node when the capacity of the first target calculation force node is monitored to be smaller than a set value; and scheduling the target video to the first target computing node and the second target computing node.
For example, for the situations of video flow change, multi-video access and the like in the service process, the scheduling subsystem is arranged to interface with the underlying k8s platform and the AI application in real time, so as to perform real-time elastic expansion and contraction of resources.
According to the embodiment of the invention, the computing nodes are increased or reduced according to the capacity change of the computing nodes, and the real-time elastic expansion and contraction of the resources are performed, so that the idle computing power is avoided, and the utilization rate of the resources is improved.
Based on the above embodiment, after the video stream information, the user demand information, and the node information of each computing node are analyzed by adopting the multi-objective optimization algorithm to determine the resource recommendation table, the method further includes:
s230, monitoring information of each optimization index of each computing node, and displaying the information of each optimization index;
s240, adjusting the scheduling information of the target video based on the information of each optimization index.
The optimization indexes comprise time delay, load, reasoning time length and the like of the computing power node. The optimization monitoring mainly compares time delay, load, inference and the like after optimizing the power calculation node with the historical conditions, monitors the optimized conditions in real time, judges the running condition, efficiency and accuracy, and can schedule and change through configuration files.
The embodiment of the invention judges the running condition, the efficiency and the accuracy by monitoring the optimization condition of the calculation nodes in real time; the information of the optimization index of each computing node is monitored in real time, and the scheduling information of the target video is adjusted based on the index information. According to the embodiment of the invention, the power calculation nodes are optimized in real time through monitoring, so that the efficiency of video analysis is improved, the utilization rate of the power calculation nodes is also improved, and the problem of idle power calculation is avoided.
Fig. 2 is a second flow chart of a resource scheduling method according to the present invention, and referring to fig. 2, the present invention also provides a resource scheduling method.
The resource collection system of the resource layer collects node information of the computing power node, including computing power information, storage information and network information, and transmits the node information to the algorithm layer, and the algorithm layer performs data summarization classification.
The user submits the requirements on the unified service platform, on the one hand, the requirements comprise collected video IP addresses, video access modes and the like, and the configuration of the part is submitted to the arrangement and scheduling subsystem; another aspect includes the SLA requirements of the user, such as AI capabilities, latency, cost estimation, etc., as desired.
The algorithm layer selects the appropriate algorithm adaptation based on the training/reasoning task and its own algorithm library.
In the algorithm layer, three parts of video stream information, power node information and user requirements are calculated through a multi-objective optimization algorithm, an optimal solution of a power node suitable for training is selected, power node recommendation is carried out, and a power node recommendation table is returned to the scheduling subsystem. The specific algorithm recommended path is as follows:
the central cloud processing can fully utilize the computing capacity and application resources of the cloud, the middle part of the central cloud processing relates to multi-stage network data forwarding, and the central cloud processing can be used for training large-scale data and models and ensures sufficient computing power.
The edge cloud processing can be used for forwarding data to nearby edge clouds for processing, and computing resources and time delay advantages of the edge clouds can be fully utilized.
The end server processes, and can forward the data to the end server closest to the data access position by using techniques such as computational force routing, etc., so as to provide access effect close to real time.
In the case of limited capacity of edge servers and end servers, because of the diversity of data processing and AI capabilities required by users, it is necessary to involve the collaboration of multiple end servers or the collaboration of cloud end.
The scheduling subsystem performs scheduling according to the resource recommendation table and mainly comprises the following steps: (1) According to the resource recommendation table, one or more AI capability platform interfaces at corresponding positions are called to prepare resources, including performing service expansion, calculation and resource storage preparation on a k8s platform at a bottom layer, performing preparation of related AI service capability on AI platform management capability, and the like; (2) The scheduling subsystem calls a network management module, and establishes network connection with the AI capability platform at the corresponding position according to video stream information such as video IP addresses submitted by users, video access modes and the like.
And arranging a scheduling subsystem to butt-joint the bottom k8s platform and the AI application in real time for the conditions of video flow change, multi-video access and the like in the service process, and carrying out real-time elastic expansion and contraction of resources.
And the AI service capability provided by the AI platform is in butt joint with the video access platform to process video data.
And (3) carrying out log presentation and monitoring on the recommended strategy, and carrying out nearby reasoning on the dispatching of the computing nodes mainly for reducing video reasoning time delay. The optimization monitoring mainly compares time delay, load, inference and the like after optimizing the power calculation node with the historical conditions, monitors the optimized conditions in real time to judge the running condition, efficiency and accuracy, and can schedule and change through configuration files.
The embodiment of the invention obtains the node information of each calculation node; analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table; and scheduling the target video based on the resource recommendation table. According to the embodiment of the invention, the resource recommendation table is determined by comprehensively considering video analysis and the self demand of the power network through the multi-target optimization algorithm, so that the power node and the AI algorithm are determined, then the target video is scheduled according to the resource recommendation table, the determined power node and the AI algorithm are utilized to analyze the target video, the recommendation strategy is simultaneously presented and monitored in a log, the power node is scheduled in real time, and the efficiency of video analysis by using the power network is improved.
Fig. 3 is a schematic block diagram of a resource scheduling system according to the present invention, and referring to fig. 3, the present invention further provides a resource scheduling system, including: a resource layer 301, an algorithm layer 302, and a presentation layer 303;
the resource layer 301 is configured to collect node information of each computing node, and send the node information to the algorithm layer 302;
the edge cloud subsystem provides cloud resources, network resources and interfaces of the scheduling subsystem, node information of the computing nodes is obtained through the interfaces, the node information comprises various computing node real-time information, node utilization rate, network resource conditions and the like, and the node information is basic data for recommending the computing nodes to the service. Meanwhile, the scheduling subsystem provides basic service capabilities such as calculation power on and network on, and after the algorithm layer determines the resource recommendation table, the scheduling subsystem calls an interface to open resource service according to the resource recommendation table.
The algorithm layer 302 is configured to receive the node information sent by the resource layer, analyze the video stream information, the user demand information, and the node information of each computing node by adopting a multi-objective optimization algorithm, determine a resource recommendation table, and send the resource recommendation table to the display layer; scheduling the target video based on the resource recommendation table;
the algorithm layer 302 performs weighted analysis on various data indexes provided by the resource layer by comprehensively considering video reasoning and the self demand of the computing network through a multi-objective optimization algorithm to form a resource recommendation table, and interacts with the scheduling subsystem through an interface to select computing nodes. The training task mainly selects the central cloud node, mainly meets the conditions of calculation power and user cost, and the reasoning task mainly screens the edge cloud node, mainly aims at reducing delay; and calling an arrangement scheduling subsystem interface to issue a task, and issuing the task to an edge cloud node for training and reasoning. And performing intelligent matching between video training reasoning and the AI capability of the analysis platform through a collaborative filtering algorithm.
And the display layer 303 is configured to receive the resource recommendation table and display the resource recommendation table.
Based on the power calculation node recommendation table, carrying out visual display on the daily use condition of the power calculation nodes in the power calculation node recommendation table, wherein the visual display is a strategy log.
Based on the strategy log, the calculation force nodes are monitored and optimized in real time, and the optimization process is optimized monitoring.
The resource scheduling system provided by the embodiment of the invention obtains the node information of each computing node; analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table; and scheduling the target video based on the resource recommendation table. The resource scheduling system determines a resource recommendation table by comprehensively considering video analysis and the self demand of the power network through a multi-target optimization algorithm, so as to determine power calculation nodes and an AI algorithm, then schedules target videos according to the resource recommendation table, analyzes the target videos by using the determined power calculation nodes and the determined AI algorithm, and improves the efficiency of video analysis by using the power calculation network.
Fig. 4 illustrates a schematic structural diagram of an electronic device, as shown in fig. 4, which may include: processor 410, communication interface 420, memory 430 and communication bus 440, wherein processor 410, communication interface 420 and memory 430 communicate with each other through communication bus 440. Processor 410 may invoke logic instructions in memory 430 to perform a resource scheduling method comprising:
acquiring node information of each computing power node;
analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table;
and scheduling the target video based on the resource recommendation table.
Further, the logic instructions in the memory 430 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-only memory (ROM), a random access memory (RAM, randomAccessMemory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform a resource scheduling method provided by the above methods, the method comprising:
acquiring node information of each computing power node;
analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table;
and scheduling the target video based on the resource recommendation table.
The apparatus embodiments described above are merely illustrative, wherein elements illustrated as separate elements may or may not be physically separate, and elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for scheduling resources, comprising:
acquiring node information of each computing power node;
analyzing video stream information, user demand information and node information of each computing node by adopting a multi-target optimization algorithm to determine a resource recommendation table;
and scheduling the target video based on the resource recommendation table.
2. The method for scheduling resources according to claim 1, wherein the analyzing the video stream information, the user demand information, and the node information of each computing node to determine the resource recommendation table using the multi-objective optimization algorithm comprises:
determining the video stream information, the user demand information and node information of each computing power node, and respectively corresponding weight values;
the video stream information, the user demand information and the node information of each computing power node are subjected to weighted analysis by adopting the multi-objective optimization algorithm, the weight value and the historical recommendation information so as to determine the resource recommendation table;
the resource recommendation table comprises an algorithm power node recommendation table and an AI algorithm recommendation table.
3. The method for scheduling a target video according to claim 2, wherein the scheduling the target video based on the resource recommendation table comprises:
determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video;
determining at least one target AI algorithm based on the AI algorithm recommendation table;
and dispatching the target video to the first target computing power node, wherein the first target computing power node performs AI analysis on the target video based on the target AI algorithm.
4. The resource scheduling method of claim 3, wherein the determining at least one first target computing node based on the computing node recommendation table comprises:
if the analysis task of the target video is a video training task, determining at least one central cloud node as the first target computing node based on the computing node recommendation table;
and if the analysis task of the target video is a video reasoning task, determining at least one edge cloud node as the first target computing node based on the computing node recommendation table.
5. The resource scheduling method of claim 3, wherein the determining at least one target AI algorithm based on the AI algorithm recommendation table comprises:
acquiring an AI service capability registry of each first target computing node;
determining the association relation between the AI service capability registry and the analysis task;
and determining at least one target AI algorithm based on the association relationship and the AI algorithm recommendation table.
6. The method of claim 3, wherein after determining at least one first target computing node based on the computing node recommendation table and the analysis task of the target video, further comprising:
if the capacity of the first target computing node is smaller than a set value, determining at least one second target computing node;
scheduling the target video to the first target computing node and the second target computing node;
wherein the first target computing node is of a different node type than the second target computing node.
7. The method for scheduling resources according to claim 1, wherein after analyzing the video stream information, the user demand information, and the node information of each computing node by using the multi-objective optimization algorithm to determine the resource recommendation table, further comprises:
monitoring information of each optimization index of each computing node, and displaying information of each optimization index;
and adjusting the scheduling information of the target video based on the information of each optimization index.
8. A resource scheduling system, comprising: a resource layer, an algorithm layer and a display layer;
the resource layer is used for collecting node information of each computing power node and sending the node information to the algorithm layer;
the algorithm layer is used for receiving the node information sent by the resource layer, analyzing the video stream information, the user demand information and the node information of each computing node by adopting a multi-objective optimization algorithm to determine a resource recommendation table, and sending the resource recommendation table to the display layer; scheduling a target video based on the resource recommendation table;
the display layer is used for receiving the resource recommendation table and displaying the resource recommendation table.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the resource scheduling method of any one of claims 1 to 7 when the program is executed by the processor.
10. A non-transitory computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the resource scheduling method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310421301.0A CN116627631A (en) | 2023-04-19 | 2023-04-19 | Resource scheduling method, system, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310421301.0A CN116627631A (en) | 2023-04-19 | 2023-04-19 | Resource scheduling method, system, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116627631A true CN116627631A (en) | 2023-08-22 |
Family
ID=87620180
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310421301.0A Pending CN116627631A (en) | 2023-04-19 | 2023-04-19 | Resource scheduling method, system, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116627631A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116863408A (en) * | 2023-09-04 | 2023-10-10 | 成都智慧城市信息技术有限公司 | Parallel acceleration and dynamic scheduling implementation method based on monitoring camera AI algorithm |
| CN117331700A (en) * | 2023-10-24 | 2024-01-02 | 广州一玛网络科技有限公司 | Computing power network resource scheduling system and method |
-
2023
- 2023-04-19 CN CN202310421301.0A patent/CN116627631A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116863408A (en) * | 2023-09-04 | 2023-10-10 | 成都智慧城市信息技术有限公司 | Parallel acceleration and dynamic scheduling implementation method based on monitoring camera AI algorithm |
| CN116863408B (en) * | 2023-09-04 | 2023-11-21 | 成都智慧城市信息技术有限公司 | Parallel acceleration and dynamic scheduling implementation method based on monitoring camera AI algorithm |
| CN117331700A (en) * | 2023-10-24 | 2024-01-02 | 广州一玛网络科技有限公司 | Computing power network resource scheduling system and method |
| CN117331700B (en) * | 2023-10-24 | 2024-04-19 | 广州一玛网络科技有限公司 | Computing power network resource scheduling system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116627631A (en) | Resource scheduling method, system, electronic equipment and storage medium | |
| CN104038540B (en) | Method and system for automatically selecting application proxy server | |
| US20010039581A1 (en) | System for balance distribution of requests across multiple servers using dynamic metrics | |
| CN104317658A (en) | MapReduce based load self-adaptive task scheduling method | |
| CN112261120B (en) | Cloud-side cooperative task unloading method and device for power distribution internet of things | |
| CN109831524A (en) | A kind of load balance process method and device | |
| Baranwal et al. | FONS: a fog orchestrator node selection model to improve application placement in fog computing | |
| CN116361006B (en) | Method and system for collaborative scheduling of computing network resources oriented to memory computing separation | |
| Li et al. | An efficient scheduling optimization strategy for improving consistency maintenance in edge cloud environment | |
| CN116662010B (en) | Dynamic resource allocation method and system based on distributed system environment | |
| CN113918240A (en) | Task unloading method and device | |
| CN115421930B (en) | Task processing method, system, device, equipment and computer readable storage medium | |
| CN105491329A (en) | Large-scale monitoring video stream converging method based on stream-oriented computation | |
| CN115604278A (en) | Dynamic load balancing method and system | |
| CN105335376B (en) | A kind of method for stream processing, apparatus and system | |
| CN106789853A (en) | The dynamic dispatching method and device of a kind of transcoder | |
| CN114531448B (en) | Calculation force determining method and device and calculation force sharing system | |
| CN106210120B (en) | A kind of recommended method and its device of server | |
| CN110442786A (en) | A kind of method, apparatus, equipment and the storage medium of prompt information push | |
| CN116668364A (en) | Route planning method and device for real-time audio and video network | |
| CN112866358B (en) | Method, system and device for rescheduling service of Internet of things | |
| CN109670691A (en) | Method, equipment and the customer service system distributed for customer service queue management and customer service | |
| Kavitha et al. | Dynamic load balancing in cloud based multimedia system with genetic algorithm | |
| CN114741200A (en) | Data center station-oriented computing resource allocation method and device and electronic equipment | |
| Grislin-Le Strugeon et al. | A multi-agent approach for vehicle-to-fog fair computation offloading |
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 |