CN113301169B - Edge network switching method based on dynamic mobile device behavior prediction - Google Patents
Edge network switching method based on dynamic mobile device behavior prediction Download PDFInfo
- Publication number
- CN113301169B CN113301169B CN202110834476.5A CN202110834476A CN113301169B CN 113301169 B CN113301169 B CN 113301169B CN 202110834476 A CN202110834476 A CN 202110834476A CN 113301169 B CN113301169 B CN 113301169B
- Authority
- CN
- China
- Prior art keywords
- mobile device
- edge computing
- computing server
- edge
- server
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- 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]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- 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]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
- G06F9/5088—Techniques for rebalancing the load in a distributed system involving task migration
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/54—Presence management, e.g. monitoring or registration for receipt of user log-on information, or the connection status of the users
Abstract
An edge network switching method based on dynamic mobile device behavior prediction belongs to the technical field of flow control and comprises the following steps: step S1, establishing an edge node architecture; step S2, establishing a first-layer service handshake linkage mechanism; then, simultaneously executing the step 3 and the step 4; step S3, if the number of mobile devices served by the edge computing server is overloaded, adjusting the deployment of the mobile devices; step S4, establishing and updating a movement prediction model: adopting a hidden Markov model and redefining parameters; step S5, the edge calculation server judges whether the mobile device moves; step S6, the edge computing server makes the mobile device and the target edge computing server establish service handshake; and the edge computing server triggers a second-time service handshake linkage mechanism. According to the scheme, the hidden Markov model is adopted to predict the network switching trend of the mobile device, so that the edge computing server for preparing service in advance is reduced.
Description
Technical Field
The invention belongs to the technical field of flow control, and particularly relates to an edge network switching method based on dynamic mobile device behavior prediction.
Background
With the development of the internet of things, mobile devices such as mobile phones and tablets have become important productivity tools. However, due to the limited resources of mobile devices, the demands of being computationally intensive, delay sensitive, and bandwidth cannot be met. Therefore, a Mobile Cloud Computing (MCC) method is proposed, which transmits tasks requiring more computing resources to a cloud data center for processing. However, long distance transmission between the mobile device and the cloud data center will result in high latency, jitter, and network congestion, which is detrimental to the core network.
Moving Edge Computing (MEC) has been proposed in recent years. The mobile edge computing adopts a three-layer structure and comprises a cloud data center, an edge computing server and a mobile device. An Edge Computing Server (ECS) is located at the end of the core network and has the same Computing service function as the data center, so that it is closer to the mobile device. The mobile device can use the calculation service function of the edge calculation server nearby, meet the requirements of timeliness calculation and content access, shorten network delay and increase the stability of communication.
However, most of the existing research on Mobile Edge Computing (MEC) assumes a quasi-static scenario, but the mobility of the ue cannot be ignored in the actual application scenario. For example, the mobile device may wander to different local area networks, and when the mobile device performs network switching in two local area networks, how the new edge computing server and the old edge computing server should cooperate with each other can implement seamless service migration, thereby reducing the time gap of handshaking and ensuring that no service interruption is caused in the network switching process.
Chinese patent publication No. CN111049917A discloses a mobile-aware multi-user offload optimization method based on mobile edge calculation, which converts an original global optimization problem into multiple local optimization problems, and decomposes the local optimization problems into sub-problems for solution. Finally, better performance can be obtained compared with other technologies under the condition of better computation complexity. However, this solution only finds an optimal offloading scheme to optimize user utility, and makes a trade-off between task latency and energy consumption, which cannot solve the problem of switching time faced by the dynamic mobile device during network switching.
Chinese patent publication No. CN113037805A discloses a multi-instance micro-service migration method for mobile edge computing, which determines a migration order of micro-services to be migrated according to a policy of large-priority migration of storage resources, then selects a server node satisfying resource constraints and having the minimum migration delay and communication delay as a target migration node for each micro-service according to the migration order, sequentially migrates the server nodes, and finally adjusts the node introducing extra delay to eliminate the extra delay. However, this solution only eliminates the communication delay, and cannot overcome the problem of short service interruption when the dynamic mobile device is switched to the network.
Disclosure of Invention
In view of the above-mentioned shortcomings of the prior art, the present invention is directed to an edge forwarding method based on dynamic mobile device behavior prediction.
In order to achieve the above object, the present invention adopts the following technical solutions.
An edge network switching method based on dynamic mobile device behavior prediction comprises the following steps:
step S1, establishing an edge node architecture: the edge node architecture comprises a cloud end, an edge computing server and a mobile device; the edge computing server is in communication connection with the cloud end; the mobile device and the edge computing server are in communication connection to form a local area network; the edge computing servers are connected with wireless base stations, and the wireless base stations responsible for any two edge computing servers are not crossed;
step S2, establishing a first layer of service handshake linkage mechanism, and tracking the state of the mobile equipment in the local network to which the edge computing server belongs by the edge computing server; the mobile device collects information including hardware, network and position, establishes an information collection list and uploads the information collection list to the edge computing server periodically; the edge computing server uploads the information of all the mobile devices to the cloud terminal for gathering;
then, simultaneously executing the step 3 and the step 4;
step S3, if the number of mobile devices served by the edge computing server is overloaded, adjusting the deployment of the mobile devices;
step S4, establishing and updating a behavior prediction model: adopting a hidden Markov model and redefining parameters to obtain a behavior prediction model;
in the behavior prediction model, a hidden state sequence I is defined as a sequence of the actual position state of the mobile device; an observation state sequence O defined as a sequence of states of the radio base stations, the sequence being a set of all radio base stations observing states; a state transition matrix A, defined as a position state transition probability matrix of the mobile device, representing a state transition matrix of hidden state transition probability; a probability matrix B generated by observing the state is defined as a state transition probability matrix of the wireless base station and represents a set of transition probabilities between a hidden state of the mobile device and an observation state of the wireless base station; hidden state probability distribution pi defined as the initial state probability of the wireless base station, and setting the parameter as 1;
step S5, the edge calculation server judges whether the mobile device moves; if the mobile device moves, calculating a target edge calculation server on a future moving path of the mobile device through a behavior prediction model, otherwise, repeating the step;
step S6, the edge computing server transmits the networking information of the target edge computing server to the mobile device after moving, and the mobile device sends the connection information to the target edge computing server, so that service handshake is established between the mobile device and the target edge computing server;
and the edge computing server triggers a service handshake linkage mechanism of a second layer, namely, the current edge computing server and the target edge computing server establish service handshake, inform the target edge computing server of starting the preloading service and transmit resource information required by the preloading service.
The cloud is configured with a database for storing information of all edge computing servers and information of all mobile devices in the framework and is responsible for finishing training of a behavior prediction model of the mobile device;
the cloud is provided with an architecture deployment unit, a mobile device behavior prediction model training unit and a database;
the architecture deployment unit adjusts the deployment of the whole architecture, calls the mobile device information in the database, and tracks the average response time between each mobile device and the edge calculation server connected with the mobile device correspondingly;
the mobile device behavior prediction model training unit is used for constructing and training a behavior prediction model of the mobile device, and updating the behavior prediction model of the mobile device after training is finished;
the database adopts a relational database management system to record all edge computing server information and all mobile device information in the framework, including deployment and networking information of the computing server information, state and networking information of the mobile device and historical moving paths of the mobile device.
The edge computing server provides resource service for nearby mobile devices, grasps the movement behaviors of the nearby mobile devices, and is in communication connection with the edge computing server nearby;
the edge computing server is provided with an architecture deployment auxiliary unit, a service providing unit, a state tracking unit, a mobile device behavior prediction unit, a preloading communication unit, a network switching handshake unit and a lightweight database.
The mobile device is user equipment for generating request data or sensing data; the mobile device is provided with a service receiving unit, a network switching holding unit and an information acquisition unit.
In step 2, the information collection list includes: a device name field, a status field, an edge calculation server field, a network information field, a location field, a sending time field, and a historical movement path field;
a device name field for recording the name of the mobile device;
the state column records the current running state of the mobile device and is divided into a normal state, a transition state and a closing state;
an edge computing server column that records an edge computing server currently deployed by the mobile device; after the mobile device is switched to the network, the location of the current column records the new edge calculation server.
A network information column for recording the name, IP address, MAC address and port of the wireless base station connected with the mobile device;
a location field recording the longitude and latitude of the mobile device;
sending a time column, recording the uploading time of the information acquisition list, wherein the time is accurate to millisecond;
and the historical moving path column records the wireless base stations which are historically connected with the mobile device by taking the numerical sequence number as an identifier.
In the step 3, the method also comprises the following steps:
step S301, an edge computing server obtains deployment information of adjacent edge computing servers from a cloud; the edge computing server judges that the number of the mobile devices served by the edge computing server currently exceeds the load; if yes, executing the next step; if not, repeating the step;
step S302, calculating the distance between the current edge calculation server and the adjacent edge calculation server, and then selecting the nearest edge calculation server which does not exceed the load as a target edge calculation server;
step S303, calculating a distance between the target edge calculation server and each mobile device in the local area network according to the position of the target edge calculation server and the position of each mobile device in the local area network, selecting the mobile device closest to the target edge calculation server as the mobile device for network switching, and then executing step 6 to establish a service connection between the mobile device for network switching and the target edge calculation server.
Step 6, the target edge computing server judges whether the target edge computing server has resource information required by the preloading service; if not, the target edge computing server requests the current edge computing server to acquire the resource through a channel between the current edge computing server and the target edge computing server, the resource is loaded after the resource is acquired, and the target edge computing server acquires the service providing capacity; if yes, the target edge computing server replies that the current edge computing server stops preloading the service flow.
According to the scheme, on the basis of a traditional three-layer architecture of mobile edge computing, organization units are additionally arranged on a cloud end, an edge computing server and a mobile device according to functions and processes required by edge transfer, service resources are deployed from appropriate devices, resource waste is reduced, and the response time of a network is favorably shortened.
The method adopts a double-layer service handshake linkage mechanism, not only applies a first layer of service handshake linkage mechanism between a current edge computing server and a mobile device and judges the position of the mobile device by uploading an information acquisition list, but also applies a second layer of service handshake linkage mechanism between the original edge computing server and a target edge computing server.
According to the scheme, a hidden Markov model is adopted to predict the network switching trend of the mobile device, in order to reduce the complexity of the model, the number of the edge calculation server is used as a historical moving path of the mobile device, and then the edge calculation server on a future possible path is predicted according to the historical moving path and the current edge calculation server, so that the edge calculation server which prepares for service in advance is reduced, a target edge calculation server is not selected in a random way, and the preloading burden of the whole framework is reduced.
According to the scheme, the first layer of service handshake linkage mechanism is implemented between the current edge computing server and the mobile devices, so that the edge computing server can master the number of the mobile devices in the local area network, and the load balance of each edge computing server is adjusted.
In the scheme, the edge computing server grasps the information of the adjacent edge computing servers, so that the edge computing server actively detects whether the mobile device moves. And if the mobile device moves, sending the information of the target edge computing server to the mobile device, requiring the mobile device to establish a connection line to the target edge computing server, and starting a first-layer service handshake linkage mechanism between the mobile device and the target edge computing server.
Drawings
FIG. 1 is an exemplary diagram of a network architecture of the present invention;
FIG. 2 is an architecture diagram of a cloud;
FIG. 3 is an architecture diagram of an edge compute server;
FIG. 4 is an architecture diagram of a mobile device;
fig. 5 is an accuracy test chart of the behavior prediction model.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
Because most of the services or computations are placed in the edge computing server under the MEC architecture and do not need to be transferred with the mobile device, however, some services or computations need to maintain connection in real time and transfer data at any time, if the mobile device switches to the local network, the services or computations needed by the mobile device need to be transferred in advance, so as to maintain the continuity of data transfer.
To accommodate dynamic wandering of mobile devices, a solution is foreseen:
the first scheme is as follows: in the time limit that the edge computing server provides the service task or the computing task, the mobile device is required to be always in the original local area network, and the mobile device can not switch the local area network until the service task or the computing task is finished.
Although the solution maintains the continuity of the service task or the computing task provided by the edge computing server as much as possible, the dynamic wandering of the mobile device may cause the communication quality between the mobile device and the edge computing server in the original office network to be reduced, and may adversely affect the information transfer between the mobile device and the edge computing server.
Scheme II: in the time limit of the service task or the computing task provided by the edge computing server, the mobile device informs the original edge computing server of switching the local area network, and requires the original edge computing server to transfer the task which is not completed to a new edge computing server, and then the new edge computing server continues to provide the service.
In this scheme, the mobile device is in the dominance of network switching, and therefore, the mobile device is required to master the real-time status of the whole system architecture (including the edge computing servers of the respective local networks), which may impose a heavy burden on the mobile device. Moreover, the migration of the required data when the mobile device requests the network switch will cause the delay and even interruption of the service.
The third scheme is as follows: and in the time limit that the edge computing server provides the service task or the computing task, the mobile device leaves the original local area network domain, uses 4G or 5G to communicate with the cloud data center instead, informs the original edge computing server to transfer the incomplete task to the cloud data center, and the cloud data center continues to provide the service.
In the scheme, the mobile device directly communicates with the cloud data center, and the communication mode returns to the architecture of Mobile Cloud Computing (MCC), so that the mobile device has the disadvantage of higher network delay compared with Mobile Edge Computing (MEC).
In the scheme of the application, the target edge computing server after the mobile device is possibly switched with the local area network at the current position is predicted by learning the historical moving path of the mobile device, so that resources which are possibly used are deployed on the target edge computing server in advance, and the service switching time during network switching is further reduced.
In the scheme, a hidden Markov model is adopted for learning the historical moving path of the mobile device. Hidden Markov models (Hidden Markov models) are relatively classical machine learning models. For this model, first, we assume that Q is the set of all possible hidden states and V is the set of all possible observed states, i.e.:
Q={q1,q2,...,qN}，V={v1,v2,...vM}；
where N is the number of possible hidden states and M is the number of all possible observed states.
For a sequence of length T, I is the corresponding hidden state sequence and O is the corresponding observed state sequence, i.e.:
I={i1,i2,...,iT},O={o1,o2,...oT}；
wherein any one of the hidden states itE.g. Q, any observation state ot∈V。
Hidden markov models, there are two important assumptions:
assume that 1: homogeneous markov property assumption. I.e. the hidden state at any moment in time depends only on its previous hidden state. If the hidden state at time t is it=qiThe hidden state at time t +1 is it+1=qjWhere j is the sequence number of the hidden state corresponding to time t +1, the hidden state transition probability a from time t to time t +1ijExpressed as:
aij=P(it+1=qj|it=qi)。
from a to aijState transition matrix a that constitutes a markov chain:
A=[aij]N×N。
assume 2: observe the independence assumption. I.e. the observed state at any moment only depends on the hidden state at the current moment. If the hidden state at time t is it=qjAnd the corresponding observed state is ot=vkWhere k is the number of the observation state corresponding to the time t, the observation state v at that timekIn a hidden state qjThe probability of lower generation is bj(k) And satisfies the following conditions:
bj(k)=P(ot=vk|it=qj)
b is formed byj(k) Forming a probability matrix B generated by observation states:
B=[bj(k)]N×M。
besides this, we need a set of hidden state probability distributions Π at time t = 1:
Π=[π(i)]Nwherein pi (i) = P (i)1=qi)。
The hidden Markov model is determined by hidden state initial probability distribution pi, a state transition probability matrix A and an observation state probability matrix B. Π, A determines the state sequence and B determines the observation sequence. Thus, the hidden markov model is represented by a triplet λ as follows:
λ=(A,B,Π)。
hidden markov models, solving three classical problems:
question 1, evaluate the observed sequence probability. I.e. given model λ = (a, B, Π) and observation sequence O = { O = { O }1,o2,...oTAnd calculating the probability P (O | lambda) of the appearance of the observation sequence O under the model lambda.
Problem 2, the prediction problem, also called the decoding problem. I.e. given model λ = (a, B, Π) and observation sequence O = { O = { O }1,o2,...oTSolving the most probable corresponding state under the condition of a given observation sequenceAnd (4) sequencing.
Problem 3, model parameter learning problem. I.e. given observation sequence O = { O = { (O) }1,o2,...oTAnd (d) estimating parameters of the model λ = (a, B, Π) so as to maximize the conditional probability P (O | λ) of the observed sequence under the model.
An edge network switching method based on dynamic mobile device behavior prediction comprises the following steps:
step S1, establishing an edge node architecture: the edge node architecture comprises a cloud end, an edge computing server and a mobile device; the edge computing server is in communication connection with the cloud end; the mobile device and the edge computing server are in communication connection to form a local area network; the edge computing servers are connected with wireless base stations, one edge computing server supports at least 1 wireless base station, and the wireless base stations responsible for any two edge computing servers are not crossed.
The cloud is a data center with a large amount of computing capacity and storage space, and can be a public cloud, a private cloud or a hybrid cloud. The cloud is configured with a database for storing information of all edge computing servers and information of all mobile devices in the framework, and is responsible for completing training of behavior prediction models of the mobile devices.
Specifically, as shown in fig. 2, the cloud is provided with a framework deployment unit, a mobile device behavior prediction model training unit, and a database.
The architecture deployment unit is used for adjusting the deployment of the whole architecture, calling the information of the mobile devices in the database, tracking the average response time between each mobile device and the edge computing server connected with the mobile device correspondingly, and reselecting a proper edge computing server for the mobile device according to the current position and network information of the mobile device and the states of the edge computing servers nearby if the average response time between the mobile device and the corresponding edge computing server exceeds a threshold value or the mobile device accesses the network for the first time, so that the situation that the mobile device cannot obtain the service of a new target edge computing server because the mobile device cannot be in butt joint with the target edge computing server is avoided.
The mobile device behavior prediction model training unit is used for constructing and training a behavior prediction model of the mobile device, and updating the behavior prediction model of the mobile device after training is finished. The behavior prediction model of the mobile device is trained, which needs to spend much effort and time, and therefore, the behavior prediction model is deployed in the cloud.
The database adopts a relational database management system (mySQL) to record all edge calculation server information and all mobile device information in the architecture, including deployment and networking information of the calculation server information, state and networking information of the mobile device, historical moving paths of the mobile device and the like.
The edge computing server has larger computing capacity and storage capacity, such as a server or a base station, provides resource service for nearby mobile devices, so as to reduce response time of communication and grasp the moving behavior of the nearby mobile devices, and meanwhile, the edge computing server is also in communication connection with the nearby edge computing server to perform information transfer between adjacent edge computing servers.
The edge computing server is configured with a wireless transmission module, and the wireless transmission module comprises but is not limited to a Wi-Fi module, a Bluetooth module and a ZigBee module. All the edge computing servers are in communication connection with the cloud end, and the communication connection adopts a wired network or a cellular network, which is a reliable communication connection mode; and the edge computing server uploads all data to a cloud for storage.
Specifically, as shown in fig. 3, the edge computing server is provided with an architecture deployment assisting unit, a service providing unit, a state tracking unit, a mobile device behavior prediction unit, a preloaded communication unit, a network switching handshake unit, and a lightweight database.
The architecture deployment auxiliary unit uploads deployment information in the local area network to the cloud end, receives a deployment instruction of the cloud end, and then adjusts deployment of the local area network.
The service providing unit is responsible for establishing communication with the mobile device and providing services for the mobile device.
The state tracking unit receives an information acquisition list containing current position information of the mobile device, and uploads the information acquisition list to the cloud after analysis.
The mobile device behavior prediction unit downloads and loads the latest cloud behavior prediction model of the mobile device, predicts a target edge calculation server possibly connected with the mobile device in real time, and triggers the preloading communication unit to act after prediction is completed.
And the preloading communication unit informs the target edge calculation server of checking whether the target edge calculation server has the resources required by the mobile device according to the behavior prediction result of the mobile device by the mobile device behavior prediction unit, and if not, the edge calculation server connected with the mobile device transmits the required resources to the target edge calculation server through a channel between the edge calculation servers.
And the network switching holding unit informs the mobile device of preparing network switching and transmitting information of the target edge computing server after switching the service channel to the mobile device when the edge computing server detects that the mobile device moves and needs to switch the local area network.
The lightweight database is used for storing the state and information of the mobile device in the office network and the information of the nearby edge computing server.
The mobile device is user equipment for generating request data or sensing data, and can be a mobile phone, a computer, an intelligent bracelet, an intelligent watch and the like. The mobile devices are all in communication connection with the edge computing server, and the communication connection adopts a wireless network, a wired network or a cellular network.
Specifically, as shown in fig. 4, the mobile device is provided with a service receiving unit, a network switching handshake unit, and an information collecting unit.
The service receiving unit corresponds to the service providing unit of the edge computing server and is used for establishing communication with the edge computing server and receiving the service content of the edge computing server.
The network switching handshake lower unit is used for receiving network switching handshake information of the network switching handshake upper unit of the edge computing server, then transferring a channel to the target edge computing server and sending handshake information to the target edge computing server.
The information acquisition unit collects the position information, the networking information and the hardware information of the mobile device, completes an information acquisition list and then periodically sends the information to the edge computing server.
According to the scheme, on the basis of a traditional three-layer architecture of mobile edge computing, organization units are additionally arranged on a cloud end, an edge computing server and a mobile device according to functions and processes required by edge transfer, service resources are deployed from appropriate devices, resource waste is reduced, and the response time of a network is favorably shortened.
Step S2, establishing a first layer of service handshake linkage mechanism, and the edge computing server tracking the state of the mobile device in the local area network to which the edge computing server belongs.
The mobile device collects information including hardware, network and position, establishes an information collection list and uploads the information collection list to the edge computing server periodically. The mobile device uploads the information to the edge computing server periodically, so that the information is real-time, the edge computing server can master the moving behavior of the mobile device, and the operation of the edge computing server such as service pre-adjustment or pre-loading is facilitated when the mobile device is switched to a network.
The edge computing server uploads the information of all the mobile devices to the cloud for gathering, so that on one hand, the cloud can master the movement of the mobile behaviors of all the mobile devices in the whole framework, and on the other hand, the cloud can also train the prediction model after updating the information of the mobile devices.
The information collection list includes: the device comprises a device name field, a state field, an edge calculation server field, a network information field, a location field, a sending time field and a historical moving path field.
A device name field for recording the name of the mobile device.
The status field records the current operating status of the mobile device, and is divided into a normal status, a transition status and a shutdown status.
The edge computing server column records an edge computing server currently deployed by the mobile device. After the mobile device is switched to the network, the location of the current column records the new edge calculation server.
The network information field records the name, address (IP address and MAC address) and Port (Port) of the wireless base station to which the mobile device is connected. The position of the wireless base station can be used as a judgment basis for the movement of the mobile device, and the moving direction of the mobile device can be known according to the position conversion of the wireless base station connected with the mobile device.
The location field records the longitude and latitude of the mobile device for readjusting the disposition of the mobile device in the framework.
And sending a time column, recording the uploading time of the information acquisition list, and enabling the time to be accurate to millisecond. The edge computing server analyzes the communication state of the mobile device by recording the uploading time of the information acquisition list: if the edge computing server does not receive the information collection list for more than a certain time, the edge computing server indicates that the mobile device has interrupted communication currently.
And the historical moving path column records the wireless base stations which are connected with the mobile device in history by taking the numerical sequence number as an identifier, and provides the wireless base stations with the historical moving path column to the current edge computing server to analyze the future moving trend of the wireless base stations.
Since the target edge computing server does not have the historical movement information of the mobile device after the mobile device is switched to the network, the historical movement path needs to be recorded by the mobile device itself.
Table 1 is a table of contents of the information collection list:
after step 2, step 3 and step 4 are performed simultaneously.
Step S3, if the number of mobile devices served by the edge computing server is overloaded, adjusting the deployment of the mobile devices.
Step S301, the edge computing server uploads an information acquisition list to a cloud end, and deployment information of adjacent edge computing servers is obtained from the cloud end; the edge computing server judges that the number of the mobile devices served by the edge computing server currently exceeds the load; if yes, executing the next step; if not, the step is repeated.
Step S302, the distance between the current edge computing server and the adjacent edge computing server is computed, and then the nearest edge computing server which does not exceed the load is selected as the target edge computing server.
Step S303, calculating a distance between the target edge calculation server and each mobile device in the local area network according to the position of the target edge calculation server and the position of each mobile device in the local area network, selecting the mobile device closest to the target edge calculation server as the mobile device for network switching, and then executing step 6 to establish a service connection between the mobile device for network switching and the target edge calculation server.
Step S4, establishing and updating a behavior prediction model:
step S401, a behavior prediction model training unit of the cloud mobile device adopts a hidden Markov model and redefines parameters to obtain a behavior prediction model.
Table 2 is a redefined parameters table:
in the behavior prediction model, the hidden state sequence I is defined as a sequence of actual position states of the mobile device, and the sequence is a set of hidden states of the mobile device and cannot be observed from the outside.
The observation state sequence O is defined as a sequence of states of the radio base stations, and this sequence is a set of all the radio base stations observing the states. This sequence reflects the mobile device movement record sequence, since the mobile device updates the state of the wireless base station to which it is connected after moving.
The state transition matrix a is defined as a position state transition probability matrix of the mobile device, and represents a state transition matrix of hidden state transition probability.
The probability matrix B generated in the observation state is defined as a state transition probability matrix of the radio base station, and represents a set of transition probabilities between the hidden state of the mobile device and the observation state of the radio base station.
The hidden state probability distribution pi is defined as an initial state probability of the radio base station, and since each radio base station is likely to become a first-time connected radio base station in the present configuration, setting this parameter to 1 represents that each state is likely to become an initial state.
The present solution does not care about the actual location of the mobile device, but rather the probability of a transition between edge calculation servers, i.e. the state transition probability of the edge calculation servers. Thus, the problem of evaluating the probability of an observed sequence (i.e., problem 1 in hidden markov models) and the model parameter learning problem (i.e., problem 3 in hidden markov models) are solved with a predictive model.
Step S402, training the parameters λ = (a, B, Π) of the behavior prediction model:
aiming at the problem of model parameter learning, a mobile device movement record sequence is obtained according to the sequence of the state of the wireless base station, a Baum-Welch algorithm is adopted, the parameters lambda = (A, B, Π) of the behavior prediction model are trained, then the updated parameters lambda = (A, B, Π) are adopted, and then a forwarder algorithm is adopted to predict the probability of occurrence of the sequence O of the state of the wireless base station. Parameters are estimated in the hidden Markov model by adopting a Baum-Welch algorithm, the specific use mode is common knowledge, and Chinese patent texts with publication numbers of CN104834995A, CN103902984A and CN109034093A are discussed and are not described any more.
Step S403, the behavior prediction model updating process of the edge computing server:
the mobile device behavior prediction model training unit at the cloud end records the latest model version number in the database after the model training is finished; the mobile device behavior prediction unit of the edge calculation server periodically requests the latest model version number to a cloud database, if the prediction model on the edge calculation server is found not to be consistent with the latest model version number, a prediction model updating process is started, and the edge calculation server requests the latest prediction model to be downloaded to the cloud; if the prediction model on the edge computing server matches the latest model version number, the step is repeated.
And through a behavior prediction model updating process, the prediction model loaded by each edge computing server is ensured to be the latest version.
In step S5, the edge calculation server determines whether the mobile device is moving.
Step S501, the edge computing server judges whether the mobile device moves according to the network information in the information acquisition list. Since the mobile device will switch the connected wireless base station after moving, the network information in the collection list will change. When the network information in the information acquisition list is different from the network information in the last information acquisition list, the mobile device is indicated to have moved, and the next step is carried out; otherwise, the edge computing server responds to the mobile device with a signal "NONE" and repeats the steps.
Step S502, the edge calculation server calculates a target edge calculation server on the future moving path of the mobile device through the behavior prediction model:
aiming at the problem of evaluating and observing sequence probability, a Forward algorithm is adopted to predict the occurrence probability of a sequence O of the state of a wireless base station, the probability of state occurrence is carried out by adopting the Forward algorithm after a mobile record sequence from the current connection of the mobile device is substituted into a behavior prediction model, and then a target edge calculation server on a future mobile path of the mobile device is calculated. The Forward algorithm is essentially a dynamic programming algorithm, which is discussed in the chinese patent texts with publication numbers CN109447182A, CN111625920A, and CN104834995A, and the specific use manner of the Forward algorithm in the hidden markov model is common knowledge and will not be described any further.
Step S6, the edge computing server transmits the networking information (IP and Port) of the target edge computing server to the moved mobile device, and the mobile device sends the connection information to the target edge computing server, so that a service handshake is established between the mobile device and the target edge computing server.
And the edge computing server triggers a service handshake linkage mechanism of a second layer, namely, the current edge computing server and the target edge computing server establish service handshake, inform the target edge computing server of starting the preloading service and transmit resource information required by the preloading service.
The target edge computing server judges whether the target edge computing server has the resource information required by the preloading service; if not, the target edge computing server requests the current edge computing server to acquire the resource through a channel between the current edge computing server and the target edge computing server, the resource is loaded after the resource is acquired, and the target edge computing server acquires the service providing capacity; if yes, the target edge computing server replies that the current edge computing server stops preloading the service flow.
Example (b): when the mobile device switches the connection of the wireless base station, the mobile device is indicated to have moved, and the edge computing server obtains the name, the website (IP address and MAC address) and the Port (Port) of the wireless base station before and after the switching by analyzing the uploaded information collection list. And recording the wireless base stations connected in the history by taking the numerical sequence number as an identifier.
Table 3 shows statistics of wireless base stations connected in history to a certain mobile device:
when the mobile device moves from the wireless base station with the MAC address of 00:1B:44:11:3A: B7 to the wireless base station with the MAC address of 01:00:5e:57:91: e6 and then to the wireless base station with the MAC address of 01:80: c2:00:00:01, the historical moving path of the mobile device is recorded as 1:2: 3. The observation of the state sequence in the behaviour prediction model therefore represents the identity sequence of the radio base station. By predicting the probability of occurrence of the sequence O of the state of the wireless base station, the wireless base stations that the mobile device may be connected to in the future and the edge calculation server corresponding to the wireless base stations are obtained, and the direction that the mobile device may move in the future is further known.
The mode of training the behavior prediction model adopts Baum-Welch algorithm. The model building method is that a model is built through historical moving paths of all mobile devices, the historical moving paths of all recorded mobile devices are called from a cloud database at first, the model is loaded, a Baum-Welch algorithm is adopted, and after training is finished, parameters lambda = (A, B, pi) of a behavior prediction model are derived, wherein the parameters lambda = (A, B, pi) are respectively a position state transition probability matrix of the mobile device, a state transition probability matrix of a wireless base station and an initial state probability of the wireless base station.
A Forward algorithm is then employed to predict the probability of occurrence of the sequence O of states of the wireless base station. Since the Forward algorithm calculates the probability of occurrence of sequences rather than the most likely states to transition, the historical moving path of the mobile device must be matched with different states and the state with the highest probability is found to represent the identity of the edge computing server that the mobile device is most likely to move in the future. For example, there are 3 possible observation states of the mobile device at present: 0,1,2, and the historical moving path of the mobile device is: 0,1, therefore the sequence during prediction is (0, 1,0), (0, 1,1), (0, 1,2), the occurrence probability of the three cases is obtained in this way, then the state with the highest transition probability is selected as the identifier of the wireless base station that the mobile device is likely to move next, and finally the edge calculation server responsible for the wireless base station is found.
Data Migration between the current edge computing server and the target edge computing server adopts Live Migration (Live Migration) of a KVM virtual machine, which completely saves the running state of the virtual machine and can quickly recover to the original hardware platform or even different hardware platforms. After recovery, the virtual machine is still running smoothly, and the user does not perceive any difference.
And testing the network switching effectiveness of the scheme.
The testing environment adopts 2 wireless network transmitters, 2 edge computing servers and 1 mobile device, takes the wireless base station of the garden network as the network environment, and respectively takes charge of different wireless base stations through 2 NVIDIA Jetson TK1 suites to form 2 edge computing servers. The testing method is that under the network domain of the 1 st wireless network emitter, the mobile device is connected with the 1 st edge computing server and starts playing streaming video, then the mobile device moves from the network domain of the 1 st wireless network emitter to the network domain of the 2 nd wireless network emitter, when the wireless network emitter connected with the mobile device is changed, the 1 st edge computing server requires the mobile device to transfer service and triggers the service handshake linkage mechanism of the second layer, namely, the 1 st edge computing server and the 2 nd edge computing server establish service handshake, inform the 2 nd edge computing server to start the preload service and transmit resource information required by the preload service.
The test movie used in the experimental environment is a Streaming movie of MP4 file, the playing time is always 238 seconds, the Streaming server is Apache Http server and matches with H264 Streaming module. The streaming movie has two specifications, i.e., 360P quality of 11.7MB and 1080P quality of 59.1 MB.
The experiment was divided into two cases. The first situation is that the 2 nd edge computing server starts the preloading service and receives the resource information needed by the preloading service; the second case is that the 2 nd edge computing server does not start the preload service.
Table 4 is a table of loading time statistics of streaming video during network switching:
in the first case, with the preload service, after the mobile device is switched to the network, the time for waiting for the film to be loaded is about 1-2 seconds, so that the handshake can be completed and the film can be played continuously. The size of the movie has little effect on the loading time.
In the second case, without the preload service, after the mobile device is switched to the network, the time for waiting for the film to be loaded is about 23-85 seconds, and then the handshake can be completed and the film can be played continuously. The loading time increases significantly as the specification of the movie is larger.
By comparing the first situation with the second situation, the scheme can effectively reduce the response time of the service during network switching.
And testing the accuracy of the behavior prediction model.
The accuracy of the behavior prediction model was tested as follows: 80% of the data was used for training the model and 20% was used as test data. And substituting the sequence in the test data into the updated behavior prediction model to predict the wireless base station on the future path of the mobile device. Then, the radio base stations corresponding to the next states of the actual sequences are compared. If the two are the same, the model prediction is successful; if the two are different, the model prediction is wrong. Repeating the prediction process until all the test data are completely predicted, and finally counting the successful prediction proportion of the model as the accuracy of the prediction model.
Fig. 5 is an accuracy test chart of the behavior prediction model, with the horizontal axis representing the state, i.e., the number of radio base stations, and the vertical axis representing the accuracy of the prediction model. Therefore, the accuracy of the prediction model is 80% -90%, and the prediction accuracy is high.
It should be understood that equivalents and modifications of the technical solution and inventive concept thereof may occur to those skilled in the art, and all such modifications and alterations should fall within the scope of the appended claims.
Claims (7)
1. An edge network switching method based on dynamic mobile device behavior prediction is characterized by comprising the following steps:
step S1, establishing an edge node architecture: the edge node architecture comprises a cloud end, an edge computing server and a mobile device; the edge computing server is in communication connection with the cloud end; the mobile device and the edge computing server are in communication connection to form a local area network; the edge computing servers are connected with wireless base stations, and the wireless base stations responsible for any two edge computing servers are not crossed;
step S2, establishing a first layer of service handshake linkage mechanism, and tracking the state of the mobile equipment in the local network to which the edge computing server belongs by the edge computing server; the mobile device collects information including hardware, network and position, establishes an information collection list and uploads the information collection list to the edge computing server periodically; the edge computing server uploads the information of all the mobile devices to the cloud terminal for gathering;
then, simultaneously executing the step 3 and the step 4;
step S3, if the number of mobile devices served by the edge computing server is overloaded, adjusting the deployment of the mobile devices;
step S4, establishing and updating a behavior prediction model: adopting a hidden Markov model and redefining parameters to obtain a behavior prediction model;
in the behavior prediction model, a hidden state sequence I is defined as a sequence of the actual position state of the mobile device; an observation state sequence O defined as a sequence of states of the radio base stations, the sequence being a set of all radio base stations observing states; a state transition matrix A, defined as a position state transition probability matrix of the mobile device, representing a state transition matrix of hidden state transition probability; a probability matrix B generated by observing the state is defined as a state transition probability matrix of the wireless base station and represents a set of transition probabilities between a hidden state of the mobile device and an observation state of the wireless base station; hidden state probability distribution pi defined as the initial state probability of the wireless base station, and setting the parameter as 1;
the model building method comprises the steps of building a model through historical moving paths of all mobile devices, firstly calling all recorded historical moving paths of the mobile devices from a cloud database, loading the model, adopting a Baum-Welch algorithm, and deriving parameters lambda = (A, B, pi) of a behavior prediction model after training, wherein the parameters lambda = (A, B, pi) are respectively a position state transition probability matrix of the mobile device, a state transition probability matrix of a wireless base station and an initial state probability of the wireless base station;
step S5, the edge calculation server judges whether the mobile device moves; if the mobile device moves, calculating a target edge calculation server on a future moving path of the mobile device through a behavior prediction model, otherwise, repeating the step;
predicting the occurrence probability of a sequence O of the state of the wireless base station by adopting a Forward algorithm, substituting a movement record sequence from the current connection of the mobile device into a behavior prediction model, predicting the occurrence probability of the state by adopting the Forward algorithm, and then calculating a target edge calculation server on a future movement path of the mobile device;
step S6, the edge computing server transmits the networking information of the target edge computing server to the mobile device after moving, and the mobile device sends the connection information to the target edge computing server, so that service handshake is established between the mobile device and the target edge computing server;
and the edge computing server triggers a service handshake linkage mechanism of a second layer, namely, the current edge computing server and the target edge computing server establish service handshake, inform the target edge computing server of starting the preloading service and transmit resource information required by the preloading service.
2. The edge network switching method based on dynamic mobile device behavior prediction as claimed in claim 1, wherein the cloud is configured with a database storing all edge computing server information and all mobile device information in the framework and is responsible for completing training of a behavior prediction model of the mobile device;
the cloud is provided with an architecture deployment unit, a mobile device behavior prediction model training unit and a database;
the architecture deployment unit adjusts the deployment of the whole architecture, calls the mobile device information in the database, and tracks the average response time between each mobile device and the edge calculation server connected with the mobile device correspondingly;
the mobile device behavior prediction model training unit is used for constructing and training a behavior prediction model of the mobile device, and updating the behavior prediction model of the mobile device after training is finished;
the database adopts a relational database management system to record all edge computing server information and all mobile device information in the framework, including deployment and networking information of the computing server information, state and networking information of the mobile device and historical moving paths of the mobile device.
3. The edge network forwarding method based on dynamic mobile device behavior prediction as claimed in claim 2, wherein the edge computing server provides resource services for nearby mobile devices and grasps the movement behaviors of nearby mobile devices, and meanwhile, the edge computing server is further connected to the edge computing server in communication with the nearby edge computing server;
the edge computing server is provided with an architecture deployment auxiliary unit, a service providing unit, a state tracking unit, a mobile device behavior prediction unit, a preloading communication unit, a network switching handshake unit and a lightweight database.
4. The edge forwarding method based on dynamic mobile device behavior prediction as claimed in claim 3, wherein the mobile device is a user equipment generating request data or sensing data; the mobile device is provided with a service receiving unit, a network switching holding unit and an information acquisition unit.
5. The edge forwarding method based on dynamic mobile device behavior prediction as claimed in claim 1, wherein in step 2, the information collection list comprises: a device name field, a status field, an edge calculation server field, a network information field, a location field, a sending time field, and a historical movement path field;
a device name field for recording the name of the mobile device;
the state column records the current running state of the mobile device and is divided into a normal state, a transition state and a closing state;
an edge computing server column that records an edge computing server currently deployed by the mobile device; after the mobile device is switched to the network, the current column records a new edge computing server;
a network information column for recording the name, IP address, MAC address and port of the wireless base station connected with the mobile device;
a location field recording the longitude and latitude of the mobile device;
sending a time column, recording the uploading time of the information acquisition list, wherein the time is accurate to millisecond;
and the historical moving path column records the wireless base stations which are historically connected with the mobile device by taking the numerical sequence number as an identifier.
6. The edge forwarding method based on dynamic mobile device behavior prediction as claimed in claim 1, wherein step 3 further comprises the following steps:
step S301, an edge computing server obtains deployment information of adjacent edge computing servers from a cloud; the edge computing server judges that the number of the mobile devices served by the edge computing server currently exceeds the load; if yes, executing the next step; if not, repeating the step;
step S302, calculating the distance between the current edge calculation server and the adjacent edge calculation server, and then selecting the nearest edge calculation server which does not exceed the load as a target edge calculation server;
step S303, calculating a distance between the target edge calculation server and each mobile device in the local area network according to the position of the target edge calculation server and the position of each mobile device in the local area network, selecting the mobile device closest to the target edge calculation server as the mobile device for network switching, and then executing step 6 to establish a service connection between the mobile device for network switching and the target edge calculation server.
7. The edge forwarding method based on dynamic mobile device behavior prediction as claimed in claim 1, wherein in step 6, the target edge computing server determines whether it has resource information required by the preloaded service; if not, the target edge computing server requests the current edge computing server to acquire the resource through a channel between the current edge computing server and the target edge computing server, the resource is loaded after the resource is acquired, and the target edge computing server acquires the service providing capacity; if yes, the target edge computing server replies that the current edge computing server stops preloading the service flow.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110834476.5A CN113301169B (en) | 2021-07-23 | 2021-07-23 | Edge network switching method based on dynamic mobile device behavior prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110834476.5A CN113301169B (en) | 2021-07-23 | 2021-07-23 | Edge network switching method based on dynamic mobile device behavior prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113301169A CN113301169A (en) | 2021-08-24 |
| CN113301169B true CN113301169B (en) | 2021-10-01 |
Family
ID=77330838
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110834476.5A Active CN113301169B (en) | 2021-07-23 | 2021-07-23 | Edge network switching method based on dynamic mobile device behavior prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113301169B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114090221B (en) * | 2022-01-20 | 2022-04-29 | 浙江万雾信息科技有限公司 | Dual migration method for tasks in edge computing environment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112218337A (en) * | 2020-09-04 | 2021-01-12 | 暨南大学 | Cache strategy decision method in mobile edge calculation |
| CN112463337A (en) * | 2020-12-08 | 2021-03-09 | 内蒙古大学 | Workflow task migration method used in mobile edge computing environment |
| CN112788109A (en) * | 2020-12-29 | 2021-05-11 | 哈尔滨工业大学 | Service migration method and system based on mobile prediction and multi-layer service deployment |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108063802B (en) * | 2017-12-01 | 2020-07-28 | 南京邮电大学 | User position dynamic modeling optimization method based on edge calculation |
| US20210135969A1 (en) * | 2019-11-06 | 2021-05-06 | Centurylink Intellectual Property Llc | Predictive resource allocation in an edge computing network utilizing beacons |
| CN111491332B (en) * | 2020-04-20 | 2021-08-27 | 中国石油大学(北京) | Dynamic service migration method and device |
-
2021
- 2021-07-23 CN CN202110834476.5A patent/CN113301169B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112218337A (en) * | 2020-09-04 | 2021-01-12 | 暨南大学 | Cache strategy decision method in mobile edge calculation |
| CN112463337A (en) * | 2020-12-08 | 2021-03-09 | 内蒙古大学 | Workflow task migration method used in mobile edge computing environment |
| CN112788109A (en) * | 2020-12-29 | 2021-05-11 | 哈尔滨工业大学 | Service migration method and system based on mobile prediction and multi-layer service deployment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113301169A (en) | 2021-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110809306B (en) | Terminal access selection method based on deep reinforcement learning | |
| US10966122B2 (en) | Method and migration managing module for managing a migration of a service | |
| CN104285470B (en) | Make it possible to improve method, computer program product and the device of the switching in mobile communications network | |
| WO2018120802A1 (en) | Collaborative content cache control system and method | |
| CN108900980B (en) | Resource allocation optimization method based on mobility prediction in heterogeneous network | |
| US10542465B2 (en) | Analytics driven wireless device session context handover in operator cloud | |
| CN113301169B (en) | Edge network switching method based on dynamic mobile device behavior prediction | |
| KR20120059175A (en) | System and method for providing mobile p2p service | |
| CN113163409B (en) | Mobile edge computing service placement method based on artificial intelligence | |
| Alghamdi et al. | Time-optimized task offloading decision making in mobile edge computing | |
| Mehrabi et al. | A survey on mobility management for MEC-enabled systems | |
| Li et al. | Smartphone-assisted energy efficient data communication for wearable devices | |
| JP2022077504A (en) | Computer-implemented method, computer system, and computer program, for predictively deploying microservices on edge devices in network | |
| US20220060956A1 (en) | Systems and methods for single uplink operation (suo) over dual connectivity networks | |
| KR101924628B1 (en) | Apparatus and Method for controlling traffic offloading | |
| CN113316087B (en) | Dynamic paging method based on terminal position prediction in LTE system | |
| WO2022001555A1 (en) | Wireless resource management method, storage medium, and electronic device | |
| Ju et al. | eDeepSave: Saving DNN inference using early exit during handovers in mobile edge environment | |
| Dai et al. | MAPCaching: A novel mobility aware proactive caching over C-RAN | |
| CN111050398B (en) | Paging method, paging device, paging equipment and storage medium | |
| Cao et al. | Mobility-Aware Routing and Caching in Small Cell Networks using Federated Learning | |
| US11659238B2 (en) | Methods, systems, and devices for providing service differentiation for different types of frames for video content | |
| Cao et al. | Mobility-aware routing and caching: A federated learning assisted approach | |
| CN110213778A (en) | A kind of active and standby method and device intelligently matched of network element | |
| WO2017097368A1 (en) | System and method for efficient caching in the access network |
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 |