A Distributed Edge Computing Information Scheduling Method Based on Information Priority

Abstract

The present invention belongs to the technical field of edge computing, and particularly relates to a distributed edge computing information scheduling method based on information priority, comprising the following steps: S1. Construct a distributed communication scenario; S2. Construct a dynamic network topology structure and an information injection model, and at the same time, the terminal node assigns different priorities to the injected messages according to the urgency of the messages; S3. The edge node broadcasts the benchmark probability of this round, and the terminal node sets the transmission probability of this round according to the benchmark probability; S4. The terminal device randomly selects a message to be sent from the message queue according to the proportion of the message priority, and transmits it at the transmission probability obtained in S3 in the transmission stage. At the same time, the edge node listens to the channel and dynamically adjusts the benchmark probability of the next round according to the channel state at this time; S5. The edge node feeds back the information successfully received in the transmission stage. Its advantage lies in solving the problem of wireless communication between multiple terminal nodes and one edge device.

Description

Technical Field

[0001] The present invention belongs to the technical field of edge computing, and particularly relates to a distributed edge computing information scheduling method based on information priority. Background Art

[0002] In order to solve the problems of high latency, instability, low bandwidth, etc. brought by traditional cloud computing, edge computing, as an emerging technology, has developed rapidly in recent years. Edge computing migrates computing resources from the center to the edge, enabling edge nodes to have the ability to process certain simple requests, thereby effectively reducing the computing burden on the cloud and improving the response speed. In an actual scenario, an edge node may need to be responsible for processing requests from multiple terminal nodes. How to solve the competition and interference problems in the wireless channel while ensuring that requests can reach the edge node from the terminal node as quickly as possible is a key challenge faced by edge computing.

[0003] However, most scheduling algorithms adopt a centralized method, that is, all requests are scheduled by a central node (usually a base station). This method can ensure the efficiency of scheduling, but in this mode, all messages need to be aggregated to the central node, the information flow is mixed, the algorithm design is relatively complex, and at the same time, there is a relatively high demand for energy consumption. As the network scale increases, it is difficult for centralized scheduling to ensure efficient communication.

[0004] Compared with the centralized method, the distributed method adopts a decentralized idea, and all nodes run the same algorithm and cooperate with each other to complete a task. This method does not need to know the global information of the network, so it is more suitable for large-scale networks and energy-constrained scenarios. However, how to design a distributed algorithm to achieve an efficiency as close as possible to that in a centralized environment is a challenge in the design of distributed algorithms.

[0005] On the other hand, whether it is a centralized or distributed scheduling algorithm, most of them do not consider the priority of request messages and treat all messages equally. However, in an actual edge computing scenario, there are often some more urgent messages that need to be responded to preferentially, such as signals detecting natural disasters. Therefore, designing a priority-based and efficient distributed scheduling algorithm is crucial for the development of edge computing. Summary of the Invention

[0006] To solve the above problems and improve the utilization rate of wireless channels, the present invention proposes a distributed edge computing information scheduling algorithm based on information priority. This algorithm assigns a priority to each message and uses a random method for transmission, which can ensure that the transmission probability of each round of messages is proportional to its priority. At the same time, the binary exponential backoff method is also used to dynamically adjust the transmission probability to solve the competition and interference problems in the wireless channel. While taking fairness into account, it can ensure that the channel utilization rate is a constant optimum.

[0007] A distributed edge computing information scheduling method based on information priority, comprising the following steps:

[0008] S1. Construct a distributed communication scenario;

[0009] S2. Construct a dynamic network topology structure and an information injection model. At the same time, according to the urgency of the message, the terminal node assigns different priorities to the injected messages;

[0010] S3. The edge node broadcasts the benchmark probability of this round, and the terminal node sets the transmission probability of this round according to the benchmark probability;

[0011] S4. The terminal device randomly selects a message to be sent from the message queue according to the proportion of the message priority and transmits it with the transmission probability of this round obtained in S3 during the transmission stage. At the same time, the edge node listens to the channel and dynamically adjusts the benchmark probability of the next round according to the channel state at this time;

[0012] S5. The edge node feeds back the information successfully received during the transmission stage.

[0013] Preferably, in step S2, the process of constructing the dynamic network topology structure model is specifically as follows:

[0014] Construct 1 edge node at a fixed position, which is equipped with a server and a half-duplex signal transmitter and is responsible for processing various edge computing requests;

[0015] Construct n terminal nodes. Each terminal node is responsible for monitoring the environment it is in charge of and may generate various information requesting edge computing services; the terminal nodes are either at fixed positions or move arbitrarily within the communication range of the edge node; both the edge node and the terminal nodes use the half-duplex method for communication, and the node can only choose to send information or listen to the channel at each moment and cannot transmit while listening.

[0016] Preferably, in step S2, the process of constructing the information injection model is specifically as follows: In each time round t, each terminal node generates new messages according to a specific probability to simulate the dynamics of information generation in the actual edge computing scenario. For a specific edge node i, it will be with π iwith a probability of generating a new message; or with a probability of 1 - π i not to generate a new message.

[0017] Preferably, the edge node assigns different priorities to the newly injected messages according to the urgency of the messages:

[0018] For a specific scenario, all possible edge computing request services are analyzed and classified according to their urgency; for each newly generated message in each round, the decision maker of the terminal node assigns an integer from 1 to K as the priority according to the type, and the larger the number, the more urgent the message is and the faster it should be processed by the edge device.

[0019] Preferably, a random algorithm is used for communication scheduling between the terminal node and the edge node. The terminal node sets the transmission probability for this round according to the reference probability obtained in each round, and at the same time, the edge node dynamically adjusts the reference probability according to the channel state in this round to improve the channel utilization rate. Specifically:

[0020] S41. The edge node broadcasts the reference probability p e , and the terminal node sets its own transmission probability for this round according to this probability: each edge node i maintains a message queue Q i , which records all the messages that have not been successfully transmitted by the edge node i; the edge node i sets the transmission probability according to the obtained reference probability where K i represents the sum of the priorities of all the messages in Q i ; the edge node i selects a message to be transmitted according to the proportion of the priority of each message in Q i ;

[0021] S42. The edge node dynamically adjusts the reference probability p according to the channel state sensed in the transmission stage according to the binary exponential backoff method e ; the edge node can observe the result of each round of transmission. If at least one terminal node selects to transmit in this round, then set Conversely, if no terminal node selects to transmit in this round, then set p e = p e *2;

[0022] S43. If the edge node successfully receives a message from the terminal node in the transmission stage, it generates a corresponding ACK signal to feedback the received message. If the terminal node receives the feedback signal for the transmitted message, it can know that the transmission is successful and remove the successfully transmitted message from the message queue.

[0023] Preferably, the SINR model is used to characterize the interference of the wireless channel. Specifically, for the transmitting node u and the receiving node v,

[0024] Signal(u, v) = P u ·d(b, v) -α

[0025]

[0026] where Signal(u, v) is the signal strength that node v senses from node u, P u represents the transmission power of node u, d(u, v) is the Euclidean distance between u and v, and α ∈ (2, 6] represents the path loss exponent, which is determined by the transmission medium and other factors in the environment; for the signal from u to v, its strength Signal(u, v) weakens with distance. When v attempts to decode the signal from u, ∑ w∈S \{u}Signal(w, v) + N is interference, where S is the set of nodes transmitting simultaneously with u, and N is the environmental noise. When SINR(u, v) ≥ β, the transmission from u to v is considered successful, where β ≥ 1 is a threshold determined by the hardware.

[0027] Preferably, both the terminal device and the edge device are equipped with a half-duplex wireless signal transmitter, that is, only one of listening to the channel or transmitting can be selected in each round; if the node chooses to listen to the channel, it can determine three states of the channel according to the sensed signal strength and the received information, namely:

[0028] State 1: Channel idle: None of the nodes in this round choose to transmit.

[0029] State 2: Successful transmission: Only one node in this round chooses to transmit, and all nodes listening to the channel will receive the transmitted information.

[0030] State 3: Channel busy: Two or more nodes transmit simultaneously, resulting in a collision. At this time, the nodes listening to the channel will not receive the transmitted information.

[0031] In a single-hop network, the channel states observed by all nodes are the same. If a node chooses to send information, it will transmit the message to be sent with a fixed power P.

[0032] Preferably, in step S5,

[0033] feedback of the message is performed. If the edge node successfully receives the message from the terminal node, it broadcasts an ACK signal for feedback; during this stage, edge node i chooses to listen to the channel. If it successfully receives the feedback on the sent message, indicating that the message transmission is successful, it can remove the message from the message queue and then update the sum of the message priorities K i ; if a new message is generated, it is added to the message queue and K is updatedi 。

[0034] Preferably, for a message with priority k, its weighted age of information A(t) in the t-th round is:

[0035]

[0036] Beneficial effects

[0037] The present invention adopts the binary exponential backoff method to implement a scheduling method for messages with different priorities in an edge computing scenario. This method uses a random algorithm for message transmission, dynamically adjusts the transmission probability according to the channel state, solves the competition and interference problems in wireless communication, considers the real scenario of edge computing applications, ensures the fairness of channel utilization among messages with different priorities, and at the same time ensures a constant optimal communication efficiency. Description of the drawings

[0038] Figure 1 is a schematic diagram of the edge computing scenario in the embodiment of the present invention;

[0039] Figure 2 is the pseudocode description of the distributed scheduling algorithm based on information priority in the embodiment of the present invention;

[0040] Figure 3 is the experimental result display of the distributed scheduling algorithm based on information priority in the embodiment of the present invention.

[0041] Figure 4 is the experimental result display chart for the maximum weighted age of information. Detailed implementation manners

[0042] For the convenience of understanding the present invention, the present invention will be described in more detail below with reference to the drawings and specific embodiments.

[0043] Appendix Figure 1It is a schematic diagram of the edge computing scenario described in the present invention. Four common scenarios for requesting edge computing services are given in the figure, namely: intelligent transportation, mobile devices, security systems, and intelligent workshops. These terminal nodes will generate various requirements according to the situations they face, and a corresponding determination program will assign specific priorities. For example, in Scenario 1, a car accident occurs, and the camera responsible for monitoring generates a message with a priority of 100. The wireless network is used for communication between the terminal nodes and the edge nodes, and the distributed edge computing information scheduling algorithm proposed by us runs in the wireless network. It should be noted that during the scheduling process, the algorithm of the present application can ensure that the priority of the information is only known to the edge node that owns the information, and is not known to other edge nodes and terminal nodes, thereby realizing a kind of privacy protection. And the algorithm of the present application can ensure that the node with a higher priority is more likely to reach the edge node first, thereby ensuring fairness between information with different priorities. After the message reaches the edge node, the information is processed according to different edge computing scenarios. For example Figure 1 Application 1 responsible for processing intelligent transportation in it may perform various operations according to the information content after receiving the information, such as notifying the nearby traffic police to go to the accident site for handling.

[0044] Specifically, the steps for constructing the edge computing scenario described in the present invention include:

[0045] S1. Construct a distributed communication scenario;

[0046] S2. Construct a dynamic network topology structure and an information injection model, and at the same time, the terminal node assigns different priorities to the injected messages according to the urgency of the messages;

[0047] (1). Construct a dynamic network topology model: Construct 1 edge node at a fixed position, which is equipped with a high-performance computing server and a half-duplex signal transmitter, and can be responsible for processing various edge computing requests. Construct n terminal nodes, and each terminal node is responsible for monitoring the environment it is in charge of and may generate various information requesting edge computing services. The terminal nodes can be at fixed positions or can move arbitrarily within the communication range of the edge node. This means that the network topology can change in real time, which is more in line with the actual situations of vehicle networking, drone swarms, etc. Both the edge node and the terminal node communicate in a half-duplex manner, which means that the node can only choose to send information or listen to the channel at each moment, and cannot send while listening.

[0048] For a specific scenario, analyze all possible edge computing request services and classify them according to the urgency. Each terminal node has a judgment program, which assigns an integer from 1 to K as the priority to each request, and the larger the number, the more urgent the message. At the same time, each terminal node i maintains a message queue Q i, which is used to store all unresponded requests. If the message queue is empty, a message with a priority of 1 is generated and added to the message queue. This message is used to notify the edge node that the situation is normal. When an emergency occurs, this message is removed.

[0049] (2) Physical interference model: Considering that in a wireless channel, the strength of a signal weakens as the distance increases, and the strengths of multiple signals will produce signal superposition at the receiving end. Therefore, the present invention uses the SINR model to characterize the interference situation of the wireless channel. Compared with the traditional graph interference model, it better describes the fading characteristics and superposition characteristics of the wireless channel. Specifically, for the transmitting node u and the receiving node v, we have:

[0050] Signal(u,v) = P u ·d(b,v) -α

[0051]

[0052] In the above equation, Signal(u,v) is the signal strength that node v senses from node u. P u represents the transmission power of node u, d(u,v) is the Euclidean distance between u and v, and α ∈ (2, 6] represents the path loss exponent, which is determined by the transmission medium and other factors in the environment. For the signal from u to v, its strength Signal(u,v) weakens with distance. When v attempts to decode the signal from u, ∑ w∈S\{u} Signal(w,v) + N is the interference, where S is the set of nodes transmitting simultaneously with u, and N is the environmental noise determined by the environment. When SINR(u,v) ≥ β, the transmission from u to v is considered successful, where β ≥ 1 is a threshold determined by the hardware.

[0053] Both the terminal device and the edge device have a half-duplex wireless signal transmitter, which means that the node can only choose to listen to the channel or perform transmission in one round. If the node chooses to listen to the channel, it can determine the three states of the channel according to the sensed signal strength and the received information, namely:

[0054] State 1: Channel idle: No node chooses to transmit in this round.

[0055] State 2: Successful transmission: Only one node chooses to transmit in this round, and all nodes listening to the channel will receive the transmitted information.

[0056] State 3: Channel busy: Two or more nodes transmit simultaneously, resulting in a conflict. At this time, the nodes listening to the channel will not receive the transmitted information.

[0057] In a single-hop network, the channel states observed by all nodes are consistent. If a node chooses to send information, it transmits the message to be sent with a fixed power P.

[0058] (3) Information injection model: In each time slot t, each terminal node generates new messages according to a specific probability to simulate the dynamics of information generation in the actual edge computing scenario. For example, for a specific edge node i, it will generate a new message with a probability of π i ; or with a probability of 1 - π i it does not generate a new message.

[0059] (4) Message feedback mechanism: After the communication process of each round ends, if the edge node successfully receives the message from the terminal node, it broadcasts an ACK signal to feedback the message. The terminal node that receives the corresponding message feedback can determine that the transmitted message has been successfully received and removes the message from the message queue.

[0060] S3. The edge node broadcasts the benchmark probability of this round, and the terminal node sets the transmission probability of this round according to the benchmark probability;

[0061] S4. The terminal device randomly selects a message to be sent from the message queue according to the proportion of the message priority, and transmits it with the transmission probability of this round obtained in S3 during the transmission stage. At the same time, the edge node listens to the channel and dynamically adjusts the benchmark probability of the next round according to the current channel state;

[0062] Appendix Figure 2 is the pseudocode description of the distributed scheduling algorithm based on information priority. As shown in Appendix Figure 2 The specific steps of the communication process of the distributed scheduling algorithm based on information priority include three stages:

[0063] Stage 1: Broadcast the benchmark probability. At this time, the edge node broadcasts the benchmark probability p of this round of transmission with power P e , and all terminal nodes choose to listen to the channel. Because there is no interference at this time, it can be ensured that all terminal nodes within the communication radius of the edge node can receive the benchmark probability p e .

[0064] Stage 2: Transmit the message. At this time, the edge node chooses to listen to the channel. Edge node i first randomly selects a message to be sent according to the priority ratio of the information in the message queue. For example, if there are three messages with priorities 1, 2, and 3 in the message queue, they are selected with probabilities respectively. According to the received benchmark probability p e , edge node i sets the transmission probability of this round In the transmission stage, the node transmits with a probability of p iSend the message selected in this round with probability 1 - p i Keep silent.

[0065] S5. Edge nodes give feedback on the information successfully received during the transmission phase.

[0066] Give feedback on the message. If an edge node successfully receives a message from a terminal node in Phase 2, it broadcasts an ACK signal for feedback. During this phase, terminal node i chooses to listen to the channel. If it successfully receives feedback on the sent message, indicating that the message transmission is successful, it can remove the message from the message queue and then update the sum of message priorities K i If a new message is generated, add it to the message queue and update K i .

[0067] Figure 3 It is a display of the experimental results of the distributed scheduling algorithm based on information priority in the embodiments of the present invention.

[0068] To better demonstrate the effect of the present invention, the experiment uses the concept of weighted age of information (WAOI) for description, which well describes the time required for information to be generated and successfully sent. It is defined as follows: for a message with priority k, its weighted age of information A(t) in the t-th round is:

[0069]

[0070] Figure 3 and Figure 4 shows the variation of the average weighted age of information and the maximum weighted age of information with the number of rounds and the arrival rate ζ of new information (i.e., the sum of arrival rates of all nodes) when the number of nodes n = 2000, 3000, 4000, 5000. Among them Figure 3 mainly considers the average waiting time required for all information to be generated and received by edge nodes, Figure 4 mainly considers the longest waiting time required for all information to be generated and received by edge nodes. It can be seen that the algorithm can remain stable when ζ does not exceed 0.09. And it can be proved that for a message with priority k, its expected number of waiting rounds is O(Kn / k), where K is the maximum value of the priority. Obviously, when the priority k is larger, the expected number of waiting rounds is smaller, reflecting the fairness of the algorithm of the present application. At the same time, the algorithm of the present application can ensure that within a sufficiently long period, the throughput of the channel is a constant, which means that the efficiency constant of the algorithm of the present application is approximately optimal.

[0071] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A distributed edge computing information scheduling method based on information priority, characterized in that, It includes the following steps: S1. Construct a distributed communication scenario; S2. Construct a dynamic network topology structure and an information injection model. Meanwhile, according to the urgency of the message, the terminal node assigns different priorities to the injected message; S3. The edge node broadcasts the benchmark probability of this round, and the terminal node sets the transmission probability of this round according to the benchmark probability; S4. The terminal device randomly selects a message to be sent from the message queue according to the proportion of the message priority, and transmits it with the transmission probability of this round obtained in S3 during the transmission phase. Meanwhile, the edge node listens to the channel and dynamically adjusts the benchmark probability of the next round according to the current channel state; S41. The edge node broadcasts the baseline probability p e , and the terminal node sets its own transmission probability for this round according to this probability: each edge node i maintains a message queue Q i , which records all the messages of edge node i that have not been successfully transmitted; edge node i sets the transmission probability according to the obtained baseline probability where K i represents the sum of the priorities of all the messages in Q i ; Edge node i selects a message to be transmitted according to the proportion of the priority of each message in Q i ; S42. The edge node dynamically adjusts the reference probability p according to the channel state sensed in the transmission stage using the binary exponential backoff method e ; The edge node can observe the result of each round of transmission. If at least one terminal node selects transmission in this round, then set Conversely, if no terminal node selects transmission in this round, then set p e = p e * 2; S43. If, during the transmission phase, the edge node successfully receives a message from the terminal node, it generates a corresponding ACK signal to feedback the received message. If the terminal node receives a feedback signal for the transmitted message, it can know that the transmission is successful and removes the successfully transmitted message from the message queue; S5. The edge node feeds back the information successfully received during the transmission phase.

2. The distributed edge computing information scheduling method based on information priority according to claim 1, wherein In step S2, when constructing the dynamic network topology structure model, the specific process is as follows: Construct 1 edge node at a fixed position, which is equipped with a server and a half-duplex signal transmitter, and is responsible for processing various edge computing requests; Construct n terminal nodes, and each terminal node is responsible for monitoring its respective responsible environment and may generate various information requesting edge computing services; The terminal nodes are either at fixed positions or move arbitrarily within the communication range of the edge node; both the edge node and the terminal nodes communicate in a half-duplex manner. A node can only choose one of sending information or listening to the channel at each moment and cannot transmit while listening.

3. A distributed edge computing information scheduling method based on information priority according to claim 1, characterized in that, In step S2, an information injection model is constructed, and the specific process is as follows: In each time round t, each terminal node generates a new message according to a specific probability to simulate the dynamics of information generation in an actual edge computing scenario. For a specific edge node i, a new message will be generated with a probability of π i ; or no new message will be generated with a probability of 1 - π i .

4. A distributed edge computing information scheduling method based on information priority according to claim 1, characterized in that, The edge node will assign different priorities to the newly injected message according to the urgency of the message: For a specific scenario, all possible edge computing request services are analyzed and classified according to their urgency; for each newly generated message in each round, the decision maker of the terminal node will assign an integer from 1 to K as the priority according to the type. The larger the number, the more urgent the message is and the faster it should be processed by the edge device.

5. A distributed edge computing information scheduling method based on information priority according to claim 1, characterized in that, The SINR model is used to characterize the interference of the wireless channel. Specifically, for the transmitting node u and the receiving node v, Signal(u,v) = P u ·d(b,v) -α where Signal(u, v) is the signal strength that node v senses from node u, and P u represents the transmission power of node u, d(u, v) is the Euclidean distance between u and v, and α ∈ (2, 6] represents the path loss exponent, which is determined by the transmission medium and other factors in the environment; for the signal from u to v, its strength Signal(u, v) weakens with distance. When v attempts to decode the signal from u, ∑ w∈S\{u} Signal(w, v) + N is the interference, where S is the set of nodes transmitting simultaneously with u, and N is the ambient noise. When SINR(u, v) ≥ β, the transmission from u to v is considered successful, where β ≥ 1 is the threshold determined by the hardware.

6. A distributed edge computing information scheduling method based on information priority according to claim 2, characterized in that, Both the terminal device and the edge device have a half-duplex wireless signal transmitter, that is, they can only choose one of listening to the channel or transmitting in each round; if they choose to listen to the channel, the node can determine the three states of the channel according to the sensed signal strength and the received information, namely: State 1. The channel is idle: No node chooses to transmit in this round; State 2. Successful transmission: Only one node chooses to transmit in this round, and all nodes listening to the channel will receive the transmitted information; State 3. The channel is busy: Two or more nodes transmit simultaneously, resulting in a conflict. At this time, the nodes listening to the channel will not receive the transmitted information; In a single-hop network, the channel states observed by all nodes are the same. If a node chooses to send information, it transmits the message to be sent with a fixed power P.

7. A distributed edge computing information scheduling method based on information priority according to claim 1, characterized in that In step S5, Perform feedback on the message. If the edge node successfully receives the message from the terminal node, it broadcasts an ACK signal for feedback. During this phase, edge node i chooses to listen to the channel. If it successfully receives the feedback on the sent message, indicating that the message transmission is successful, it can remove the message from the message queue and then update the sum of message priorities K i ; If a new message is generated, add it to the message queue and update K i .

8. A distributed edge computing information scheduling method based on information priority according to claim 1, characterized in that For a message with priority k, its weighted age of information A(t) at the t-th round is as follows:

Images