Online single-period task offloading method for 5g small base station network

Abstract

The application discloses an online single-period task offloading method for a 5G small base station network and belongs to the field of task offloading. The method provides a single-period task model for describing a wireless device connecting a 5G small base station network scene and gives a task offloading method based on the model. The method can adjust and update a task offloading strategy according to real-time data feedback without depending on any prior knowledge of the network, solves the single-period task offloading problem of the wireless device connecting the 5G small base station network scene, fills the vacancy of the task offloading method under the wireless device connecting the 5G small base station network scene, and can obtain a high task reward while guaranteeing quality of service (QoS) and has a high performance ratio.

Description

Technical Field

[0001] This invention relates to the field of task offloading, and more particularly to an online single-cycle task offloading method for 5G small cell networks. Background Technology

[0002] Fifth-generation mobile communication technology (5G) is a new generation of broadband mobile communication technology characterized by high speed, low latency and massive connectivity. 5G communication facilities are the network infrastructure for realizing the interconnection of people, machines and things. Among them, macro base stations, as the core nodes of the 5G network, have a large coverage area and a large data processing capacity, while small base stations, as the edge nodes of the 5G network, have a lower coverage area and computing capacity, and complement macro base stations.

[0003] As edge nodes in 5G networks, small base stations are typically located close to wireless devices and possess a significantly faster capacity to process large amounts of data compared to wireless devices. Applications such as security monitoring, virtual reality, and autonomous driving often generate massive amounts of data and are highly sensitive to data processing latency. Compared to macro base stations and cloud servers, which are core network nodes, small base stations are better positioned to meet the low-latency, high-computing demands of these scenarios.

[0004] The problem of how to transfer computing tasks generated by devices to network edge nodes for processing is known as the "task offloading" problem. Many task offloading methods have been proposed to date, but due to the characteristics of 5G networks, such as the unique features of small cell nodes, dynamic network changes, overlapping coverage by multiple base stations, and unstable signals, these methods are not entirely applicable. A task offloading method specifically for scenarios where wireless devices connect to 5G small cell networks has not yet been proposed. Summary of the Invention

[0005] The main objective of this invention is to provide an online single-cycle task offloading method for 5G small cell networks. The method provides a model that describes the scenario of a wireless device connecting to a 5G small cell network and requesting a single-cycle computing task. It incorporates characteristics such as dynamic changes in the 5G network, limited resources, multiple coverage areas, and uncertain signal strength, thus reflecting the complex and ever-changing reality of the scenario. This invention can adjust and update the task offloading strategy based on real-time data feedback without relying on any prior network knowledge, thereby solving the problem of single-cycle task offloading in scenarios where wireless devices connect to 5G small cell networks.

[0006] This invention provides a single-cycle task model for a scenario where a wireless device connects to a 5G small cell network:

[0007] There are m small base station nodes in a 5G small base station network. A series of wireless devices are distributed in the network and connected to small base station nodes. Wireless devices can switch to another small base station at any time, and a wireless device may be covered by the signals of multiple small base stations. Represents t time slots, Indicates the first The set of all tasks that need to be processed within a time slot, and each task can be processed within a single time slot (i.e., a single-cycle task). Indicates the first Within the time slot and in the time slot A set of tasks within the coverage area of ​​a small base station. For the first The maximum number of wireless devices that a small base station can cover, where c is the number of small base stations. The maximum number of wireless devices that a small base station can connect to simultaneously, β being the number of small base stations. The maximum computing resources that a small base station can access simultaneously, where α is the number of base stations. The minimum computing resources required for a small base station to guarantee Quality of Service (QoS) at the same time.

[0008] A computational task is characterized by three features: input data size, output data size, and time delay requirement. These three features are called the context of task k. Indicates the first The small base station in the first Each time slot completes the context as φ k The rewards for the task, Indicates the first The small base station in the first Each time slot completes the context as φ k The probability of the task. Indicates the first The small base station in the first Each time slot completes the context as φ k Resource consumption of the task, variables Indicates the first The small base station in the first The probability of executing task k in a given time slot.

[0009] Based on the above single-cycle task model, this invention provides an online single-cycle task offloading method for 5G small cell networks, which learns probabilities. Determine the task uninstallation strategy, which includes the following steps:

[0010] Step 1: The context of all tasks constitutes the task context space. The task context space is then divided into n subspaces, where n is the input parameter. Let the set of subspaces after the division be denoted as . The subspace f includes the context of one or more tasks, and each task k corresponds to a task context subspace. Each task context subspace can correspond to one or more tasks;

[0011] Step 2: Define the context subspace for each task For each small base station The weights are initialized accordingly.

[0012] Step 3: For each small base station Initialize auxiliary variables: Random initial variables, in, The two auxiliary variables represent the weights of the completion probability and resource consumption, respectively. This refers to the unique parameters that distinguish this device from others. This represents the weighting adjustment factor. This represents the resource consumption correction factor.

[0013] Step 4: Configure the time slot For all tasks that need to be processed, perform the following steps:

[0014] Step 4.1: Calculate each small base station In the time slot The probability of executing each task k Constructing vectors

[0015] Step 4.2: Based on the probability vector Greedily allocate time slots Task assigned to small base stations

[0016] Step 4.3: Update the auxiliary variables based on the task feedback results.

[0017] Step 5: Repeat step 4 to allocate and update auxiliary variables for all tasks to be processed in the next time slot in real time until the tasks are unloaded.

[0018] Furthermore, step 4.1 calculates each small base station. In the time slot The probability of executing each task The steps are as follows:

[0019] Step 4.1.1: Obtain small base stations In the time slot The set of tasks within the coverage area

[0020] Step 4.1.2: Obtain the task set Corresponding context set The aforementioned context set Includes task set The input data size, output data size, and time delay requirements for each task;

[0021] Step 4.1.3: For each task k, find its corresponding task context subspace.

[0022] Step 4.1.4: Determine Check if the condition is true. If yes, proceed to step 4.1.5; otherwise, proceed to step 4.1.6. Indicates in time slot Task context subspace j for small base stations The weight, Indicates in time slot The context subspace f corresponding to task k k For small base stations The weights;

[0023] Step 4.1.5: Find the equation that satisfies the condition Let the variable ∈, and let the auxiliary set Record the task context subspace f in the auxiliary set k New weights Proceed to step 4.1.7;

[0024] Step 4.1.6: Let the auxiliary set

[0025] Step 4.1.7: Record the new weights of the subspaces outside the task context subspace in the auxiliary set.

[0026] Step 4.1.8: Calculate the probability of each task k being executed, using the following formula:

[0027]

[0028] in, Indicates in time slot The context subspace f corresponding to task k k For small base stations The new weights Indicates in time slot Task k was attacked by a small base station The probability of execution;

[0029] Step 4.1.9: Return the probability that each task k will be executed. The vectors formed in sequence K represents the time slot. The total number of tasks that need to be processed;

[0030] Furthermore, step 4.2 is based on the probability vector. Greedily allocate time slots The steps for assigning the task to the small base station are: select the largest c. In the time slot Within, the corresponding c tasks are assigned to the small base station. c represents the first The maximum number of wireless devices that a small base station can connect to at the same time.

[0031] Furthermore, step 4.3, updating the auxiliary variables based on the task feedback results, is as follows:

[0032] Step 4.3.1: After completing the calculation, obtain the time slot. Inside, small base stations The context is φ k Rewards for Task k Completion probability Resource consumption

[0033] Step 4.3.2: Update each task according to the following formula Rewards Completion probability Resource consumption

[0034]

[0035]

[0036]

[0037] in, The updated rewards, completion probability, and resource consumption. For indicator functions, In the time slot Inside, allocated to small base stations A set of tasks.

[0038] Step 4.3.3: Calculate the time slot Within each task context subspace, for each small base station Compound rewards Composite completion probability Composite resource consumption The calculation formula is:

[0039]

[0040]

[0041]

[0042] in, Indicates time slot Inside, context φ k The task k corresponding to the task context subspace f, where k is a variable;

[0043] Step 4.3.4: For each satisfying For the task context subspace f, update its corresponding weights using the following formula:

[0044]

[0045] Step 4.3.5: For each satisfying For the task context subspace f, update its corresponding weights using the following formula:

[0046]

[0047] Step 4.3.6: Update the iterative auxiliary variable using the following formula:

[0048]

[0049]

[0050] in, For the first The small base station in the first The probability of completing each task in each time slot is represented by a vector composed of its components. For the first The small base station in the first The probability of executing each task in each time slot is a vector composed of components. For the first The small base station in the first The resource consumption for completing each task in each time slot is represented by a vector composed of components, [·]. + This means taking the larger of the value of the expression within the parentheses and 0.

[0051] This invention solves the problem of single-cycle task offloading in the scenario of wireless devices connecting to 5G small base station networks, fills the gap in single-cycle task offloading methods in the scenario of wireless devices connecting to 5G small base station networks, and can obtain high task rewards while ensuring quality of service (QoS), thus having a high performance ratio. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the task unloading method of the present invention;

[0053] Figure 2 A graph showing the cumulative task rewards for embodiments and comparative examples of the invention;

[0054] Figure 3 The diagram shows the results of violating the minimum computing resource requirement constraint for the embodiments and comparative examples of the invention;

[0055] Figure 4 The diagram shows the results of violating the upper limit of computing resources for small base stations in the embodiments and comparative examples of the invention.

[0056] Figure 5 The graph shows the performance comparison results of the embodiments and comparative examples of the invention;

[0057] The implementation and functional features of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0058] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0059] Example 1

[0060] This embodiment is implemented in a network with the following model parameters: the network has m = 30 small base stations, and the task set within the coverage area of ​​each small base station in each time slot is... The coverage area is randomly distributed, and the number of wireless devices within the coverage area of ​​each small base station is... The range is randomly distributed. Within each time slot, each small base station can connect a maximum of c=20 wireless devices. The reward for each task is... The range is evenly distributed, and the probability of completion is also... The area is evenly distributed, and the resource consumption is... The data is uniformly distributed within the range. Task input is randomly distributed within the range of [5, 20] Mbit, and task output is randomly distributed within the range of [1, 4] Mbit. The maximum computing resource that a small base station can call at the same time is β = 27, and the minimum computing resource required to call at the same time to ensure Quality of Service (QoS) is α = 15. Data changes within t = 10000 time slots are statistically analyzed in the network.

[0061] like Figure 1 As shown, the steps in this embodiment are as follows:

[0062] Step 1: Divide the task context space into 3 subspaces, and denote the set of subspaces after the division as follows:

[0063] Step 2: For each task context subspace For each small base station Initialize the corresponding weights

[0064] Step 3: For each small base station Initialize auxiliary variables:

[0065] Step 4: For each time slot Perform the following steps:

[0066] Step 4.1: Calculate each small base station In the time slot The probability vector for performing each task

[0067] Step 4.2: Based on the probability vector Greedily allocate time slots The task is assigned to the small base station;

[0068] Step 4.3: Update the auxiliary variables based on the task feedback results;

[0069] In this embodiment, step 4.1 calculates each small base station. In the time slot The probability of executing each task The steps are as follows:

[0070] Step 4.1.1: Obtain small base stations Task Collection

[0071] Step 4.1.2: Obtain the task set Corresponding context set

[0072] Step 4.1.3: For each task k, find its corresponding task context subspace.

[0073] Step 4.1.4: Determine Check if the condition is met. If yes, proceed to step 4.1.5; otherwise, proceed to step 4.1.6.

[0074] Step 4.1.5: Find the equation that satisfies the condition Let the variable ∈, and let the auxiliary set Record the task context subspace f in the auxiliary set k New weights Proceed to step 4.1.7;

[0075] Step 4.1.6: Let the auxiliary set

[0076] Step 4.1.7: Record the new weights of the subspaces outside the task context subspace in the auxiliary set.

[0077] Step 4.1.8: Calculate the probability of each task k being executed, using the following formula:

[0078]

[0079] Step 4.1.9: Return the probability that each task k will be executed. The vectors formed in sequence K represents the time slot. The total number of tasks that need to be processed;

[0080] In this embodiment, step 4.2 is based on the probability vector. Greedily allocate time slots The steps for assigning the task to the small base station are: select the largest c. In the time slot Within, the corresponding c tasks are assigned to the small base station. c represents the first The maximum number of wireless devices that a small base station can connect to at the same time.

[0081] In this embodiment, step 4.3, updating the auxiliary variables based on the task feedback results, is as follows:

[0082] Step 4.3.1: After completing the calculation, obtain the time slot. Inside, small base stations The context is φ k Rewards for Task k Completion probability Resource consumption

[0083] Step 4.3.2: Update each task according to the following formula Rewards Completion probability Resource consumption

[0084]

[0085]

[0086]

[0087] in, The updated rewards, completion probability, and resource consumption. For indicator functions, In the time slot Inside, allocated to small base stations A set of tasks.

[0088] Step 4.3.3: Calculate the time slot Within each task context subspace, for each small base station Compound rewards Composite completion probability Composite resource consumption The calculation formula is:

[0089]

[0090]

[0091]

[0092] in, Indicates time slot Inside, context φ k The task k corresponding to the task context subspace f, where k is a variable;

[0093] Step 4.3.4: For each satisfying For the task context subspace f, update its corresponding weights using the following formula:

[0094]

[0095] Step 4.3.5: For each satisfying For the task context subspace f, update its corresponding weights using the following formula:

[0096]

[0097] Step 4.3.6: Update the iterative auxiliary variable using the following formula:

[0098]

[0099]

[0100] in, For the first The small base station in the first The probability of completing each task in each time slot is represented by a vector composed of its components. For the first The small base station in the first The probability of executing each task in each time slot is a vector composed of components. For the first The small base station in the first The resource consumption for completing each task in each time slot is represented by a vector composed of components, [·]. + This means taking the larger of the value of the expression within the parentheses and 0.

[0101] Record the sum of all task rewards for the task offloading allocation scheme in step 4.2 within each time slot, and add it to the sum of task rewards for all time slots before the current time slot as the cumulative task reward. Record the number of times the task offloading allocation scheme in step 4.2 violates the constraint of the minimum service quality required computing resources α within each time slot, and add it to the sum of the number of violations for all time slots before the current time slot as the cumulative number of violations of the minimum service quality required computing resources α. This measures the quality of service (QoS), and a lower value is better. Record the number of times the task offloading allocation scheme in step 4.2 violates the constraint of the upper limit of small base station computing resources β within each time slot, and add it to the sum of the number of violations for all time slots before the current time slot as the cumulative number of violations of the upper limit of small base station computing resources. This measures the quality of service (QoS), and a lower value is better. Calculate the cumulative task reward / (cumulative number of violations of the lower limit of computing resources α + cumulative number of violations of the upper limit of computing resources β) as the performance ratio. This measures service efficiency, and a higher value is better.

[0102] refer to Figure 2 , 3 Figures 4 and 5 show the results of this embodiment. It can be seen that the cumulative task reward of this invention has reached the level of the current advanced method Oracle. Although the cumulative task rewards of vUCB and FML are greater than those of this method, both ignore the resource lower limit α constraint and the resource upper limit β constraint, making it difficult to guarantee the quality of service (QoS) in practice. After running for a period of time and gaining sufficient experience, this method consistently maintains the lowest number of violations of the resource lower limit α constraint and the lowest number of violations of the resource upper limit β constraint, resulting in the best quality of service (QoS). This method consistently maintains the lowest performance ratio and the highest overall service efficiency.

[0103] Comparative Example 1

[0104] In contrast, on the same network as in Example 1, task offloading was performed using the state-of-the-art Oracle method, and cumulative task rewards were recorded, referencing... Figure 2 This is the result of the comparative comparison.

[0105] Comparative Example 2

[0106] In contrast, on the same network as in Example 1, task offloading was performed using the state-of-the-art method vUCB, and the cumulative task rewards, cumulative number of violations of the lower limit of computing resources α, cumulative number of violations of the upper limit of computing resources β, and performance ratio were recorded, with reference to... Figure 2 , 3 Figures 4 and 5 represent the results of this comparative study.

[0107] Comparative Example 3

[0108] In contrast, on the same network as in Example 1, task offloading was performed using the state-of-the-art method FML, and the cumulative task rewards, cumulative number of violations of the lower limit of computing resources α, cumulative number of violations of the upper limit of computing resources β, and performance ratio were recorded, with reference to... Figure 2 , 3 Figures 4 and 5 represent the results of this comparative study.

[0109] Comparative Example 4

[0110] In contrast, on the same network as in Example 1, a random method was used for task offloading. This method randomly assigns tasks and records cumulative task rewards, cumulative number of violations of the lower limit of computing resources α, cumulative number of violations of the upper limit of computing resources β, and performance ratio, with reference to... Figure 2 , 3 Figures 4 and 5 represent the results of this comparative study.

[0111] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0112] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0113] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for online single-cycle task offloading for 5G small cell networks, characterized in that, Includes the following steps: Step 1: Obtain single-cycle, multi-time-slot tasks. The context of all tasks constitutes the task context space. Divide the task context space into... Each subspace, each task context subspace It contains context information for one or more tasks, and the context information for each task exists only in one task context subspace; Step 2: Initialize each task context subspace For each small base station weight ,in, This represents a collection of 5G small cell networks. Indicates the first One 5G small base station, Represents the set of task context subspaces; Step 3: Initialize each small base station Auxiliary variables and random initialization variables ;in, The two auxiliary variables represent the weights of the completion probability and resource consumption, respectively; Step 4: Configure the time slot All tasks that need to be processed are allocated, and the weights and auxiliary variables of each task context subspace for each small base station are updated based on the task feedback results. Specifically, it includes: Step 4.1: Calculate each small base station In the time slot Execute each task probability , forming a vector The calculation formula is: ； in, Indicates in time slot ,Task Corresponding context subspace For small base stations The new weights Indicates in time slot ,Task small base stations The probability of execution; Step 4.2: From the probability vector Choose the largest indivual In the time slot Within, the corresponding c tasks are assigned to the small base station. ,in, Indicates the first The maximum number of wireless devices that a small base station can connect to at the same time; Step 4.3: Update the auxiliary variables based on the task feedback results; Step 5: Repeat step 4 to allocate all tasks to be processed in the next time slot in real time until the tasks are unloaded.

2. The online single-cycle task offloading method for 5G small base station networks according to claim 1, characterized in that, The aforementioned , The calculation formula is as follows: ； ； in, Indicates the device's unique parameters. ; This represents the weighting adjustment factor. This represents the resource consumption correction factor.

3. The online single-cycle task offloading method for 5G small base station networks according to claim 1, characterized in that, Step 4.1 includes: Step 4.1.1: Obtain small base stations In the time slot The set of tasks within the coverage area ; Step 4.1.2: Obtain the task set Corresponding context set The aforementioned context set Includes task set The input data size, output data size, and time delay requirements for each task; Step 4.1.3: For each task Find its corresponding task context subspace ; Step 4.1.4: Determine Check if the condition is true. If yes, proceed to step 4.1.5; otherwise, proceed to step 4.1.

6. Indicates in time slot Task context subspace j for small base stations The weight, Indicates in time slot ,Task Corresponding context subspace For small base stations The weight, Indicates the first The maximum number of wireless devices that a small base station can connect to at the same time; Step 4.1.5: Find the equation that satisfies the condition variables Let the auxiliary set Record the task context subspace in the auxiliary set. New weights Proceed to step 4.1.7; Step 4.1.6: Let the auxiliary set ; Step 4.1.7: Record the new weights of the subspaces outside the task context subspace in the auxiliary set. ; Step 4.1.8: Calculate each task Probability of execution ; Step 4.1.9: Return to each task The probability vector of execution K represents the time slot. The total number of tasks that need to be processed.

4. The online single-cycle task offloading method for 5G small base station networks according to claim 1, characterized in that, In step 4.3, each small base station The auxiliary variable update process includes: Step 4.3.1: Obtain the time slot Inside, small base stations Complete the context as Task Rewards Probability of completion Resource consumption ; Step 4.3.2: Based on each task small base stations The probability of execution and allocation to small base stations Task collection, updated rewards Probability of completion Resource consumption Record the update result as ; Step 4.3.3: Calculate the time slot Within each task context subspace, for each small base station Compound rewards Probability of completion of composite Complex resource consumption ; Step 4.3.4: For each satisfying Task context subspace Update its corresponding weights using the following formula: ； Step 4.3.5: For each satisfying Task context subspace Update its corresponding weights using the following formula: ； Step 4.3.6: Update the iterative auxiliary variable using the following formula: ； ； in, For the first The small base station in the first The probability of completing each task in each time slot is represented by a vector composed of its components. For the first The small base station in the first The probability of executing each task in each time slot is a vector composed of components. For the first The small base station in the first The resource consumption for each task to be completed in each time slot is a vector composed of components. This means taking the larger of the value of the expression within the parentheses and 0.

5. The online single-cycle task offloading method for 5G small base station networks according to claim 4, characterized in that, The aforementioned reward Probability of completion Resource consumption The update formula is as follows: ； ； ； in, The updated rewards, completion probability, and resource consumption. For indicator functions, In the time slot Inside, allocated to small base stations A set of tasks.

6. The online single-cycle task offloading method for 5G small base station networks according to claim 4, characterized in that, Compound rewards Probability of completion of composite Complex resource consumption The calculation formula is as follows: ； ； ； in, Indicates time slot Internal context The task context subspace Corresponding task , For variables.

Images