Resource configuration method for differentiated user application requests

Abstract

The present invention relates to the field of computing matching technology, and specifically to a resource configuration method for differentiated user application requests. In an edge cloud computing architecture, the influence of factors such as the upper limit of edge node resources, the upper limit of the fronthaul of data uploaded by edge nodes to a large edge cloud, and the properties of wireless links are taken into account. In order to truly reflect the differentiated needs of users, factors such as the user's sensitivity to latency and the urgency of task completion are digitized to form influencing factors affecting user satisfaction. Furthermore, latency and energy cost are taken as the main factors to consider in user satisfaction. Under the 'first-come, first-served' resource configuration principle, overall user satisfaction is maximized, thereby improving the shortcomings of existing offload configuration calculation methods.

Description

Technical Field

[0001] The present invention relates to the field of computing and matching technology, and in particular to a resource configuration method oriented to differentiated user application requests. Background Art

[0002] With the rapid development of wireless communication technology, diverse network applications, such as high-definition video and VR gaming, are becoming increasingly popular. Users of these latency-sensitive, computationally intensive applications require ultra-low latency services. However, existing mobile devices are limited by their computing power. Running these computationally intensive applications can take longer than expected, making it difficult to guarantee low-latency services at all times.

[0003] Computation offloading provides an effective solution to the above problems, which is to allow mobile terminal devices to transfer part of the application calculation to other platforms with more sufficient computing resources for processing, thereby reducing the computing burden and energy consumption of the terminal devices. Early cloud computing, as a traditional application offloading architecture, supports user terminal devices to transfer computing tasks to large edge clouds for processing through wide area networks. However, when people urgently need to ensure ultra-low latency services, uploading data to large edge clouds and then receiving downlink data cannot meet everyone's expectations. Edge cloud computing, as an extension and expansion of cloud computing, deploys a large amount of computing resources in wireless access networks close to users, allowing them to receive application calculations uploaded by users and return processing results, which can achieve lower latency network services, but the complex system also brings challenges to the allocation of computing resources.

[0004] In reality, when allocating computing resources to users, the system must consider factors such as the deployment status of edge nodes, the varying computational workloads and latency sensitivities of user applications, the limited computing capacity of edge nodes, and the varying pricing structures of large edge clouds and edge nodes. However, whether for single-objective or multi-objective optimization, the design of algorithms still lacks compatibility with edge cloud computing applications. Therefore, when designing algorithms, while pursuing optimal goals, it is crucial to ensure low complexity and high performance. Summary of the Invention

[0005] The purpose of the present invention is to provide a resource configuration method for differentiated user application requests. Based on an edge computing system consisting of a large edge cloud, a small edge cloud, and users, the method takes latency and energy cost as the main factors to consider in user satisfaction, and maximizes overall user satisfaction under the "first come, first served" resource configuration principle.

[0006] To achieve the above object, the present invention provides a resource configuration method for differentiated user application requests, comprising the following steps:

[0007] Initialize the user layer and select the initial computing platform attribute parameters;

[0008] Determine the calculation configuration model and estimate the profit model;

[0009] Maximizing total user benefits in problem planning;

[0010] The computing platform completes user configuration based on a 'first come, first served' configuration principle.

[0011] Optionally, the computing platform includes two platforms: a large edge cloud and a small edge cloud. The computing platform farther away from the user is a large edge cloud with sufficient computing resources. The computing platform closer to the user is composed of small distributed edge nodes L=={1,2,...,L}, which have limited storage and computing capabilities and are connected to the large edge cloud through a fronthaul link.

[0012] Optionally, the user layer is composed of users M={1, 2, ..., M}, has heterogeneous application requests, and performs application offloading to edge nodes or large edge clouds.

[0013] Optionally, in the computing configuration model, all users design strategies based on computing service prices and latency factors during one allocation process. The computing task allocation decision variable for user m is: x m ={x mi |i∈{1, 2, ..., L+1}, x mi ∈{0,1}}, m∈M;

[0014] Among them, x ml =1, l∈{1,2,···,L} means that user m∈M offloads the computing task to the edge node l∈L, x m(L+1) =1 means that user m uploads the computing task to the large edge cloud.

[0015] Optionally, in the revenue model, the revenue of user m∈M is:

[0016]

[0017] Among them, ω m α m The user's satisfaction in completing the task, the unit power consumption price of large edge cloud is p c , the power consumption required to transmit a unit bit of data is θ c , calculate the power consumption per bit of data φ c , the unit power consumption price p of edge node l∈L l , the power consumption required to transmit a unit bit of data is θ l , calculate the power consumption per bit of data φ l The price that users need to pay for edge node calculation is pl φ l α m +p l θ l β m D m , the price to pay for choosing a large edge cloud is p C φ C α m +p C θ C β m D m , where β m D m Indicates the size of the data packet after the calculation is completed. The impact of application response delay on user m's satisfaction is shown in Figure 2. Delay includes the time for data transmission, queuing, and calculation.

[0018] Optionally, in the process of maximizing the total user benefit in problem planning, the objective optimization problem P1 is formulated as:

[0019]

[0020] Define x = {x m |m∈M}, constraint C1 constrains the business to select edge nodes in two situations: success or failure. C2 constrains each user's application to request only a unique computing node. C3 constrains the application's requested resource size to be below the threshold of the available computing resources of the computing node.

[0021] The present invention provides a resource configuration method for differentiated user application requests. In the edge cloud computing architecture, the influence of factors such as the upper limit of edge node resources, the fronthaul limit of edge node uploading data to the large edge cloud, and the nature of the wireless link are taken into account. In order to truly reflect the differentiated needs of users, factors such as the user's sensitivity to delay and the urgency of task completion are digitized to form influencing factors affecting user satisfaction. Furthermore, delay and energy cost are taken as the main considerations for user satisfaction. Under the "first come, first served" resource configuration principle, the overall user satisfaction is maximized, thereby improving the shortcomings of the existing offload configuration calculation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1It is a flow chart of a resource configuration method for differentiated user application requests of the present invention.

[0024] Figure 2 It is a structural diagram of an implementation model of the present invention.

[0025] Figure 3 It is a schematic diagram of an algorithm flow of a resource configuration method for differentiated user application requests of the present invention.

[0026] Figure 4 It is a schematic diagram of the number of iterations required to complete matching under different numbers of users and edge nodes in a specific embodiment of the present invention.

[0027] Figure 5 It is a bar chart comparing the total user revenue of the algorithm of the present invention and a single uninstall target in a specific embodiment of the present invention.

[0028] Figure 6 This is a bar chart comparing the total user revenue of the algorithm of the present invention and the method of only offloading to edge nodes under different numbers of edge nodes in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0030] See also Figure 1 The present invention provides a resource configuration method for differentiated user application requests, comprising the following steps:

[0031] S1: Initialize the user layer and select the initial computing platform attribute parameters;

[0032] S2: Determine the calculation configuration model and estimate the profit model;

[0033] S3: Maximize total user benefits in problem planning;

[0034] S4: The computing platform completes user configuration based on a 'first come, first served' configuration principle.

[0035] The following is a further explanation based on specific terms and implementation procedures:

[0036] See also Figure 2 , Figure 2This diagram illustrates the implementation model structure of the resource configuration method for differentiated user application requests. User application offloading can choose between two computing platforms: a large edge cloud and a small edge cloud. The computing platform located relatively farther from the user is the large edge cloud, which contains basic hardware such as a central processing unit (CPU), random access memory (RAM), and disks, and has sufficient computing resources. The computing platform located relatively close to the user consists of small, distributed edge nodes L=={1,2,...,L}, deployed on devices such as micro base stations and wireless network controllers. These nodes have limited storage and computing capabilities and are connected to the large edge cloud via fronthaul links. The user layer consists of users M={1,2,...,M}, who request heterogeneous applications and offload them to edge nodes or the large edge cloud. Users use their terminal devices to offload applications to edge nodes in the small edge cloud or the large edge cloud via wireless transmission. Each edge node provides a unified interface protocol for the user terminal devices it covers, enabling data from different user terminal devices to be shared.

[0037] In the model, user m∈M has an application computing task that needs to be performed, which requires computing resources α m , the size of the data packet transmitted in the network is D m Large edge cloud and edge nodes can execute user computing requests. The total amount of user request computing processed by edge node l∈L cannot exceed its own computing resource limit S l Users need to pay a price for using the computing services of large edge clouds or edge nodes. This price is affected by two factors: the energy consumed by large edge clouds or edge nodes in processing computing tasks and their different energy charging standards.

[0038] In addition to the service price paid, the user's choice is also influenced by the sensitivity to latency. Since edge nodes are closer to users than large edge cloud segments, users who are eager to pursue low-latency services are more inclined to choose to offload applications to edge nodes. The present invention considers the amount of data transmitted through the link between the user and the computing platform, the wireless transmission speed R between the user and the edge computing node and the large edge cloud segment, and the time delay requirement. ml 、R F (m∈M, l∈L), the waiting time before edge nodes and large edge clouds process the calculation, as well as the time for edge nodes and large edge clouds to process the calculation tasks, constitute the computing task completion delay t of user m∈M ml (l∈L),t mc To represent the latency required for application requests to be computed in edge nodes or large edge clouds. Introducing the user's sensitivity coefficient to latency ξ m , making the delay factor affect user choices. Specifically, the overall user satisfaction will be significantly reduced as the delay increases.

[0039] User upload D m After the data volume, β m D m The amount of data, where β m The coefficient for comparing task processing results with application data.

[0040] Specifically, the following four aspects are explained:

[0041] 1. Computational Configuration Model

[0042] Assume that the user satisfaction incentive coefficient is ω m , will ω m α m As the user's satisfaction after completing the task. Calculate the request size D m The data packets are uploaded to the computing platform. In one allocation, all users design strategies based on the computing service price and latency factors, with the goal of maximizing user satisfaction. The decision variables for computing task allocation for user m are:

[0043] x m ={x mi |i∈{1, 2, ..., L+1}, x mi ∈{0,1}},m∈M

[0044] Among them, x ml =1, l∈{1,2,···,L} means that user m∈M offloads the computing task to the edge node l∈L, x m(L+1) =1 means that user m uploads the computing task to the large edge cloud. The user is only allowed to offload the computing task request to one target, that is, when When , it means that the computing task request of user m∈M fails. The user-side incentive coefficient ξ for delay sensitivity m This reflects the varying degrees of latency requirements for different users. User satisfaction is expressed as the difference between user benefit and price paid.

[0045] 2. Revenue Model

[0046] User computing tasks can request edge node set L = {1, 2, ···, L} or large edge cloud computing. In this model, the large edge cloud computing resources S c Sufficient, (S1, S2, ···, S L ) are the resource limits of L edge nodes. Users need to pay for the computing services of edge nodes and large edge clouds, which is expressed as the product of the energy consumed by the computing platform to perform computing tasks and transmit data packets and its energy price. It is known that the unit power consumption price p of large edge clouds is c , the power consumption required to transmit a unit bit of data is θc , calculate the power consumption per bit of data φ c The unit power consumption price p of edge node l∈L l , the power consumption required to transmit a unit bit of data is θ l , calculate the power consumption per bit of data φ l The price a user needs to pay for edge node calculation is p l φ l α m +p l θ l β m D m , the price to pay for choosing a large edge cloud is p C φ C α m +p C θ C β m D m , where β m D m Indicates the size of the data packet after the calculation is completed. Consider that the transmission rate from user m∈M to node l∈L is R ml , the data transmission rate of the link from the user to the large cloud is R F , the waiting delay of the user selecting the edge node l∈L is The user selects the waiting delay of a large edge cloud as The delay of user m∈M is:

[0047]

[0048] In summary, the benefit of user m∈M is:

[0049]

[0050] In the above formula, the first term represents the satisfaction each user receives from completing the computation. The second and third terms represent the cost the user pays for the computation at the edge node or in a large edge cloud. These terms represent the product of the energy spent on executing the computation and transmitting the data packets, and the energy price. The impact of application completion delay on user m's satisfaction is shown in Figure 2. Delay includes the time for data transmission, queuing, and calculation.

[0051] 3. Problem Planning

[0052] Based on the above single-user benefit formula, the present invention maximizes the total user benefit by designing a calculation configuration algorithm. The target optimization problem P1 is as follows:

[0053]

[0054] Define x = {xm |m∈M}, constraint C1 restricts the business's selection of edge nodes to either success or failure. Considering the limited availability of resources, C2 constrains each user's application to request only a single compute node, and C3 constrains the application's requested resource size to be below the threshold for the available compute resources on the compute node.

[0055] 4. Algorithm derivation based on the "first come, first served" configuration principle

[0056] It's easy to configure each user with a corresponding available app uninstallation platform and calculate the total user benefit under this configuration. However, finding the configuration with the highest total benefit from these different user configuration combinations requires traversing all possible combinations, which is clearly an NP-hard problem. Using an exhaustive search method—listing all app uninstallation scenarios for all users and then finding the optimal solution—would require a complexity of (total number of users, total number of edge nodes). Furthermore, due to the limited computing resources of edge computing nodes, the benefits of different users are coupled, and within a single user's own benefits, the optimal choices for latency and payment costs are also coupled. These factors make the problem very difficult to solve.

[0057] In the present invention, a low-complexity computing configuration algorithm is designed based on the 'first come, first served' computing configuration principle. Specifically, the user sends an application uninstallation request to the computing platform with the highest revenue (do not select the node with negative revenue) according to the revenue that the user can obtain on each computing platform; then, each computing platform selects the users that can be accommodated and rejects other users according to its own computing resource limitations and the 'first come, first served' rule, thus completing the configuration of one user. For rejected users, the suboptimal available computing platform is selected again according to the above rules and the computing platform decides whether to accept them. This iteration ends until all users are matched or the upper limit of the number of iterations is reached. Please refer to the algorithm flow diagram. Figure 3 .

[0058] The specific algorithm is shown in Table 1:

[0059]

[0060] Furthermore, in order to verify the convergence speed and higher total user benefits of the present invention, the present invention proposes a simulation embodiment:

[0061] The model parameters are set as follows. The transmission rate from user to edge node and cloud is 50Kb / s. Unit resource benefit ω m The value range is [0.1, 0.2], and the resource size requested by a single task is α m The value range is [100,1000]Kb, and the size of the data packet transmitted in the link for each request is D mThe value range is [1,9]Kb. User waiting satisfaction coefficient ω mt The value range is [20, 40]. Calculate the data multiplication coefficient β m = 0.1. The delay caused by requesting edge nodes and large-scale edge cloud computing t d1 , t d2 The value range is [0.1, 0.2]s, and the upper limit of edge node computing resources S l =5000Kb. Considering 1Kb of data as unit data and 1J of energy as unit energy, the energy consumed by edge nodes and large edge cloud to transmit unit data is θ l =3J, calculate the energy consumed per unit data Unit energy price of edge node p l The value range is [3,5], and the unit energy price p in large edge cloud is c =2.

[0062] Table 2(a) Performance of the algorithm of the present invention in the embodiment

[0063]

[0064] Table 2(b) Performance of exhaustive search method in the embodiment

[0065]

[0066] As shown in Table 2, when the number of users is 20 and the number of edge computing nodes is 20, the total user revenue changes with the number of iterations. When the algorithm of the present invention is used, the total user revenue will reach the maximum after the system iterates twice, and the total revenue is close to the maximum total revenue calculated by the exhaustive search method. However, the exhaustive search method requires (12) 100 Therefore, the complexity of the present invention is much smaller than that of the exhaustive search method.

[0067] Figure 4 This figure shows the number of iterations required to achieve maximum system benefit for different numbers of users and edge nodes. As shown, regardless of the ratio of users to edge nodes, the number of iterations does not change dramatically, demonstrating the rapid convergence and stability of our algorithm.

[0068] Figure 5 This chart shows the overall user benefit after iterations using our algorithm, considering only offloading to edge nodes, only offloading to large edge clouds, and so on. By comparing each set of histograms, we can observe that the overall user benefit after using our algorithm is greater than the other two scenarios, demonstrating the significant effectiveness of our algorithm in optimizing overall benefits.

[0069] Figure 6The difference in total user revenue between using the proposed algorithm and offloading only to edge nodes under different edge node computing capacities is demonstrated. It can be seen that when edge node computing capacity is small, that is, when computing resources are scarce, the user revenue obtained using the proposed algorithm is far greater than that obtained by selecting only edge nodes. However, when edge node computing resources are sufficient, the proposed algorithm still has certain advantages, indicating that the proposed algorithm coordinates large and small edge clouds to better configure user application requests.

[0070] In summary, the present invention has the following advantages:

[0071] 1) The method proposed in this paper is based on an edge computing system consisting of a large edge cloud, small edge cloud segments, and users. It investigates the problem of heterogeneous multi-user application offloading, considering the coupling relationship between user benefits and service costs, and maximizing total user benefits.

[0072] 2) By designing a user benefit formula, adapting to the influencing factors of different factors, and reducing the algorithm running time in a targeted manner, the algorithm of the present invention provides a solution for the reasonable offloading of a large number of heterogeneous delay-sensitive and computationally intensive applications.

[0073] 3) To reflect the limited computing power and energy of edge nodes and ensure the quality of service for latency-sensitive applications, this paper pursues the design of a low-complexity and fast-convergence algorithm. Simulation comparisons show that while maintaining these characteristics, this paper also maximizes total user benefits.

[0074] The above disclosure is only a preferred embodiment of the present invention, and certainly cannot be used to limit the scope of the rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A resource configuration method for differentiated user application requests, characterized in that: The following steps are involved: Initialize the user layer and select the initial computing platform attribute parameters; Determine the calculation configuration model and estimate the profit model; In the revenue model, the revenue of user m∈M is: Among them, ω m α m The user's satisfaction in completing the task, the unit power consumption price of large edge cloud is p c , the power consumption required to transmit a unit bit of data is θ c , calculate the power consumption per bit of data φ c , the unit power consumption price p of edge node l∈L l , the power consumption required to transmit a unit bit of data is θ l , calculate the power consumption per bit of data φ l The price that users need to pay for edge node calculation is p l φ l α m +p l θ l β m D m , the price to pay for choosing a large edge cloud is p C φ C α m +p C θ C β m D m , where β m D m Indicates the size of the data packet after the calculation is completed. The impact of application completion delay on user m's satisfaction is represented by the time it takes for data transmission, queuing, and calculation. Maximizing total user benefits in problem planning; The computing platform completes user configuration based on a 'first come, first served' configuration principle.

2. The resource configuration method for differentiated user application requests according to claim 1, characterized in that: The computing platform includes two platforms: large edge cloud and small edge cloud. The computing platform farther away from the user is the large edge cloud with sufficient computing resources. The computing platform closer to the user is composed of small distributed edge nodes L=={1,2,...,L}, with limited storage and computing capabilities, and is connected to the large edge cloud through a fronthaul link.

3. The resource configuration method for differentiated user application requests according to claim 2, characterized in that: The user layer consists of users M={1, 2, ..., M}, which have heterogeneous application requests and perform application offloading to edge nodes or large edge clouds.

4. The resource configuration method for differentiated user application requests according to claim 3, characterized in that: In the computing configuration model, all users design strategies based on computing service prices and latency factors during one allocation process. The computing task allocation decision variable for user m is: m ={x mi |i∈{1, 2, ..., L+1}, x mi ∈{0,1}}, m∈M; Among them, x ml =1, l∈{1,2,···,L} means that user m∈M offloads the computing task to the edge node l∈L, x m(L+1) =1 means that user m uploads the computing task to the large edge cloud.

5. The resource configuration method for differentiated user application requests according to claim 1, wherein: In the process of maximizing the total user benefits in problem planning, the objective optimization problem P1 is formulated as: Where, define x={x m |m∈M}, constraint C1 constrains the business to select edge nodes in two situations: success or failure. C2 constrains each user's application to request only a unique computing node. C3 constrains the application's requested resource size to be below the threshold of the available computing resources of the computing node.

Images