CN115935563A - Network bandwidth prediction method and device based on graph neural network - Google Patents
Network bandwidth prediction method and device based on graph neural network Download PDFInfo
- Publication number
- CN115935563A CN115935563A CN202211495900.9A CN202211495900A CN115935563A CN 115935563 A CN115935563 A CN 115935563A CN 202211495900 A CN202211495900 A CN 202211495900A CN 115935563 A CN115935563 A CN 115935563A
- Authority
- CN
- China
- Prior art keywords
- network
- link
- node
- nodes
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/50—Reducing energy consumption in communication networks in wire-line communication networks, e.g. low power modes or reduced link rate
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a network bandwidth prediction method and a device based on a graph neural network, which comprises the steps of firstly, acquiring data under different network topological structures, wherein the data comprises the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, and network performance related information such as link connection relation between nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and the like; constructing a feature engineering, namely constructing a feature vector of the network node according to the information of the network topology node, and constructing a feature vector of each link according to the information of each link between the nodes; thirdly, abstract representation of a network system is carried out, the network is abstracted into a graph represented by nodes and edges, the weights of the nodes are assigned by the characteristic vectors of the nodes, and the edges are assigned by the characteristic vectors of the links between the nodes; fourthly, modeling the node characteristics and the link characteristics in the network by adopting a graph neural network algorithm, and training to obtain new node characteristics and link characteristic information; and finally, predicting the bandwidth between the links by adopting a machine learning algorithm according to the updated node characteristics and the updated link characteristic information.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a network bandwidth prediction method and device based on a graph neural network.
Background
With the rapid development of new-generation information technologies such as internet of things, cloud computing, edge computing, artificial intelligence and the like, the robot has realized the intelligent progress from perception to cognition, reasoning and decision making, becomes one of the main trends of future development in various industry fields, and has great application value in many aspects such as industrial manufacturing, life service and the like. The existing robot technology integrates cloud computing and edge computing to generate a cloud robot and a fog robot, the cloud robot provides sufficient computing power for the robot by using the strong computing power of a cloud server, the fog robot aims to solve the problem of poor real-time response capability of the cloud computing, and partial cloud side communication, computing, data storage and processing capabilities are achieved by using the edge computing. The current research trend is to adopt a computing architecture with the fusion of a robot body (end), an edge server (edge) and a cloud server (cloud) to provide a distribution relation which can adapt to the computing, storage and cooperative work of the robot for executing different computing tasks, and provide computing power support at the cloud side and the edge side so as to realize more effective and economic computing power deployment. Due to the fact that a large amount of data transmission is involved between the robot body and the edge server and the cloud server, when the edge cloud cooperative computing is carried out, network bandwidth limitation is one of the most critical limiting factors. The influence of the network must be considered in the modeling process, and with the increasing size and the increasing complexity of the existing network scale, the factors influencing the normal operation of the network are increased, the network bandwidth is also influenced by network jitter, packet loss and the like, and the requirement for real-time prediction of the network bandwidth is increased.
Aiming at the prediction research of network bandwidth, the existing modeling technology cannot realize accurate estimation of the network bandwidth, and has the defects of huge calculated amount, ideal model construction, incomplete network information consideration and the like. The research on network performance analysis can be divided into a measurement method, an analytical method and a simulation method. The measurement method is a measurement for monitoring relevant characteristics by simply using software and hardware tools. The analysis method describes the relationship between the performance characteristics and the system parameters by using a mathematical theory method to obtain a parameter solution (such as queuing theory, network calculation and the like) of performance estimation. The simulation method is to establish a network system model in an abstract way and predict the network performance by combining with relevant technologies such as mathematical description and network modeling.
The above method has several problems as follows:
(1) The measurement method is difficult to meet the strict requirements of the novel network on measurement and analysis;
(2) The analytical method is an ideal model, cannot reflect the real situation in the network, and has a certain difference with the actual complex network.
(3) Existing simulation models generally perform in more widely used non-Euclidean spaces (e.g., in real-world networks with complex and irregular connection information)
With the rise of deep learning and the development of data driving in recent years, more and more researches start to adopt a deep neural network to predict the network performance. The recurrent neural network RNN has a certain time memory function and is largely used in flow prediction by researchers, but a model based on the RNN only considers time sequence characteristics and ignores spatial characteristics. However, the conventional convolutional neural network can only process euclidean space data (such as images, texts and voice), and for graph data with unstructured characteristics like the network, the graph neural network can generate a graph from the unstructured data, the output of the graph is not transferred along with the input sequence of nodes, edges represent the link connection relationship between two nodes, the states of the nodes and the edges can be updated depending on the surrounding states, and the graph neural network has practical significance for the performance prediction research of the network.
Disclosure of Invention
In order to solve the defects of the prior art, realize the prediction of the bandwidth on each link under different network topology structures and avoid the time and labor consumption of actual measurement, the invention provides a network bandwidth prediction method based on a graph neural network.
A network bandwidth prediction method based on a graph neural network comprises the following steps:
s101, data acquisition, namely acquiring data under different network topology structures, wherein the data comprises the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, link connection relation between nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and other network performance related information;
s102, constructing a characteristic project, preprocessing acquired data, constructing a characteristic vector of a network node according to information of a network topology node, and constructing a characteristic vector of a link according to information of each link between nodes;
s103, abstract representation of a network system, wherein the network is abstracted into a graph represented by nodes and edges, the weights of the nodes are assigned by the feature vectors of the nodes, and the edges are assigned by the feature vectors of links among the nodes;
s104, modeling the node characteristics and the link characteristics in the network by using a graph neural network model based on model training of the graph neural network, training to obtain new node characteristics and link characteristic information, and representing the node characteristics and the link characteristic information in the network topology more accurately;
and S105, predicting the network bandwidth, and predicting the bandwidth among the links by adopting a machine learning algorithm according to the updated node characteristics and the link characteristic information.
Further, the data acquisition of S101 includes the following steps:
The star topology is a center and a plurality of sub-nodes. The multiple nodes are connected with the central node in a point-to-point mode.
The mesh topology is such that nodes are interconnected by transmission lines and each node is connected to at least two other nodes.
The top end of the tree-shaped topological structure is provided with a tree root, branches are arranged below the tree root, each branch can be provided with a branch, and the tree root receives data sent by each site and broadcasts the data to the whole network.
The hybrid topology structure is a network topology structure formed by mixing two or more network topology structures.
the network node refers to various devices in a network topology structure, and may be a network device such as a router and a switch, or a computer device such as a server, a desktop, a development board, and the like having a network connection function.
The output size of each data packet in different queues on the node refers to the size of bytes generated by each data packet, and the unit is Byte. The execution scheduling policy of the data packets generated by the different queues includes, but is not limited to, first-in-first-out (FIFO), strict priority scheduling (PQ), weighted fair queue scheduling (WFQ), round robin scheduling (RR), weighted round robin scheduling (WRR), differential round robin scheduling (DRR), and differential weighted round robin scheduling (DWRR).
The weight information only endows the weight value aiming at the condition that the scheduling strategy has weight.
the link is a path connecting every two network nodes, the maximum bandwidth capacity of the link is the maximum bandwidth limit of the full link under different network topological structures, and the packet loss rate on the single link is the ratio of the number of data packets lost in transmission on the single link to the number of data packets sent, and is obtained by measurement; the throughput on the single link is initialized as the maximum bandwidth capacity of the full link divided by the total number of links.
Further, the construction characteristic engineering of S102 includes the following steps:
s301, constructing a feature vector x of the network node according to the node information of the network topology v Including the number of queues queue _ num on the node, the length of each queue _ length, the size of the output data packet queue _ size under each queue, the packet scheduling policy schedule _ policy of the queue, and scheduling weight information schedule _ weight which may be generated under different policies, and is expressed as
S302, constructing a feature vector x of the network link according to the information of each link between the nodes l Including the maximum bandwidth capacity max _ capacity of the link, the packet loss rate link _ loss on the single link, and the throughput link _ throughput of the single link, denoted as x l ＝{c max ,l loss ,l througput }；
S303, constructing a routing relation matrix R according to the link connection relation between the nodes n×n And n is the number of nodes.
Further, the routing relation matrix of S303 reflects the routing connection relation between the network nodes.
Further, the network system abstract representation of S103 is to abstract a graph G = (V, L) represented by nodes and edges according to a topology structure, and represent any node i in the topology as V = (V, L) i The nodes are combined into
n v The feature vector of a node is represented as ^ er for the number of nodes>
Is the weight value of the node. Denote any link j in the topology as l j Link set is taken as>
n l For the number of edges, the feature vector of the link is represented as
Is the weight value of the edge.
Further, the graph neural network-based model training of S104 includes the following steps:
s401, obtaining an input node feature vector x v And link feature vector x l ；
The node feature vector x v And link feature vector x l The node feature vector x constructed for the S301 v And the link characteristic vector x constructed by the S302 l 。
S402, converting the input feature vector into a fixed length expression in a coding mode;
the encoding is to convert the high-dimensional feature vectors with different lengths into fixed-length graph embedding expressions with low dimensions, and then to perform subsequent training and updating on the basis.
S403, starting from a target node/edge, randomly sampling neighbor nodes/edges, setting a threshold value S and a sampling hop number K, wherein the number of neighbors sampled in each hop is not more than S, and if the first hop samples 3 neighbor nodes/edges, the second hop samples 2 neighbor nodes/edges;
s404, generating the embedded expression of the target node/edge: aggregating the characteristics of the K-th hop neighbor to generate the embedded expression of the K-1 hop neighbor, and then aggregating the embedded expression of the K-1 hop until the embedded expression of the target node/edge is generated;
and S405, updating the node characteristics and the link characteristics according to the embedded expression of the target node/edge generated in S404.
The graph neural network-based model training process of S104 can be summarized as follows:
1. randomly sampling neighbor nodes/edges of each node/edge in the graph
2. Aggregating feature information of neighbor nodes/edges according to aggregation function
3. Obtaining new embedded expression of each node/edge in the graph for use by downstream tasks
Further, the network bandwidth prediction of S105 is to predict the bandwidth of the link between the nodes by modeling the updated node characteristics and link characteristics of S405 using a machine learning algorithm. The machine learning algorithm includes: neural networks, GBDT, random forests, support vector machine regression, etc.
The invention also includes a network bandwidth prediction device based on the graph neural network, which comprises: the data acquisition module, the characteristic engineering module, the network system abstraction representation module, the graph neural network model training module and the network bandwidth prediction module that connect gradually, its characterized in that:
the data acquisition module acquires data under different network topology structures, including the number of queues on each network node, the length of each queue and the output size of each data packet, and information on implementation scheduling strategies and weights, link connection relations between nodes, the maximum bandwidth capacity of links, packet loss rate on links and other network performance related information;
the characteristic engineering module is used for preprocessing the acquired data, constructing a characteristic vector of a network node according to the information of the network topology node, and constructing a characteristic vector of a link according to the information of each link between the nodes;
the network system abstraction representation module abstracts the network into a graph represented by nodes and edges, wherein the weights of the nodes are assigned by the feature vectors of the nodes, and the edges are assigned by the feature vectors of the links between the nodes;
the graph neural network model training module is used for modeling the node characteristics and the link characteristics in the network by utilizing the graph neural network model, training to obtain new node characteristic and link characteristic information and representing the node characteristics and the link characteristic information in the network topology more accurately;
and the network bandwidth prediction module predicts the bandwidth between the links by adopting a machine learning algorithm according to the updated node characteristics and the link characteristic information.
The invention has the advantages and beneficial effects that:
when network bandwidth prediction is carried out, the existing modeling technology cannot realize accurate estimation of the network bandwidth, and the problems of huge calculation amount, ideal model construction, incomplete network information consideration and the like exist. In order to solve the defects of the prior art, the network system is abstracted and modeled, a complex network topological structure is converted into graph data for description, the neighbor nodes are randomly sampled by adopting an algorithm based on a graph neural network and feature information of the neighbor nodes is gathered for multiple times, so that the accurate expression of the feature information of the network nodes and edges is realized, and the accurate prediction of the network bandwidth is further completed by a post-connected machine learning algorithm.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a flow chart of the operation of the data collection module in an embodiment of the present invention.
Fig. 3 is a schematic diagram of network model data flow in the embodiment of the present invention.
FIG. 4 is a flow chart of the operation of the feature engineering module in an embodiment of the present invention.
FIG. 5 is a flowchart illustrating operation of a model training module based on a graph neural network according to an embodiment of the present invention.
Fig. 6 is a schematic diagram of information convergence visualization of the neural network model in the embodiment of the present invention.
Fig. 7 is a schematic diagram of the structure of the device of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are given by way of illustration and explanation only, not limitation.
A network bandwidth prediction method based on a graph neural network is characterized in that a convolutional neural network is good at processing images, the core concept behind the convolutional neural network is a convolutional layer, and spatial local features are identified through a set of kernel-form receptive fields. However, convolutional neural networks can only process structured data, and are difficult to process for unstructured graph data. The Graph Neural Network (GNN) is a deep learning method which is excellent in graph data performance, can realize the prediction of nodes, edges or graphs, and can realize the task which cannot be processed by the traditional Convolutional Neural Network (CNN). For unstructured data such as a communication network, modeling can be performed by adopting a graph neural network method, the connection relation between network equipment and the equipment is mapped to nodes and edges of a graph, the nodes and the edges are mapped to a low-dimensional embedding space, the characteristic vector expressions of the nodes and the edges reflect various performance parameters of the network, and more accurate embedding expressions of the nodes and the edges are obtained through a series of graph convolution, graph aggregation, graph updating and other operations, so that the bandwidth condition of each link in the network is further predicted by adopting a machine learning method. The method can improve the prediction precision of the network bandwidth after prediction.
Specifically, as shown in fig. 1, the method of the present invention comprises the steps of:
s101, data acquisition, namely acquiring data under different network topology structures, including the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, link connection relation between nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and other network performance related information.
In this embodiment, as shown in fig. 2, the data acquisition includes the following steps:
specifically, as shown in fig. 3, the network node information in step 202 includes:
the different network devices may be network devices such as a router and a switch, or may be computer devices such as a server, a desktop, and a development board having a network connection function.
Each network device corresponds to a network node, and each network device has a certain number of queues. For example, in fig. 3, there are 4 queues on network device 1, 3 queues on network device 2, and 2 queues on network device n, and each queue has a computation task queued for execution. Some calculation tasks need low time delay, require small cache and are scheduled preferentially; if some calculation tasks need zero packet loss, the larger the cache is, the better the cache is; some pursue network throughput and utilization; some pursuits of fairness require that queue resources be distributed as evenly as possible. Therefore, data packets generated by different queues may correspond to different execution scheduling policies according to computing task requirements, including but not limited to first-in first-out (FIFO), strict priority scheduling (PQ), weighted fair queue scheduling (WFQ), round robin scheduling (RR), weighted round robin scheduling (WRR), differential round robin scheduling (DRR), and differential weighted round robin scheduling (DWRR), where in the case of weighting the scheduling policies, different weights need to be given to the corresponding queues, for example, for fair weighted queue scheduling, weighted round robin scheduling, and differential weighted round robin scheduling, there is a weight list associated with each queue.
the link refers to a path for connecting every two network nodes, the maximum bandwidth capacity of the link refers to the maximum bandwidth limit of the full link under different network topology structures, and the packet loss rate on the single link refers to the ratio of the number of lost data packets in the transmission on the single link to the sent data packets, and is obtained by measurement; the throughput on the single link is initialized as the maximum bandwidth capacity of the full link divided by the total number of links.
There is data transmission between each connected device, and as shown in fig. 3, the bandwidth on each link is different, so that accurate bandwidth information needs to be obtained through repeated updating. The different queues on different devices may have the same type of data packets, such as data packets of queue 1 of network device 1 and queue 3 of network device 2, and data packets of queue 2 of network device 1 and queue 1 of network device 2, belonging to the same type of data packets.
S102, constructing a feature project, preprocessing acquired data, constructing a feature vector of a network node according to information of network topology nodes, and constructing a feature vector of a link according to information of each link between the nodes;
in this embodiment, as shown in fig. 4, the constructing of the feature engineering includes the following steps:
s301, constructing a feature vector x of the network node according to the node information of the network topology v Including the number of queues queue _ num on the node, the length of each queue _ length, the size of the output data packet queue _ size under each queue, the packet scheduling policy schedule _ policy of the queue, and scheduling weight information schedule _ weight which may be generated under different policies, and is expressed as
S302, constructing a feature vector x of the network link according to the information of each link between the nodes l Including the maximum bandwidth capacity max _ capacity of the link, the packet loss rate link _ loss on the single link, and the throughput link _ throughput of the single link, denoted as x l ＝{c max ,l loss ,l througput }；
S303, constructing a routing relation matrix R according to the link connection relation between the nodes n×n And n is the number of nodes.
Further, the routing relation matrix of S303 reflects the routing connection relation between the network nodes.
Further, the abstract representation of the network system of S103 is abstracted into a node and an edge table according to a topology structure for a given networkThe graph G = (V, L), representing any node i in the topology as V i The nodes are combined into
n v The feature vector of a node is represented as ^ er for the number of nodes>
And is the weight value of the node. Denote any link j in the topology as l j Link set is taken as>
n l For the number of edges, the feature vector of the link is represented as
The weight value of the edge.
In this embodiment, as shown in fig. 5, the model training based on the graph neural network includes the following steps:
s401, obtaining an input node feature vector x v And link feature vector x l ；
The node feature vector x v And link feature vector x l The node feature vector x constructed for the S301 v And the link characteristic vector x constructed by the S302 l 。
S402, converting the input feature vector into a fixed length expression in a coding mode;
the encoding is to convert the high-dimensional feature vectors with different lengths into fixed-length graph embedding expressions with low dimensions, and then to perform subsequent training and updating on the basis.
S403, starting from a target node/edge, randomly sampling neighbor nodes/edges, setting a threshold value S and a sampling hop number K, wherein the number of neighbors sampled in each hop is not more than S, and if the first hop samples 3 neighbor nodes/edges, the second hop samples 2 neighbor nodes/edges;
s404, generating the embedded expression of the target node/edge: aggregating the characteristics of the K-th hop neighbor to generate the embedded expression of the K-1 hop neighbor, and then aggregating the embedded expression of the K-1 hop until the embedded expression of the target node/edge is generated;
and S405, updating the node characteristics and the link characteristics according to the embedded expression of the target node/edge generated in S404.
Specifically, as shown in fig. 6, the operation flow of the neural network mainly includes the following three steps:
graph 6.a randomly samples neighbor nodes/edges of each node/edge in the graph
Graph 6.b aggregates feature information implied by neighbor nodes/edges according to aggregation function
The graph 6.c gets a new embedded representation of each node/edge in the graph for use by downstream tasks
The method adopts the concept of graph sage in the graph neural network, all neighbor nodes and edges are not required to be aggregated, but in consideration of computational efficiency, a certain number of neighbor nodes/edges are sampled for each node/edge to serve as nodes/edges of information to be aggregated, for example, K in the graph 6.a refers to the number of layers of the network and represents the hop count of the neighbor nodes which can be aggregated by each node, and each node can learn the embedded expression of the node according to the information of the 2-hop neighbor nodes when K = 2.
FIG. 6.b reflects the aggregation process for one of the node information. Assuming that K times of aggregation is needed, K aggregation functions are needed, and each time of aggregation, features of neighboring nodes of each sample obtained by a previous layer are aggregated once to obtain features of the layer. After repeating the aggregation K times, the final feature of the node, namely the embedded expression of 6.c of the graph is obtained.
The aggregation function may employ an averaging aggregator, an LSTM aggregator, and a pooling aggregator.
Further, the network bandwidth prediction of S105 is to predict the bandwidth of the link between the nodes by modeling the updated node characteristics and link characteristics of S405 using a machine learning algorithm. The machine learning algorithm includes: neural networks, GBDT, random forests, support vector machine regression, etc.
As shown in fig. 7, a network bandwidth prediction apparatus based on a graph neural network includes: the device comprises a data acquisition module, a feature engineering module, a network system abstraction representation module, a graph neural network model training module and a network bandwidth prediction module which are connected in sequence, and is characterized in that:
the data acquisition module acquires data under different network topological structures, wherein the data include the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, link connection relation between the nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and other network performance related information;
the characteristic engineering module is used for preprocessing the acquired data, constructing a characteristic vector of a network node according to the information of the network topology node, and constructing a characteristic vector of a link according to the information of each link between the nodes;
the network system abstraction representation module abstracts the network into a graph represented by nodes and edges, wherein the weights of the nodes are assigned by the feature vectors of the nodes, and the edges are assigned by the feature vectors of the links between the nodes;
the graph neural network model training module is used for modeling the node characteristics and the link characteristics in the network by utilizing the graph neural network model, training to obtain new node characteristic and link characteristic information and representing the node characteristics and the link characteristic information in the network topology more accurately;
and the network bandwidth prediction module predicts the bandwidth between the links by adopting a machine learning algorithm according to the updated node characteristics and the link characteristic information.
As shown in fig. 7, at the hardware level, the network bandwidth prediction apparatus based on the neural network includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, but may also include hardware required by other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the method described in fig. 1 above. Of course, besides the software implementation, the present invention does not exclude other implementations, such as logic devices or combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
Improvements to a technology can clearly be distinguished between hardware improvements (e.g. improvements to the circuit structure of diodes, transistors, switches, etc.) and software improvements (improvements to the process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is an integrated circuit whose Logic functions are determined by a user programming the Device. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as ABEL (Advanced Boolean Expression Language), AHDL (alternate Hardware Description Language), traffic, CUPL (core universal Programming Language), HDCal, jhddl (Java Hardware Description Language), lava, lola, HDL, PALASM, rhyd (Hardware Description Language), and vhigh-Language (Hardware Description Language), which is currently used in most popular applications. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the units may be implemented in one or more of software and/or hardware in implementing the invention.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
All the embodiments in the invention are described in a progressive manner, and the same and similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present invention and is not intended to limit the present invention. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present 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 solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A network bandwidth prediction method based on a graph neural network and machine learning is characterized by comprising the following steps:
s101, data acquisition, namely acquiring data under different network topological structures, wherein the data comprises the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, link connection relation between nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and other network performance related information;
s102, constructing a feature project, preprocessing acquired data, constructing a feature vector of a network node according to information of network topology nodes, and constructing a feature vector of a link according to information of each link between the nodes;
s103, abstract representation of a network system, wherein the network is abstracted into a graph represented by nodes and edges, the weights of the nodes are assigned by the characteristic vectors of the nodes, and the edges are assigned by the characteristic vectors of links between the nodes;
s104, modeling the node characteristics and the link characteristics in the network by using a graph neural network model based on model training of the graph neural network, training to obtain new node characteristics and link characteristic information, and representing the node characteristics and the link characteristic information in the network topology more accurately;
and S105, predicting the network bandwidth, and predicting the bandwidth between the links by adopting a machine learning algorithm according to the updated node characteristics and the link characteristic information.
2. The method of claim 1, wherein the network topology of step S101 is selected from a star topology, a mesh topology, a tree topology, and a hybrid topology.
3. The network bandwidth prediction method based on the graph neural network as claimed in claim 1, wherein the data collection of S101 includes the following steps:
step 201, collecting various network topology structures;
step 202, collecting network node information under different network topology structures, including: the number of queues on the node, the length of each queue and the output size of each data packet, and the execution scheduling strategy and weight information of the data packets generated by different queues on the node;
step 203, collecting link information under different network topology structures, including: maximum bandwidth capacity of the link, packet loss rate on the single link, and throughput of the single link.
4. The method according to claim 1, wherein the step of constructing feature engineering of S102 comprises the following steps:
s301, constructing a feature vector x of the network node according to the node information of the network topology v The method comprises the number of queues queue _ num on a node, the length of each queue _ length, the size of an output data packet queue _ size under each queue, a data packet scheduling policy schedule _ policy of the queues, and scheduling weight information schedule _ weight which can be generated under different policies, wherein the scheduling weight information schedule _ weight is represented as
S302, constructing a feature vector x of the network link according to the information of each link between the nodes l The method comprises the maximum bandwidth capacity max _ capacity of the link, the packet loss rate link _ loss on the link, and the throughput link _ throughput, which is denoted as x l ＝{c max ,l loss ,l througput }；
S303, constructing a routing relation matrix R according to the link connection relation between the nodes n×n And n is the number of nodes.
5. The method according to claim 1, wherein the step S103 comprisesThe network system abstract representation is that for a given network, a graph G = (V, L) represented by nodes and edges is abstracted according to a topological structure, and any node i in the topology is represented as V i The nodes are combined into
n v The feature vector of a node is represented as ^ er for the number of nodes>
Is the weight value of the node. Denote any link j in the topology as l j Link set is taken as>
n l For the number of edges, the feature vector of the link is expressed as ≥>
Is the weight value of the edge.
6. The method according to claim 1, wherein the graph neural network-based network bandwidth prediction method of S104 is characterized in that: a graph convolution neural network based on neighbor node aggregation, comprising: the system comprises an input layer, an embedding layer, a neighbor node sampling layer, a neighbor node convergence layer and an output layer.
7. The method according to claim 1, wherein the model training of S104 is to convert a network bandwidth prediction problem into an edge prediction problem in a graph, and the specific steps include:
s401, obtaining an input node feature vector x v And link feature vector x l ；
S402, converting the input feature vector into a fixed length expression in a coding mode;
s403, starting from a target node/edge, randomly sampling neighbor nodes/edges, setting a threshold value S and a sampling hop number K, wherein the number of neighbors sampled in each hop is not more than S, and if the first hop samples 3 neighbor nodes/edges, the second hop samples 2 neighbor nodes/edges;
s404, generating the embedded expression of the target node/edge: aggregating the characteristics of the K-th hop neighbor to generate the embedded expression of the K-1 hop neighbor, and then aggregating the embedded expression of the K-1 hop until the embedded expression of the target node/edge is generated;
and S405, updating the node characteristics and the link characteristics according to the embedded expression of the target node/edge generated in S404.
8. The method of claim 7, wherein the network bandwidth prediction of S105 is that the updated node characteristics and link characteristics of step S404 are modeled by a machine learning algorithm to predict the bandwidth of the link between nodes.
9. The method of claim 8, wherein the machine learning algorithm is selected from the group consisting of: neural networks, GBDT, random forests, support vector machine regression.
10. A network bandwidth prediction device based on a graph neural network comprises: the data acquisition module, the characteristic engineering module, the network system abstraction representation module, the graph neural network model training module and the network bandwidth prediction module that connect gradually, its characterized in that:
the data acquisition module acquires data under different network topological structures, wherein the data include the number of queues on each network node, the length of each queue, the output size of each data packet, an execution scheduling strategy and weight information, link connection relation between the nodes, the maximum bandwidth capacity of a link, packet loss rate on the link and other network performance related information;
the characteristic engineering module is used for preprocessing the acquired data, constructing a characteristic vector of a network node according to the information of the network topology node, and constructing a characteristic vector of a link according to the information of each link between the nodes;
the network system abstraction representation module abstracts a network into a graph represented by nodes and edges, wherein the weights of the nodes are assigned by the characteristic vectors of the nodes, and the edges are assigned by the characteristic vectors of the links between the nodes;
the graph neural network model training module is used for modeling the node characteristics and the link characteristics in the network by utilizing the graph neural network model, training to obtain new node characteristic and link characteristic information and representing the node characteristics and the link characteristic information in the network topology more accurately;
and the network bandwidth prediction module predicts the bandwidth between the links by adopting a machine learning algorithm according to the updated node characteristics and the link characteristic information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211495900.9A CN115935563A (en) | 2022-11-24 | 2022-11-24 | Network bandwidth prediction method and device based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211495900.9A CN115935563A (en) | 2022-11-24 | 2022-11-24 | Network bandwidth prediction method and device based on graph neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115935563A true CN115935563A (en) | 2023-04-07 |
Family
ID=86648278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211495900.9A Pending CN115935563A (en) | 2022-11-24 | 2022-11-24 | Network bandwidth prediction method and device based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115935563A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117151279A (en) * | 2023-08-15 | 2023-12-01 | 哈尔滨工业大学 | Isomorphic network link prediction method and system based on line graph neural network |
-
2022
- 2022-11-24 CN CN202211495900.9A patent/CN115935563A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117151279A (en) * | 2023-08-15 | 2023-12-01 | 哈尔滨工业大学 | Isomorphic network link prediction method and system based on line graph neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mao et al. | Routing or computing? The paradigm shift towards intelligent computer network packet transmission based on deep learning | |
| CN114286413B (en) | TSN network joint routing and stream distribution method and related equipment | |
| Li et al. | DeepNFV: A lightweight framework for intelligent edge network functions virtualization | |
| CN112422443B (en) | Adaptive control method, storage medium, equipment and system of congestion algorithm | |
| CN112532530B (en) | Method and device for adjusting congestion notification information | |
| CN111245718A (en) | Routing optimization method based on SDN context awareness | |
| CN115935563A (en) | Network bandwidth prediction method and device based on graph neural network | |
| CN116680062B (en) | Application scheduling deployment method based on big data cluster and storage medium | |
| CN115665227B (en) | Universal heterogeneous integrated computing network resource intelligent adaptation network architecture and method | |
| CN111404815B (en) | Constrained routing method based on deep learning | |
| CN112910789A (en) | Congestion control method and related equipment | |
| Zanbouri et al. | A new fog-based transmission scheduler on the Internet of multimedia things using a fuzzy-based quantum genetic algorithm | |
| CN116185596B (en) | Method and device for improving task execution efficiency of wide-area multi-cluster distributed system | |
| Wei et al. | Drl-deploy: adaptive service function chains deployment with deep reinforcement learning | |
| CN114422453B (en) | Method, device and storage medium for online planning of time-sensitive stream | |
| CN115567466A (en) | Time-sensitive traffic online scheduling method and device based on deep reinforcement learning | |
| Zhang et al. | Accelerate deep learning in IoT: Human-interaction co-inference networking system for edge | |
| CN112333102B (en) | Software defined network routing method and system based on knowledge graph | |
| Hsieh et al. | Deep reinforcement learning-based task assignment for cooperative mobile edge computing | |
| CN116418808A (en) | Combined computing unloading and resource allocation method and device for MEC | |
| CN109756386B (en) | Communication optimization method and device based on cloud network vehicle cooperation | |
| Abdulazzaq et al. | A deep learning based system for traffic engineering in software defined networks | |
| Liu et al. | Scalable Deep Reinforcement Learning-Based Online Routing for Multi-Type Service Requirements | |
| WO2020051918A1 (en) | Neuronal circuit, chip, system and method therefor, and storage medium | |
| Yang et al. | A Multi-Policy Deep Reinforcement Learning Approach for Multi-Objective Joint Routing and Scheduling in Deterministic Networks |
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 |