CN116594784B - Method, device and system for scheduling edges and readable storage medium - Google Patents
Method, device and system for scheduling edges and readable storage medium Download PDFInfo
- Publication number
- CN116594784B CN116594784B CN202310874351.4A CN202310874351A CN116594784B CN 116594784 B CN116594784 B CN 116594784B CN 202310874351 A CN202310874351 A CN 202310874351A CN 116594784 B CN116594784 B CN 116594784B
- Authority
- CN
- China
- Prior art keywords
- node
- historical
- pod
- information
- communication
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000004891 communication Methods 0.000 claims abstract description 170
- 239000011159 matrix material Substances 0.000 claims description 64
- 230000006978 adaptation Effects 0.000 claims description 28
- 238000012545 processing Methods 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 16
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 230000003862 health status Effects 0.000 claims description 7
- 230000036541 health Effects 0.000 claims description 5
- 239000000306 component Substances 0.000 description 32
- 230000005012 migration Effects 0.000 description 9
- 238000013508 migration Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 6
- 239000004973 liquid crystal related substance Substances 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 239000008358 core component Substances 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 210000001638 cerebellum Anatomy 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000006317 isomerization reaction Methods 0.000 description 1
- 230000003287 optical effect Effects 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
- G06F9/5072—Grid computing
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
- H04W16/02—Resource partitioning among network components, e.g. reuse partitioning
- H04W16/10—Dynamic resource partitioning
-
- 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)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The application provides a method, a device, a system and a readable storage medium for dispatching edges, which belong to the technical field of cloud computing, wherein the method for dispatching edges is applied to a first node and comprises the following steps: acquiring communication information of a container group Pod when running on the first node; receiving node information of a second node sent by first equipment; under the condition that the network state of the first node cannot meet the application requirement of the Pod, determining a first target node meeting the application requirement according to the communication information and the node information; scheduling the Pod to the first target node; wherein the second node comprises the first target node. The scheme of the application can ensure the stability of the communication performance of the node when the node performs edge scheduling and enable the edge equipment to easily realize the autonomous node scheduling.
Description
Technical Field
The application relates to the technical field of cloud computing, in particular to a method, a device and a system for dispatching edges and a readable storage medium.
Background
The Kubernetes (K8S) serves as an open source platform for managing containerized workloads and services, and the containerization has higher portability across clouds and systems than virtualized deployment applications, thereby meeting the development requirements of cloud computing. Kubernetes performs the workload by placing containers in a group of containers (Pod) running on a Node (Node), and schedules the Pod to which the load belongs to the Node using a scheduling policy, forming its own set of container orchestration rules.
In edge scenarios, kubernetes has high demands on network performance, and the performance of edge devices limits the versatility of Kubernetes, resulting in KubeEdge. The Kubeedge optimizes the edge node agent of the Kubernetes, reduces the resource occupation of edge components, can support more architecture types, and adds an offline autonomous function.
Although KubeEdge provides a containerized solution in an edge network scenario, with the development of the edge computing field, the edge network has an increasingly wide application field. In order to ensure communication efficiency, communication at the same time is a main means of communication between Pod, and network performance of the edge device limits application development in the edge field. Although KubeEdge supports offline autonomy, the stability of Pod itself can be maintained, but the degree of performance optimization for the network-on-edge is very limited. In addition, for Pod with frequent communication at the same time, the KubeEdge adopts a Kubernetes scheduling scheme, scheduling is performed only according to the resource performance of the node, the influence of network factors is ignored, and certain trouble is caused to communication at the same time.
With the development of digital economy, computing power gradually shows the trend of kernel diversification and distribution generalization. In addition to general purpose computing, the advent of high performance computing, intelligent computing, and computing power cores has evolved continuously toward the isomerization of graphics processors (graphics processing unit, GPU), programmable array logic (Field Programmable Gate Array, FPGA), and network processors (Neural-network Processing Unit, NPU). In recent years, with the prosperous development of the internet of things and edge computing, a large number of terminals are connected into a network, computing power gradually extends to the edge side and the end side, and the edge computing power is gradually enriched. The computing force is integrally presented as a cloud edge three-level architecture, and the cloud edge three-level architecture has the characteristic of super-concentration of cloud computing force and super-distribution of edge computing force.
The ubiquitous computing force can break through the performance limit of Shan Diansuan force through network connection, and the cluster advantage of the computing force is exerted. Through global intelligent scheduling and optimization of the computing network resources, the flow of computing power can be effectively promoted, and the demand of business on the use of computing power as required is met. Meanwhile, with the extreme requirements of industry application on the end-to-end quality of the network, the network needs to evolve from best effort to end-to-end deterministic guarantee, and network protocols also need to be innovatively developed. In order to meet the support of edge computing power to a computing power network, the computing network brain scheduling scheme is sunk, the edge scheduling flow is refined, the control of a heterogeneous network to an edge is supported, the existing scheduling scheme must be optimized, and a scheduling frame is improved.
Therefore, the existing KubeEdge technology has the following disadvantages: although the KubeEdge can ensure the Pod running state through offline autonomy, the stability of the communication state with the cloud and the performance of Pod communication at the same time cannot be ensured. In summary, the existing edge scheduling method has the problems that the communication performance is poor, and the network performance of edge equipment is difficult to realize node autonomous scheduling.
Disclosure of Invention
The application aims to provide a method, a device, a system and a readable storage medium for scheduling edges, which are used for solving the problems that the existing edge scheduling method has poor communication performance and the network performance of edge equipment is difficult to realize autonomous node scheduling.
In order to solve the technical problems, the embodiment of the application provides the following technical scheme.
In a first aspect, an embodiment of the present application provides an edge scheduling method, which is applied to a first node, where the method includes:
acquiring communication information of a container group Pod when running on the first node;
receiving node information of a second node sent by first equipment;
under the condition that the network state of the first node cannot meet the application requirement of the Pod, determining a first target node meeting the application requirement according to the communication information and the node information;
scheduling the Pod to the first target node;
wherein the second node comprises the first target node.
Optionally, the determining, according to the communication information and the node information, a first target node that meets the application requirement includes:
obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained Singular Value Decomposition (SVD) model;
determining at least one second target node in the second nodes according to the adaptation degree between the Pod and each second node;
and determining a first target node meeting the application requirement in the at least one second target node according to the sequence of the adaptation degree from high to low.
Optionally, the method further comprises:
constructing a history communication matrix according to the history communication information;
constructing a history weight matrix according to the history communication information and the history node information;
training a preset initial SVD model according to the historical communication matrix, the historical weight matrix and the adaptation degree between the historical Pod and the historical nodes to obtain a trained SVD model.
Optionally, the constructing a historical communication matrix according to the historical communication information includes:
constructing an initial historical communication matrix according to the historical Pod name, the historical Pod name space, the communication time between the historical Pod and the historical node and the name of the historical node;
processing the initial historical communication matrix according to a scheduling strategy corresponding to the historical Pod to obtain the historical communication matrix;
wherein the historical communication information comprises the historical Pod name, the historical Pod namespace, communication time between the historical Pod and a historical node and the name of the historical node.
Optionally, the constructing a historical weight matrix according to the historical communication information and the historical node information includes:
obtaining the historical weight matrix according to the successful communication times between the historical Pod and the historical nodes, the historical node bandwidth information and the historical node network component health state information;
wherein the historical node information includes the historical node bandwidth information and the historical node network component health status information.
In a second aspect, an embodiment of the present application further provides a method for edge scheduling, which is applied to a first device, where the method includes:
transmitting node information of the second node to the first node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
In a third aspect, an embodiment of the present application further provides an edge scheduling apparatus, applied to a first node, where the apparatus includes:
the acquisition module is used for acquiring communication information of the container group Pod when the container group Pod runs on the first node;
the receiving module is used for receiving the node information of the second node sent by the first equipment;
the determining module is used for determining a first target node meeting the application requirement according to the communication information and the node information under the condition that the network state of the first node can not meet the application requirement of the Pod;
a scheduling module, configured to schedule the Pod to the first target node;
wherein the second node comprises the first target node.
In a fourth aspect, an embodiment of the present application further provides an edge scheduling apparatus, applied to a first device, where the apparatus includes:
a transmitting module for transmitting node information of the second node to the first node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
In a fifth aspect, an embodiment of the present application further provides an edge scheduling system, including a first node and a first device;
wherein the first node is configured to perform the steps in the edge-to-edge scheduling method according to any one of the first aspects;
the first device is configured to perform the steps in the edge-to-edge scheduling method according to the second aspect.
In a sixth aspect, an embodiment of the present application further provides a readable storage medium, on which a program or an instruction is stored, which when executed by a processor, implements the steps of the method for edge scheduling according to any one of the first aspects, or implements the steps of the method for edge scheduling according to the second aspect.
The technical scheme of the application has the following beneficial effects:
according to the edge scheduling method provided by the scheme of the application, the communication information of the container group Pod when running on the first node is obtained through the first node, the node information of the second node sent by the first device is received, the stability of the communication performance of the node when performing edge scheduling can be ensured, and under the condition that the network state of the first node can not meet the application requirement of the Pod, the first target node meeting the application requirement is determined according to the communication information and the node information, and the Pod is scheduled to the first target node, so that the edge device can easily realize autonomous node scheduling.
Drawings
Fig. 1 is a flowchart of an edge scheduling method applied to a first node according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a system for edge-to-edge scheduling according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a training SVD model according to an embodiment of the present application;
fig. 4 is a flowchart of a method for scheduling edges applied to a first device according to an embodiment of the present application;
FIG. 5 is a flowchart of a method for edge-to-edge scheduling according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of an edge scheduling device applied to a first node according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a side-to-side scheduling apparatus applied to a first device according to an embodiment of the present application;
FIG. 8 is a second schematic diagram of a system for edge-to-edge scheduling according to an embodiment of the present application.
Detailed Description
In order to make the technical problems, technical solutions and advantages to be solved more apparent, the following detailed description will be given with reference to the accompanying drawings and specific embodiments.
It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment," "in an embodiment," or "in an alternative embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
In various embodiments of the present application, it should be understood that the sequence numbers of the following processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
In addition, the terms "system" and "network" are often used interchangeably herein.
In the embodiments provided herein, it should be understood that "B corresponding to a" means that B is associated with a from which B may be determined. It should also be understood that determining B from a does not mean determining B from a alone, but may also determine B from a and/or other information.
In order to solve the problems of poor communication performance and unscientific scheduling scheme of the existing side-by-side communication method, the embodiment of the application provides a side-by-side scheduling method, a side-by-side scheduling device, a side-by-side scheduling system and a readable storage medium.
As shown in fig. 1, an embodiment of the present application provides an edge scheduling method, which is applied to a first node, and includes:
step 101: and acquiring communication information of the container group Pod when the container group Pod runs on the first node.
It should be noted that, the first Node provided in the embodiment of the present application is an edge Node (Node), and the edge Node may also be referred to as an edge device.
In the step, a user firstly creates a work load on a cloud device and configures a corresponding scheduling policy, the cloud device schedules a Pod according to the scheduling policy, creates the Pod on an edge node (namely a first node) meeting the conditions, the Pod normally operates on the first node, and obtains and stores communication information when the Pod operates on the first node.
Wherein the communication information includes at least one of:
the name of the Pod; a namespace of the Pod; the name of the first node; a communication time between the Pod and the first node; whether communication between the Pod and the first node is successful; a communication frequency between the Pod and the first node; a communication success rate between the Pod and the first node; a communication delay between the Pod and the first node; destination IP.
The embodiment of the application provides an Edge scheduling system, as shown in fig. 2, which comprises Cloud equipment (Cloud) and Edge nodes (Edge). The cloud device comprises a cloud core component (CloudCore), wherein the cloud core component comprises a controller (Controllers) and a cloud structure body (CloudHub), a user is connected with a Server (Server) through a Cobernett (K8S) application programming interface (Application Programming Interface, API), a workload is created, a corresponding scheduling strategy is configured, pod is scheduled according to the scheduling strategy, and Pod is created on an edge node (namely a first node) which meets the conditions. The edge node side components include a monitor edge communication component (EdgeMonitor), an information backup component (DataStore), and an execute edge schedule component (execute edge schedule component) that includes an edge core component (EdgeCore) that includes an edge fabric (EdgeHub), an edge container application management component (Edged), and a meta manager component (MetaManager). The EdgeMonitor monitors and obtains the communication and stores the communication record in the DataStore.
Step 102: and receiving node information of the second node sent by the first equipment.
It should be noted that, in the embodiment of the present application, the first device is a cloud device, and the second node is an edge node except the first node.
In this step, the cloud device may send the node information of the second node to the first node at regular time or at intervals for a preset period (for example, one minute), and the edge monitor of the first node receives the node information of the second node and makes a backup on the first node, and stores the node information of the second node into the DataStore, where the node information of the first node is also stored.
With continued reference to fig. 2, cloudhub sends node information for the second node to the EdgeHub of the edge node. The EdgeMonitor obtains node information and stores the node information record in the DataStore.
Wherein the node information includes at least one of:
node bandwidth information and node network component health status information; node network stability information; and (5) node network time delay.
Step 103: and under the condition that the network state of the first node can not meet the application requirement of the Pod, determining a first target node meeting the application requirement according to the communication information and the node information. Wherein the second node comprises the first target node.
In this step, the first node monitors its own network state, and when the network state (such as network bandwidth, network stability, network delay, etc.) cannot meet the application requirement of Pod, the first target node meeting the scheduling condition (i.e. meeting the application requirement) is found in the second node according to the node information and communication information of the second node backed up in the DataStore.
Step 104: and dispatching the Pod to the first target node.
In this step, the first node migrates the Pod schedule to the first target node, that is, the edge device may autonomously apply the scheduling policy to schedule.
Through the steps, communication scheduling among the edge nodes can be realized, the communication performance is ensured, and the scientificity of a scheduling strategy is improved.
In an alternative embodiment of the present application, step 102 includes:
and sending node request information to the first equipment, optionally, the monitoring edge communication component of the first node sends the node request information to the first equipment at regular time, the first equipment sends node information of a second node to the first node according to the node request information, and the first node receives the node information of the second node sent by the first equipment according to the node request information.
It should be noted that, for the power calculation network, the operator scheduling scheme affects the usability of the whole network, plays a role in the whole scheme, and if the operator scheduling scheme is combined with the artificial intelligence (Artificial Intelligence, AI) capability, the whole quality of the power calculation network can be greatly improved.
In an optional embodiment of the application, the determining, according to the communication information and the node information, a first target node that meets the application requirement includes:
obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained Singular Value Decomposition (SVD) model;
determining at least one second target node in the second nodes according to the adaptation degree between the Pod and each second node;
and determining a first target node meeting the application requirement in the at least one second target node according to the sequence of the adaptation degree from high to low.
Specifically, in this optional embodiment, training a singular value decomposition model (Singular Value Decomposition, SVD) model to obtain a trained singular value decomposition model SVD model, where the input of the trained SVD model is communication information of Pod and Node information of second nodes, and the output of the trained SVD model is the fitness between Pod and each second Node, and ordering the fitness to obtain a Pod/Node scoring list, optionally selecting nodes with three top scores (i.e. at least one second target Node in the second nodes), and for nodes with three top scores, analyzing, by the first Node, the possibility of migration (i.e. whether the application requirement is met) according to Node information backed up in DataStore and scheduling strategies corresponding to Pod itself in sequence, if migration can be performed, directly performing application migration by the first Node, and not needing to issue Pod through cloud equipment; if the migration is not possible, the next Pod is judged in a forward direction until all the second target nodes are judged or the second target nodes which can be migrated are determined.
Further, the process of training the SVD model is as follows:
constructing a history communication matrix according to the history communication information, namely forming a history communication matrix by the history Pod and the history Node according to the history communication information stored in the DataStore; the historical communication information can be counted once per minute for the historical Pod, and the node bandwidth occupancy rate and the state of the network component are recorded;
constructing a historical weight matrix according to the historical communication information and the historical node information, namely constructing a historical weight matrix according to the historical communication information stored in the DataStore and the historical node information of the second node monitored by the component in real time;
training a preset initial SVD model according to the historical communication matrix, the historical weight matrix and the adaptation degree between the historical Pod and the historical nodes to obtain a trained SVD model, specifically, placing the historical communication matrix into the preset initial SVD model for training, adding the historical weight matrix as a bias matrix for training, and obtaining the trained SVD model according to the adaptation degree between the historical Pod and the historical nodes.
Further, referring to fig. 3, the construction of the historical communication matrix according to the historical communication information includes:
according to the historical Pod name (shouldUsing names), a historical Pod name space, communication time between the historical Pod and the historical nodes and names of the historical nodes to construct an initial historical communication matrix; wherein the historical communication information includes the historical Pod name (Podm) A name space of the historical Pod, a communication time between the historical Pod and a historical Node, and a name (Node) of the historical Noden）。
Specifically, an initial history communication matrix is constructed according to the history Pod names, history Pod namespaces, communication time between the history Pod and the history nodes and the names of the history nodes backed up in the DataStoreThe matrix is as follows>Every element->Representing PodmNode within one minutenNumber of communications:
;
processing the initial historical communication matrix according to a scheduling policy corresponding to the historical Pod to obtain the historical communication matrix, specifically, processing the initial historical communication matrix according to node information backed up in a DataStore in combination with the historical Pod scheduling policyThe Pod/Node element in which the scheduling policy is not satisfied is set to 0. The specific processing formula is as follows:
;
wherein, the liquid crystal display device comprises a liquid crystal display device,qrepresenting the processing parameters, taking the value of 0 or 1,representing historical communicationsMatrix (S)>Representing elements in the historical communications matrix.
Further, with continued reference to fig. 3, the construction of the historical weight matrix according to the historical communication information and the historical node information includes:
obtaining a history weight matrix according to the successful times of communication between the history Pod and the history node, the history node bandwidth information (bandwidth occupation) and the history node network component health state information (component state); wherein the historical node information includes the historical node bandwidth information and the historical node network component health status information.
Specifically, a historical weight matrix is constructed according to the node information backed up in the DataStore and the communication success timesThe specific formula is as follows:
;
wherein, the liquid crystal display device comprises a liquid crystal display device,prepresenting the elements in the historical weight matrix,cthe number of node communication failures is represented, and the more the number of communication failures is, the worse the network communication is;trepresenting the number of recordings processed according to the time stamp of the monitoring information;representing the bandwidth occupation ratio corresponding to each record, wherein the higher the occupation ratio is, the more busy the network is; />And the component state of the corresponding network is recorded each time, the component is normally 0, the component abnormality is 1, and the more the component abnormality times are, the worse the component stability is. Element(s)Representation NodenThe higher the application mobility, the more nodesApplication migration is required.
Thereafter, the history communication matrixTraining by putting a preset initial SVD model, and training a historical weight matrix +.>Training is added as a bias matrix, and the training formula is as follows:
;
wherein, the liquid crystal display device comprises a liquid crystal display device,μthe excitation function is represented by a function of the excitation,respectively represent training bias, ++>Indicating the degree of adaptation between the history Pod and the history node,/->Comprises->Personal->。
Inputting communication information when Pod operates on the first node and node information of the second node into a trained SVD model to obtain a resultWherein each element->Representing PodmFor NodenA migration application list is generated.
Thereafter, the method further comprises:
after scheduling the Pod to the first target node, updating communication information of the backed up Pod in the DataStore, and synchronizing communication information of the Pod by the first device to the first target node.
As shown in fig. 4, the embodiment of the present application further provides a method for scheduling edges, which is applied to a first device, where the method includes:
step 401: transmitting node information of the second node to the first node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
It should be noted that, in the embodiment of the present application, the first device is a cloud device, the first Node is an edge Node (Node), the edge Node may also be referred to as an edge device, and the second Node is an edge Node other than the first Node.
In the step, a user firstly creates a work load on a cloud device and configures a corresponding scheduling policy, the cloud device schedules a Pod according to the scheduling policy, creates the Pod on an edge node (namely a first node) meeting the conditions, the Pod normally operates on the first node, and obtains and stores communication information when the Pod operates on the first node.
The cloud device may send node information of the second node to the first node at regular time or at intervals of a preset time period (for example, one minute), where the edge monitor of the first node receives the node information of the second node and makes a backup on the first node, and stores the node information of the second node into the DataStore, and meanwhile, the node information of the first node is also stored in the DataStore.
The first node monitors the network state of the first node, and when the network state (such as network bandwidth, network stability, network delay and the like) cannot meet the application requirement of Pod, the first target node meeting the scheduling condition (namely meeting the application requirement) is searched in the second node according to the node information and the communication information of the second node backed up in the DataStore.
The first node migrates the Pod schedule to the first target node, that is, the edge device can autonomously apply the scheduling policy to schedule.
Through the steps, communication scheduling among the edge nodes can be realized, the communication performance is ensured, and the scientificity of a scheduling strategy is improved.
Optionally, the node information includes at least one of:
node bandwidth information and node network component health status information; node network stability information; and (5) node network time delay.
Optionally, the communication information includes at least one of:
the name of the Pod; a namespace of the Pod; the name of the first node; a communication time between the Pod and the first node; whether communication between the Pod and the first node is successful; a communication frequency between the Pod and the first node; a communication success rate between the Pod and the first node; and communication delay between the Pod and the first node.
In an optional embodiment of the application, the sending node information of the second node to the first node comprises:
receiving node request information sent by the first node;
and sending the node information of the second node to the first node according to the node request information.
The following specifically describes a specific flow of the edge scheduling method provided by the embodiment of the present application with reference to fig. 5:
1. the user creates a load: a user creates a workload at the cloud device and configures a corresponding scheduling policy;
2. scheduling Pod according to a scheduling policy: the cloud creates Pod on the edge node meeting the conditions according to the scheduling strategy;
pod normal operation: pod operates normally on edge nodes;
4. monitoring Pod edge communication: monitoring Pod edge communication, backing up Pod name, communication time, destination ip and whether successful communication is carried out in a DataStore, and backing up information such as communication frequency, communication success rate, communication time delay and the like;
5. synchronizing node information: the edge node requests node information from the cloud end at regular time;
6. issuing node information: the cloud sends the node information to the edge node and makes a backup at the edge node;
7. performing edge-to-edge scheduling: monitoring the network state of the edge node, and searching other nodes meeting the scheduling conditions according to the node information backed up in the component when the network state (network bandwidth, network stability, network delay and the like) of the node cannot meet the Pod requirement;
8. component update information: when the node Pod changes, pod information in the component is updated;
9. cloud synchronization node information: after the migration is completed, the cloud synchronizes the Pod information of the nodes.
According to the side scheduling system provided by the embodiment of the application, for the Pod with higher network requirements, the components provide the functions of side communication backup, pod intelligent migration and the like, the quality of the Pod side communication is optimized, and the use experience of a user is improved.
According to the edge scheduling method provided by the embodiment of the application, according to the node network state migration application, the node network performance is analyzed and quantized by monitoring edge communication and combined with the SVD model, so that the method is a more scientific scheduling method, and the node network performance is prevented from affecting the application operation quality; based on the full node information of the edge backup, enabling the edge node to have the possibility of realizing scheduling autonomously, and sinking the decision right of cloud edge scheduling to the edge; the method is applied to a computational power network, realizes the sinking of a computational network task scheduling scheme, forms a computational network cerebellum, helps to reasonably schedule edge operators, and realizes the unification of physical distribution logic of resources and capabilities; based on the existing scheduling strategy, based on node network performance and bandwidth requirements of the container, the application of intelligent scheduling is realized by combining a deep learning model, and the scientificity of the scheduling strategy is improved; the autonomy of the edge side in the whole network is enhanced, and the high requirement of the edge side on the network quality is reduced. For the whole computational power network, the computational power sinking scheme with high feasibility meets the end-to-end communication requirement of the computational power network, refines the scheduling link of the network to the edge computational power, improves the availability of the network on the edge side, and improves the service efficiency.
As shown in fig. 6, an embodiment of the present application further provides an edge scheduling apparatus, which is applied to a first node, and the apparatus includes:
an obtaining module 601, configured to obtain communication information when the container group Pod runs on the first node;
a receiving module 602, configured to receive node information of a second node sent by a first device;
a determining module 603, configured to determine, according to the communication information and the node information, a first target node that meets the application requirement when the network state of the first node cannot meet the application requirement of the Pod;
a scheduling module 604, configured to schedule the Pod to the first target node;
wherein the second node comprises the first target node.
Optionally, the determining module 603 includes:
the first processing unit is used for obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained singular value decomposition model SVD model;
a first determining unit, configured to determine at least one second target node among the second nodes according to an adaptation degree between the Pod and each of the second nodes;
and the second determining unit is used for determining a first target node meeting the application requirement in the at least one second target node according to the sequence from high to low of the adaptation degree.
Optionally, the determining module 603 further includes:
the second processing unit is used for constructing a historical communication matrix according to the historical communication information;
the third processing unit is used for constructing a history weight matrix according to the history communication information and the history node information;
and the fourth processing unit is used for training a preset initial SVD model according to the history communication matrix, the history weight matrix and the adaptation degree between the history Pod and the history nodes to obtain a trained SVD model.
Optionally, the second processing unit is specifically configured to:
constructing an initial historical communication matrix according to the historical Pod name, the historical Pod name space, the communication time between the historical Pod and the historical node and the name of the historical node;
processing the initial historical communication matrix according to a scheduling strategy corresponding to the historical Pod to obtain the historical communication matrix;
wherein the historical communication information comprises the historical Pod name, the historical Pod namespace, communication time between the historical Pod and a historical node and the name of the historical node.
Optionally, the third processing unit is specifically configured to:
obtaining the historical weight matrix according to the successful communication times between the historical Pod and the historical nodes, the historical node bandwidth information and the historical node network component health state information;
wherein the historical node information includes the historical node bandwidth information and the historical node network component health status information.
It should be noted that, the edge scheduling device applied to the first node provided in the embodiment of the present application is a device capable of executing the edge scheduling method applied to the first node, so all embodiments of the edge scheduling method applied to the first node are applicable to the device, and the same or similar technical effects can be achieved.
As shown in fig. 7, an embodiment of the present application further provides an edge scheduling apparatus, which is applied to a first device, where the apparatus includes:
a transmitting module 701, configured to transmit node information of the second node to the first node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
It should be noted that, the edge scheduling device applied to the first device provided in the embodiment of the present application is a device capable of executing the edge scheduling method applied to the first device, and all embodiments of the edge scheduling method applied to the first device are applicable to the device, and the same or similar technical effects can be achieved.
As shown in fig. 8, the embodiment of the present application further provides an edge scheduling system, including a first node 801 and a first device 802;
wherein the first node 801 is configured to perform:
acquiring communication information of a container group Pod when running on the first node;
receiving node information of a second node sent by first equipment;
under the condition that the network state of the first node cannot meet the application requirement of the Pod, determining a first target node meeting the application requirement according to the communication information and the node information;
scheduling the Pod to the first target node;
wherein the second node comprises the first target node.
Optionally, the first node 801 is configured to:
obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained Singular Value Decomposition (SVD) model;
determining at least one second target node in the second nodes according to the adaptation degree between the Pod and each second node;
and determining a first target node meeting the application requirement in the at least one second target node according to the sequence of the adaptation degree from high to low.
Optionally, the first node 801 is further configured to:
constructing a history communication matrix according to the history communication information;
constructing a history weight matrix according to the history communication information and the history node information;
training a preset initial SVD model according to the historical communication matrix, the historical weight matrix and the adaptation degree between the historical Pod and the historical nodes to obtain a trained SVD model.
Optionally, the first node 801 is specifically configured to:
constructing an initial historical communication matrix according to the historical Pod name, the historical Pod name space, the communication time between the historical Pod and the historical node and the name of the historical node;
processing the initial historical communication matrix according to a scheduling strategy corresponding to the historical Pod to obtain the historical communication matrix;
wherein the historical communication information comprises the historical Pod name, the historical Pod namespace, communication time between the historical Pod and a historical node and the name of the historical node.
Optionally, the first node 801 is specifically configured to:
obtaining the historical weight matrix according to the successful communication times between the historical Pod and the historical nodes, the historical node bandwidth information and the historical node network component health state information;
wherein the historical node information includes the historical node bandwidth information and the historical node network component health status information.
The first device 802 is configured to perform:
transmitting node information of the second node to the first node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
In addition, a specific embodiment of the present application further provides a computer readable storage medium, on which a computer program is stored, where the program when executed by a processor implements the steps in the edge scheduling method applied to the first node as set forth in any one of the above, or implements the steps in the edge scheduling method applied to the first device as set forth in any one of the above.
In the several embodiments provided in the present application, it should be understood that the disclosed methods and apparatus may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may be physically included separately, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in hardware plus software functional units.
The integrated units implemented in the form of software functional units described above may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform part of the steps of the transceiving method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
While the foregoing is directed to the preferred embodiments of the present application, it will be appreciated by those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present application, and such modifications and adaptations are intended to be comprehended within the scope of the present application.
Claims (6)
1. An edge-to-edge scheduling method, applied to a first node, the method comprising:
acquiring communication information of a container group Pod when running on the first node;
receiving node information of a second node sent by first equipment;
under the condition that the network state of the first node cannot meet the application requirement of the Pod, determining a first target node meeting the application requirement according to the communication information and the node information; the second node comprises the first target node;
scheduling the Pod to the first target node;
wherein the determining, according to the communication information and the node information, a first target node that meets the application requirement includes:
obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained Singular Value Decomposition (SVD) model;
determining at least one second target node in the second nodes according to the adaptation degree between the Pod and each second node;
determining a first target node meeting the application requirement in the at least one second target node according to the sequence of the adaptation degree from high to low;
wherein the method further comprises:
constructing a history communication matrix according to the history communication information;
constructing a history weight matrix according to the history communication information and the history node information;
training a preset initial SVD model according to the historical communication matrix, the historical weight matrix and the adaptation degree between the historical Pod and the historical nodes to obtain a trained SVD model.
2. The edge scheduling method of claim 1, wherein constructing a historical communication matrix from historical communication information comprises:
constructing an initial historical communication matrix according to the historical Pod name, the historical Pod name space, the communication time between the historical Pod and the historical node and the name of the historical node;
processing the initial historical communication matrix according to a scheduling strategy corresponding to the historical Pod to obtain the historical communication matrix;
wherein the historical communication information comprises the historical Pod name, the historical Pod namespace, communication time between the historical Pod and a historical node and the name of the historical node.
3. The edge scheduling method of claim 2, wherein constructing a historical weight matrix from the historical communication information and historical node information comprises:
obtaining the historical weight matrix according to the successful communication times between the historical Pod and the historical nodes, the historical node bandwidth information and the historical node network component health state information;
wherein the historical node information includes the historical node bandwidth information and the historical node network component health status information.
4. An edge scheduling apparatus for use with a first node, the apparatus comprising:
the acquisition module is used for acquiring communication information of the container group Pod when the container group Pod runs on the first node;
the receiving module is used for receiving the node information of the second node sent by the first equipment;
the determining module is used for determining a first target node meeting the application requirement according to the communication information and the node information under the condition that the network state of the first node can not meet the application requirement of the Pod; the second node comprises the first target node;
a scheduling module, configured to schedule the Pod to the first target node;
wherein, the determining module includes:
the first processing unit is used for obtaining the adaptation degree between the Pod and each second node according to the communication information, the node information and the trained singular value decomposition model SVD model;
a first determining unit, configured to determine at least one second target node among the second nodes according to an adaptation degree between the Pod and each of the second nodes;
the second determining unit is used for determining a first target node meeting the application requirement in the at least one second target node according to the sequence from high to low of the adaptation degree;
wherein, the determining module further includes:
the second processing unit is used for constructing a historical communication matrix according to the historical communication information;
the third processing unit is used for constructing a history weight matrix according to the history communication information and the history node information;
and the fourth processing unit is used for training a preset initial SVD model according to the history communication matrix, the history weight matrix and the adaptation degree between the history Pod and the history nodes to obtain a trained SVD model.
5. An edge scheduling system is characterized by comprising a first node and first equipment;
wherein the first node is configured to perform the steps in the edge scheduling method of any one of claims 1 to 3;
the first device is configured to perform node information for transmitting to a first node a second node;
the node information of the second node is used for indicating that under the condition that the network state of the first node cannot meet the application requirement of a Pod set Pod running on the first node, determining a first target node meeting the application requirement according to the communication information and the node information when the Pod runs on the first node, and scheduling the Pod to the first target node; the second node includes the first target node.
6. A readable storage medium having stored thereon a program or instructions which when executed by a processor performs the steps in the method of edge scheduling of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310874351.4A CN116594784B (en) | 2023-07-17 | 2023-07-17 | Method, device and system for scheduling edges and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310874351.4A CN116594784B (en) | 2023-07-17 | 2023-07-17 | Method, device and system for scheduling edges and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116594784A CN116594784A (en) | 2023-08-15 |
| CN116594784B true CN116594784B (en) | 2023-10-20 |
Family
ID=87612105
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310874351.4A Active CN116594784B (en) | 2023-07-17 | 2023-07-17 | Method, device and system for scheduling edges and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116594784B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114756340A (en) * | 2022-03-17 | 2022-07-15 | 中国联合网络通信集团有限公司 | Computing power scheduling system, method, device and storage medium |
| CN114880100A (en) * | 2022-05-27 | 2022-08-09 | 中国工商银行股份有限公司 | Container dynamic scheduling method and device, computer equipment and storage medium |
| CN115664988A (en) * | 2022-11-04 | 2023-01-31 | 西北工业大学 | Kubernetes-oriented complex service deployment and scheduling method |
| CN116225609A (en) * | 2022-11-30 | 2023-06-06 | 重庆紫光华山智安科技有限公司 | Pod scheduling method, device and equipment of Kubernetes cluster and computer readable medium |
| WO2023109015A1 (en) * | 2021-12-16 | 2023-06-22 | 苏州浪潮智能科技有限公司 | Container application scheduling method and apparatus, electronic device and storage medium |
-
2023
- 2023-07-17 CN CN202310874351.4A patent/CN116594784B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023109015A1 (en) * | 2021-12-16 | 2023-06-22 | 苏州浪潮智能科技有限公司 | Container application scheduling method and apparatus, electronic device and storage medium |
| CN114756340A (en) * | 2022-03-17 | 2022-07-15 | 中国联合网络通信集团有限公司 | Computing power scheduling system, method, device and storage medium |
| CN114880100A (en) * | 2022-05-27 | 2022-08-09 | 中国工商银行股份有限公司 | Container dynamic scheduling method and device, computer equipment and storage medium |
| CN115664988A (en) * | 2022-11-04 | 2023-01-31 | 西北工业大学 | Kubernetes-oriented complex service deployment and scheduling method |
| CN116225609A (en) * | 2022-11-30 | 2023-06-06 | 重庆紫光华山智安科技有限公司 | Pod scheduling method, device and equipment of Kubernetes cluster and computer readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116594784A (en) | 2023-08-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109034396B (en) | Method and apparatus for processing deep learning jobs in a distributed cluster | |
| CN112069265B (en) | Synchronization method of configuration data, business data system, computer system and medium | |
| CN101652750B (en) | Data processing device, distributed processing system and data processing method | |
| CN110716744A (en) | Data stream processing method, system and computer readable storage medium | |
| CN110769278A (en) | Distributed video transcoding method and system | |
| CN102591921A (en) | Scheduling and management in a personal datacenter | |
| CN109117252B (en) | Method and system for task processing based on container and container cluster management system | |
| CN112114950A (en) | Task scheduling method and device and cluster management system | |
| CN109656690A (en) | Scheduling system, method and storage medium | |
| CN114610474B (en) | Multi-strategy job scheduling method and system under heterogeneous supercomputing environment | |
| CN113742031A (en) | Node state information acquisition method and device, electronic equipment and readable storage medium | |
| US20210194774A1 (en) | System and method for a generic key performance indicator platform | |
| CN113033814A (en) | Method, apparatus and storage medium for training machine learning model | |
| CN111104548B (en) | Data feedback method, system and storage medium | |
| CN111435315A (en) | Method, apparatus, device and computer readable medium for allocating resources | |
| CN113014649B (en) | Cloud Internet of things load balancing method, device and equipment based on deep learning | |
| Liu et al. | KubFBS: A fine‐grained and balance‐aware scheduling system for deep learning tasks based on kubernetes | |
| CN115361382B (en) | Data processing method, device, equipment and storage medium based on data group | |
| CN116594784B (en) | Method, device and system for scheduling edges and readable storage medium | |
| Tang et al. | A survey on scheduling techniques in computing and network convergence | |
| CN111190731A (en) | Cluster task scheduling system based on weight | |
| CN117215764A (en) | Computing power resource processing method, device, equipment and storage medium | |
| CN113434591B (en) | Data processing method and device | |
| CN115378937A (en) | Distributed concurrency method, device and equipment for tasks and readable storage medium | |
| CN115001692A (en) | Model updating method and device, computer readable storage medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |