Multi-cooperative server deployment method in edge computing scenarios

Abstract

The present invention discloses a method for deploying multiple cooperative servers in an edge computing scenario. The present invention deploys edge servers based on a two-stage incremental deployment method, which includes the following two stages: the first stage is a server incremental deployment stage based on a greedy strategy; the second stage is a load distribution stage based on a convex optimization approximation method; the optimal server deployment solution is found by iterating the above two steps with an increasing number of servers. In response to the problems of redundant resources and excessive overhead in edge server deployment, the present invention proposes a deployment framework for cooperative services and a two-stage incremental deployment method to improve server resource utilization and reduce overall server deployment overhead. The present invention is applicable to the field of edge computing.

Description

Technical Field

[0001] The present invention belongs to the field of Internet of Things and edge computing technology, and specifically relates to a multi-cooperative server deployment method for edge computing scenarios. Background Art

[0002] With the continuous development of 5G communication technology, an increasing number of compute-intensive and latency-sensitive applications, such as real-time video streaming analysis, autonomous driving, and augmented and virtual reality, have emerged. However, Internet of Things (IoT) devices lack the computing power to run these applications locally. While cloud computing offers powerful computing capabilities, the long transmission distances result in excessive end-to-end latency, making it unable to meet the extremely low latency requirements of these applications. The combination of 5G and edge computing is considered a promising solution for empowering the IoT with intensive computing capabilities.

[0003] Edge computing is a computing paradigm that moves computing resources to the edge of the network. This overcomes the significant latency inherent in cloud computing and can simultaneously meet the demands of applications with high computational load and extremely low latency. The development of 5G communication technology has significantly reduced the transmission latency of the last hop in edge computing, further improving edge network performance. However, the new features introduced by 5G communication technology also present numerous challenges for edge computing.

[0004] Excessive deployment overhead is one of the main challenges facing 5G edge computing. This means that a large number of powerful edge servers are required to achieve full network coverage and complete terminal tasks. Since 5G communication range is significantly reduced, more servers must be deployed to cover the same area as 4G, which increases deployment overhead. Furthermore, considering the rapid increase in latency-sensitive applications in edge computing, a denser deployment of servers is required to meet the extremely high Quality of Service (QoS) requirements of these applications, and the actual deployment cost may be even higher.

[0005] Furthermore, in edge computing, the peak computing resource demands of terminal nodes at a given moment far exceed their average demands. If all edge servers were deployed based on the peak resource demands of terminal nodes, resource utilization would be very low most of the time, a major factor in increasing deployment costs. Reducing overall deployment costs (especially in 5G scenarios) while ensuring user experience has become a major challenge.

[0006] Based on this, a multi-cooperative server deployment method for edge computing scenarios that can reduce the deployment cost of edge servers while meeting users' real-time experience needs is urgently needed in this field. Summary of the Invention

[0007] The present invention provides a multi-cooperative server deployment architecture and method in an edge computing scenario. By allowing multiple edge servers to utilize the real-time load differences between each other to collaboratively process the fluctuating loads of terminals in the interlaced areas, the architecture and method can solve the problem of deployment resource waste caused by the large difference between the peak demand and average demand of terminal resources, especially in 5G edge networks, reduce overall deployment costs, and improve resource utilization.

[0008] The present invention is achieved through the following technical solutions:

[0009] A multi-cooperative server deployment method in an edge computing scenario includes the following steps: Step S10: discretizing the region according to the effective communication range of the edge server and the location information of the regional nodes; Step S20: deploying the edge server according to the two-stage incremental deployment method, wherein Step S20 specifically includes the following two stages: the first stage is the server incremental deployment stage based on the greedy strategy; the second stage is the load distribution stage based on the convex optimization approximation method; Step S30: completing the full coverage of the terminal nodes in the target edge network according to the deployment plan obtained by the aforementioned two-stage incremental deployment method.

[0010] In one embodiment, the region discretization is further performed based on the average workload of the region.

[0011] In one embodiment, in the first stage, a location with the greatest optimization possibility and the greatest cooperation potential is selected as the deployment location of the new server for incremental deployment.

[0012] In one embodiment, in the first stage, a minimum number of servers are used to achieve full coverage of IoT devices, and then the candidate set locations with the greatest optimization possibility are selected based on this basic deployment plan for incremental deployment of servers to reduce the total server deployment overhead.

[0013] In one embodiment, the first stage further includes: selecting a server with the greatest possibility of optimization; specifically, calculating the optimization possibility index of each server and finding a deployed server with the greatest possibility of cooperation.

[0014] In one embodiment, the optimization possibility index of each server is calculated by calculating the difference between the peak computing resource request amount and the average computing resource request amount of the server; the larger the difference, the greater the optimization possibility of the server.

[0015] In one embodiment, after the selected server is determined, a new server deployment location is determined based on the cooperation capability indicators of all candidate sets around the server, and the deployment location is the candidate deployment location with the greatest cooperation potential.

[0016] In one embodiment, the cooperation capability index of all candidate sets around the server is calculated based on the difference between the sum of the peak values of independent computing requests of all IoT devices that can be covered by deploying the server at the candidate deployment location and the peak computing requests after server aggregation; the larger the difference, the more suitable the candidate deployment location is for deploying a new server.

[0017] In one embodiment, in the second stage, multiple rounds of random server selection are performed, and multiple time slots with the highest number of resource requests for the selected servers are optimized to obtain a solution with the minimum number of overall computing resource deployments.

[0018] In one embodiment, the solution for obtaining the minimum number of overall computing resource deployments specifically includes the following steps:

[0019] Step 3.1) Model the problem of balancing the workload of each time slot as a non-convex optimization problem;

[0020] Step 3.2) For the non-convex optimization problem, use the log-sum-exp approximation function to convert the non-convex optimization problem into a convex optimization problem;

[0021] Step 3.3) For the transformed convex optimization problem, use the KKT condition to solve it.

[0022] Based on the above two-stage incremental deployment method, the present invention proposes an implementation plan for server location and resource number with the least resource deployment overhead for a given network.

[0023] The present invention has the following advantages and beneficial effects:

[0024] 1. To address the problem of waste of deployment resources caused by the large difference between the peak demand and average demand of terminal resources in 5G edge networks, the present invention proposes a solution that uses multiple edge servers to utilize the real-time load differences between each other to collaboratively handle the fluctuating loads of shared terminals, thereby effectively improving resource utilization and reducing the overall deployment overhead of edge servers.

[0025] 2. The present invention proposes a two-stage incremental deployment method to plan the above-mentioned server deployment problem, jointly optimizes the two coupled sub-problems of server deployment location and workload distribution, and decouples the two sub-problems, achieving a good compromise between multi-server collaboration and service coverage. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention.

[0027] Figure 1This is a schematic diagram of the cooperation mechanism of multiple cooperating edge servers proposed in the present invention;

[0028] Figure 2 A schematic diagram of the process of the two-stage incremental deployment method proposed in the present invention;

[0029] Figure 3 Example of selecting an incremental server location for the first phase of a two-phase incremental deployment approach. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0031] First, the multi-cooperative server deployment architecture for edge computing scenarios in the present invention is introduced. The edge computing scenario of the present invention takes the 5G edge computing scenario as an example, but is not limited to this. The architecture of this embodiment allows multiple edge servers to utilize the real-time load differences between each other to cooperate to process the fluctuating load of shared terminals, thereby meeting the computing needs of terminals in the region with fewer hardware resources and higher resource utilization, so as to solve the deployment resource waste problem caused by the large difference between the peak demand and average demand of terminal resources in the 5G edge network, reduce the overall deployment cost, and improve resource utilization.

[0032] In one embodiment of the present invention, the 5G edge architecture has the following specific features:

[0033] 1. The total amount of computing resources of the edge server is based on the common computing power unit on the market. The resources of the total server are multiples of this basic unit, and its deployment complies with the ETSI standard.

[0034] 2. Different types of terminal device workloads can be offloaded to different edge servers for collaborative processing. The workload requests of each terminal device vary over time. Through this edge server collaboration, peak workloads can be distributed to multiple surrounding edge servers, effectively reducing the deployment of redundant resources and lowering deployment costs.

[0035] In addition, during specific implementation, it is necessary to consider that the overall deployment cost is divided into two major parts: the infrastructure cost caused by the number of edge server deployments and the computing power cost caused by the number of resources.

[0036] During the edge server deployment phase, opportunities for multi-server cooperative scheduling are considered, and an effective balance is made between server coverage and server cooperation opportunities. The deployment scale, location, and hardware resources of edge servers, as well as the load distribution between servers, are rationally planned to achieve an efficient and economical edge server deployment strategy.

[0037] Figure 1 A schematic diagram of the cooperation mechanism for multiple cooperating edge servers is presented, along with two specific load distribution methods. In one embodiment of the present invention, the collaborative processing mechanism for multiple cooperating servers in the edge architecture described above is utilized. The edge servers in this invention support two load distribution mechanisms: "single terminal device tasks are simultaneously offloaded to multiple edge servers" and / or "single terminal device tasks are offloaded to a single edge server."

[0038] 1. "Single Terminal Device Task Simultaneously Offloaded to Multiple Edge Servers" Mechanism: This invention innovatively leverages the data-parallel nature of terminal device tasks. This means that the computing load can be partitioned and offloaded to multiple edge servers for parallel processing. This mechanism leverages the weak correlation between resource requests from different terminal devices over the same period. Through load balancing, it effectively handles peak computing resource requests from terminal devices, reducing the deployment of redundant resources and overall deployment overhead.

[0039] 2. "Single-Terminal Device Task Offloading to a Single Edge Server" Mechanism: Not all terminal nodes in this invention implement the "Single-Terminal Device Task Offloading to Multiple Edge Servers" mechanism; most nodes still implement the traditional offloading mechanism. For situations where a terminal device has only one edge server within its communication range, or for situations where a terminal device has multiple edge devices within its communication range but its resource requests are relatively stable, the terminal device can perform one-to-one offloading using the traditional mechanism.

[0040] refer to Figure 2 This embodiment is a specific deployment method of the edge deployment architecture mentioned above - a two-stage incremental deployment method, which is as follows:

[0041] The two-stage incremental deployment method determines the number and deployment location of servers through self-incremental traversal, so that the server can improve resource utilization while meeting the service quality and resource requests of IoT terminal devices, achieving the goal of minimizing the overall server deployment overhead. That is, iteratively optimize server deployment and load distribution to achieve a trade-off between cooperation opportunities and coverage. The execution process of this method is as follows: Figure 2 As shown, the specific steps include:

[0042] Step S10: discretize candidate points according to the regional node location information and the regional average workload, and discretize the continuous target area into a candidate set.

[0043] First, the continuous target region needs to be discretized into a candidate set. Unlike other approaches that use peak workload as the region discretization criterion, this method uses the average regional workload and evaluates the cooperation opportunities of the candidate set based on optimization potential and cooperation ability. Specifically, the region is discretized based on the effective communication range of the edge server, regional node location information, and / or the average regional workload. Then, proceed to step 2.

[0044] Step S20: Deploy edge servers using a two-stage incremental deployment method. This step consists of two layers: an inner and an outer layer. The outer layer loops through the process of incrementally increasing the number of servers to the maximum number of deployed servers, i.e., the total number of edge servers. The inner layer uses a two-stage method to incrementally deploy new servers based on the previous round of deployment plans, selecting locations with the greatest optimization space and cooperation potential. This method then allocates computing tasks to find the minimum deployment cost.

[0045] The two-stage incremental deployment method includes a first stage and a second stage. An example of the incremental server location selection operation in the first stage of the two-stage incremental deployment method is as follows: Figure 3 shown.

[0046] The first phase involves incremental server deployment based on a greedy strategy. This involves first using the minimum number of servers to achieve full coverage of IoT devices. Then, based on this basic deployment plan, incremental server deployment is performed in candidate locations with the highest potential for optimized cooperation, minimizing overall server deployment overhead. This phase utilizes two metrics, deployment optimization potential and server cooperation capability, to guide the selection of incremental server deployment locations.

[0047] Specifically include:

[0048] Step 2.1) Select the server with the greatest potential for optimization cooperation.

[0049] To identify deployed servers with the highest potential for collaboration, we need to calculate each server's optimization probability index to facilitate server selection. This is calculated as the difference between the server's peak and average computing resource requests. Obviously, for servers with a large difference between peak and average load, balancing the peak load by adding new deployment nodes can reduce overall deployment overhead.

[0050] Step 2.2) Select the deployment location of the newly added server that can be optimized.

[0051] It is necessary to find a candidate deployment location with the greatest cooperation potential around the server selected in the previous step. Here, it is necessary to calculate the cooperation capability index of all candidate sets around the server to select the new server deployment location. The specific index is calculated by the difference between the sum of the peak computing requests of all IoT devices that can be covered by deploying the server at the candidate deployment location and the peak computing request after server aggregation.

[0052] Obviously, if the two are equal, it means that the peak request times of the nodes within the coverage area of the server overlap and there is no room for optimization; and the larger the difference between the two, the more effective the load balancing can be achieved by deploying servers at this location, and it is suitable for deploying new servers.

[0053] The second phase involves load allocation based on a convex optimization approximation method. This involves randomly selecting servers over multiple rounds and optimizing for the time slots with the highest resource requests for the selected servers. This transforms the continuous-time optimization problem into a workload allocation problem within a single time slot, ultimately minimizing the overall number of computing resources deployed. Specifically, the solution includes:

[0054] Step 3.1) The problem of balancing the workload of each time slot is modeled as a non-convex optimization problem.

[0055] Step 3.2) For the non-convex optimization problem, use the log-sum-exp approximation function to convert the non-convex optimization problem into a convex optimization problem.

[0056] Step 3.3) For the transformed convex optimization problem, use the KKT condition to solve it.

[0057] Finally, after each step in the above loop is completed, an optimal solution for the deployment cost corresponding to the number of servers will be obtained. After comparison and updating, the deployment plan of the optimal solution is saved.

[0058] In other words, it can be formally expressed as a mixed integer nonlinear programming problem, which includes two coupled sub-problems of server deployment and load distribution, and the optimization goal is to minimize the edge server deployment overhead.

[0059] By iterating the above two steps with increasing the number of servers, the optimal server deployment solution is found.

[0060] Step S30: Complete full coverage of terminal nodes in the target edge network according to the deployment plan obtained by the aforementioned two-stage incremental deployment method.

[0061] In the above scheme, the contents that are not specifically introduced, such as modeling as a non-convex optimization problem, converting the non-convex optimization problem into a convex optimization problem, solving the KKT conditions, regional discretization, etc., themselves (but not the entire scheme constituted by them) are conventional technical means in this field, and the present invention will not elaborate on them.

[0062] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-cooperative server deployment method in an edge computing scenario, characterized in that: The following steps are involved: Step S10: discretize the region according to the effective communication range of the edge server and the location information of the regional nodes; Step S20: Deploy edge servers according to a two-stage incremental deployment method. Step S20 specifically includes the following two stages: The first phase is the incremental server deployment phase based on a greedy strategy. Locations with the greatest optimization potential and cooperation potential are selected as new server deployment locations for incremental deployment. The minimum number of servers is used to achieve full coverage of IoT devices. Based on this basic deployment plan, the candidate set locations with the greatest optimization potential are then selected for incremental server deployment to reduce the total server deployment overhead. The server with the greatest optimization potential is selected by calculating the optimization possibility index for each server and finding the deployed server with the greatest cooperation potential. The optimization possibility index for each server is calculated by taking the difference between the peak computing resource request amount and the average computing resource request amount for that server. The larger the difference, the greater the optimization possibility of that server. The second stage is a load allocation stage based on a convex optimization approximation method. Through multiple rounds of random server selection, optimization is performed on multiple time slots with the highest number of resource requests from the selected servers to obtain a solution with the minimum number of overall computing resource deployments. The solution for obtaining the minimum number of overall computing resource deployments specifically includes the following steps: Step 3.1) Model the problem of balancing the workload of each time slot as a non-convex optimization problem; Step 3.2) For the non-convex optimization problem, use the log-sum-exp approximation function to convert the non-convex optimization problem into a convex optimization problem; Step 3.3) For the transformed convex optimization problem, use the KKT condition to solve it; Step S30: Complete full coverage of terminal nodes in the target edge network according to the deployment plan obtained by the aforementioned two-stage incremental deployment method.

2. The multi-cooperative server deployment method in the edge computing scenario according to claim 1 is characterized in that: The regional discretization is further performed based on the average workload of the region.

3. The multi-cooperative server deployment method in the edge computing scenario according to claim 1 is characterized in that: After the selected server is determined, the deployment location of the new server is determined based on the cooperation capability indicators of all candidate sets around the server. The deployment location is the candidate deployment location with the greatest cooperation potential.

4. The multi-cooperative server deployment method in the edge computing scenario according to claim 3 is characterized by: Based on the difference between the sum of the peak values of independent computing requests of all IoT devices covered by the server deployed at the candidate deployment location and the peak computing requests after server aggregation, the cooperation capability index of all candidate sets around the server is calculated; the larger the difference, the more suitable the candidate deployment location is for deploying a new server.

Images