An edge server type deployment decision method

Abstract

The application discloses an edge server type deployment decision method, which comprises the following steps: rasterizing a map of a city in which an edge server needs to be deployed; then performing discrete Fourier transformation on discrete changes of a task number of each grid; then performing K-means clustering on results after the discrete Fourier transformation of each grid, so as to divide a plurality of grids of the map into K clusters; and finally performing time domain analysis on the K clusters to obtain an edge server type deployment decision result of each cluster. The application can select the most suitable edge server for a region in the city, and avoid waste of computing resources or deficiency of computing resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication device deployment, and in particular to an edge server type deployment decision method. BACKGROUND

[0002] With the development of the Internet of Things, a large number of sensing devices have been produced explosively, which have generated a large amount of data. However, most of the Internet of Things devices have limited resources for computing, storage and communication, and therefore have to turn to cloud computing (CC). Centralized cloud computing cannot meet the strict requirements of 5G and the Internet of Things (IoT), and edge computing (EC) has emerged as the times require.

[0003] With the growth of mobile terminal devices, Internet of Things devices are no longer limited to a certain place, which poses great challenges to edge computing. In order to solve the above challenges, in 2014, the European Telecommunications Standards Institute (ETSI) first proposed the concept of a new architecture, i.e. mobile edge computing (MEC). According to ETSI, the definition of MEC can be described as "mobile edge computing provides an IT service environment and cloud computing function at the edge of the mobile network, within the radio access network (RAN) and close to the edge of the mobile user". MEC provides cloud computing functions in the RAN. MEC allows direct mobile traffic between the core network and the end user, and directly connects the user to the nearest edge network supporting cloud services. Deploying MEC at the base station can enhance computing power and avoid bottlenecks and system failures. Unmanned aerial vehicles (UAVs) have achieved great success in military and civilian applications due to their easy deployment, flight in low altitude without affecting the ground and high altitude, low cost and mobility. Considering the large-scale deployment of Internet of Things devices and limited communication capabilities, the application of UAVs in the Internet of Things ecosystem has attracted widespread attention. Based on the advantages of UAVs, UAVs can assist MEC systems, especially in areas with obstacles and scattered and highly mobile nodes. UAVs can play many roles in MEC, such as MEC servers, users, relays, data collectors or dispatchers, support entities, and attackers. UAVs can be well equipped with MEC-related devices to provide mobile edge computing services in large-scale Internet of Things scenarios to improve the quality of service (QoS) of users. Ground Internet of Things nodes can offload heavy computing tasks to mobile clouds through uplink and downlink with UAVs.

[0004] However, there is no good solution to the problem of deploying edge servers at present, i.e. choosing to deploy mobile edge servers or fixed site edge servers, which also leads to the waste of computing resources of edge servers.

[0005] Xu analyzed the traffic pattern of five regions in a city by 9600 cell towers, from time and frequency domain. He distinguished the cell towers that were not very obvious in time domain by frequency domain.

[0006] Zhang analyzed the time series of mobile traffic data by 6400 cell towers, and divided the city into five traffic patterns.

[0007] Yang proposed a multi-UAV deployment mechanism based on differential evolution (DE), modeled the access problem as a generalized assignment problem (GAP), and then solved it by a near-optimal solution algorithm.

[0008] Wang proposed a two-layer optimization method to jointly optimize UAV deployment and task scheduling, aiming to minimize system energy consumption. In the upper layer, he proposed a differential evolution algorithm with a cancellation operator to optimize the deployment of UAVs. In the lower layer, based on the given UAV deployment, he converted the task scheduling into a 0-1 integer programming problem. He also established constraints to optimize the non-convex problem.

[0009] Hu proposed an alternating optimization algorithm to jointly optimize computing resource scheduling, bandwidth allocation, and UAV trajectory under the constraints of task constraints, information causal relationship constraints, bandwidth allocation constraints, and UAV trajectory constraints, to minimize the weighted sum of energy consumption of UAVs and UEs.

[0010] Qian decomposed the problem into three sub-problems when solving non-convex optimization problems, and solved them respectively using integer programming and continuous convex optimization methods.

[0011] Zhang proposed a simplified lo-norm algorithm to solve non-convex optimization problems.

[0012] Xu, Zhang only analyzed the traffic changes in the city, but did not mention how to use this change to deploy edge servers. Yang, Wang, Hu, Qian, Zhang all directly deployed edge servers or clustered the load based on distance when deploying edge servers, without classifying the changes in the number of tasks to be offloaded in different regions.

[0013] For the city, the number of tasks to be unloaded varies in different areas. In the city center, the number of tasks to be unloaded is significantly different between day and night, while in the suburbs far from the city center, the number of tasks to be unloaded is not significantly different between day and night. For the city center, using fixed sites can easily cause waste of computing resources or not enough computing resources to unload the number of tasks. For the suburbs, using edge computing servers carried by drones can easily cause the problem of insufficient endurance of drones. Therefore, it is very important to analyze and classify the number of tasks in the city and use the corresponding edge server. SUMMARY

[0014] The purpose of the present application is to overcome the shortcomings of the prior art and provide an edge server type deployment decision method.

[0015] To achieve the above purpose, the technical solution provided by the present application is:

[0016] An edge server type deployment decision method comprises:

[0017] The map of the city where the edge server needs to be deployed is rasterized;

[0018] Discrete Fourier transform is performed on the discrete variation of the number of tasks of each grid to obtain a sample data set The result of the discrete Fourier transform of the discrete variation of the number of tasks of the mth grid is represented;

[0019] K-means clustering is performed on the result of the discrete Fourier transform of each grid, thereby dividing the plurality of grids of the map into K clusters;

[0020] Time domain analysis is performed on the K clusters to obtain the edge server type deployment decision result of each cluster.

[0021] Further, the K-means clustering of the result of the discrete Fourier transform of each grid comprises:

[0022] Input the sample data set The number of clusters K and the maximum number of iterations N;

[0023] Step 1: randomly select K samples from the sample data set D as the initial K centroid vectors: {u1, u2,..., u K};

[0024] Step 2:

[0025] Step 2-1: iteration number n = 1, initialize cluster division C to

[0026] Step2-2: Calculate sample data x i and the distance of each centroid μ j :

[0027]

[0028] Where, i = 1, 2,..., m, m is the total number of grids obtained after the grid of the map; j = 1, 2,..., K;

[0029] According to the principle of proximity, the sample data x i is divided into the cluster where the nearest centroid is located and the

[0030] Step2-3: Recalculate the new centroid of all sample data in C j

[0031] Step2-4: If all K centroids have not changed or n is equal to N, go to Step3, otherwise, go to Step2-2;

[0032] Step 3: Output cluster division C = {C1, C2,..., CK}. K

[0033] Further, the number of clusters K is determined by drawing a residual sum of squares curve and using the elbow method.

[0034] Further, the K clusters are analyzed in the time domain. For clusters where the number of tasks changes significantly during the day and night, deploy mobile edge servers. For clusters where the number of tasks does not change significantly during the day and night, deploy fixed site edge servers.

[0035] Compared with the prior art, the principles and advantages of the technical solution are as follows:

[0036] The scheme rasterizes the map of the city where the edge server needs to be deployed; then performs discrete Fourier transform on the discrete changes of the number of tasks of each grid; then performs K-means clustering on the results of the discrete Fourier transform of each grid, thereby dividing the multiple grids of the map into K clusters; finally, the K clusters are analyzed in the time domain to obtain the edge server type deployment decision result of each cluster.

[0037] The technical solution can select the most suitable edge server for the area in the city, avoiding waste of computing resources or insufficient computing resources. BRIEF DESCRIPTION OF DRAWINGS

[0038] ​​In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the services required to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0039] Figure 1 A schematic diagram of the principle flow chart of the edge server type deployment decision method of the present application;

[0040] Figure 2 A schematic diagram of the map gridding of the city where the edge server needs to be deployed;

[0041] Figure 3 A schematic diagram of the change of SSE (sum of squared errors) with K value in K-means clustering;

[0042] Figure 4 A distribution diagram of a city from the city center to the suburbs;

[0043] Figure 5 A schematic diagram of the change of SSE with K value when the fifth cluster is clustered by distance-based K-means;

[0044] Figure 6 An effect diagram of the fifth cluster;

[0045] Figure 7 A schematic diagram of the average number of tasks of each grid in the twelve clusters in a day;

[0046] Figure 8 A schematic diagram of the change of the number of tasks of each cluster in a day;

[0047] Figure 9 A schematic diagram of the change of the number of tasks of each cluster in a week. DETAILED DESCRIPTION

[0048] The present application will be further described below in combination with specific embodiments:

[0049] As shown in the drawings, the edge server type deployment decision method described in the present embodiment comprises the following steps: Figure 1

[0050] A1, gridding the map of the city where the edge server needs to be deployed, as shown in the drawings; Figure 2

[0051] A2, performing discrete Fourier transform on the discrete change of the number of tasks of each grid to obtain a sample data set represents the result of the discrete Fourier transform of the discrete change of the number of tasks of the mth grid.​​

[0052] A3, K-means clustering is performed on the result of the discrete Fourier transform of each grid, thereby dividing the plurality of grids of the map into K clusters;

[0053] wherein the K-means clustering performed on the result of the discrete Fourier transform of each grid comprises:

[0054] an input sample data set a number K of clusters, a maximum number N of iterations;

[0055] Step 1: randomly select K samples from the sample data set D as initial K centroids: {μ1, μ2,..., μK} ; K};

[0056] Step 2:

[0057] Step 2-1: iteration number n = 1, initialize cluster division C to

[0058] Step 2-2: calculate the distance between sample data x i and each centroid μ j :

[0059]

[0060] wherein i = 1, 2,..., m, m is the total number of grids obtained after gridding the map; j = 1, 2,..., K;

[0061] According to the nearest principle, sample data x i is divided into the cluster wherein the distance is closest to the centroid, and

[0062] Step 2-3: recalculate the new centroid j for all sample data in C

[0063] Step 2-4: if all K centroids have not changed or n equals N, go to Step 3, otherwise, go to Step 2-2;

[0064] Step 3: output cluster division C = {C1, C2,..., CK}. K

[0065] In the above, the number K of clusters is determined by drawing a residual sum of squares curve and selecting the elbow method.

[0066] ​This step uses K-means clustering in the frequency domain to group rasters with similar task number variations into one category, and then selects an appropriate edge computing server for deployment within that category.

[0067] A4. Perform time-domain analysis on K clusters. For clusters where the number of tasks changes significantly between day and night, deploy mobile edge servers. For clusters where the number of tasks does not change significantly between day and night, deploy fixed-site edge servers.

[0068] To demonstrate the authenticity and effectiveness of the above method, the following simulation experiment was conducted:

[0069] In this experiment, a city was considered as a 60×80 rectangular area with 4800 grid cells. The sequence of tasks performed on these 4800 grid cells every 10 minutes over 28 days was used to deploy multiple drone-mounted mobile edge computing servers and fixed-site edge servers across this area, seeking the optimal deployment method.

[0070] First, by analyzing the task count sequence of these 4800 rasters every 10 minutes over 28 days, we obtain the daily and weekly raster count sequences for these 4800 rasters over 28 days. Then, we perform a Fast Fourier Transform on these raster count sequences for each of the 10-minute, daily, and weekly periods over 28 days, respectively, to obtain three-dimensional pulses. Finally, we perform the first K-means clustering on these 4800 rasters using these three-dimensional pulses.

[0071] Figure 3 The diagram shows the variation of SSE with the value of K in K-means clustering, indicating that K-means clustering performs best when the value of K is 5. Therefore, setting the value of K to 5 in K-means clustering yields the following clustering results: Figure 5 As shown.

[0072] Figure 4 It resembles a city distribution map from the city center to the suburbs, with five clusters nested one inside the other, which strongly suggests that K-means clustering with 4800 grids in the frequency domain is effective. Within the same cluster, the number of tasks exhibits similar characteristics and trends.

[0073] As can be seen, the fifth cluster has a large grid, but the number of tasks per grid is small. This is detrimental to deployment when multiple drones are deployed on this cluster, and it also places high demands on the coverage radius of the drone server. Therefore, a second K-means clustering is performed on this cluster based on distance.

[0074] Figure 5The change of SSE with K value when the fifth cluster is clustered by distance-based K-means is shown. The results show that when the K value is 8, the clustering effect is better, so the K value is selected as 8 for distance-based K-means clustering of the fifth cluster, and the clustering effect is as shown in Figure 5 .

[0075] Figure 6 The clustering effect of the fifth cluster reaches the expected effect, and the enclosure formed by the fifth cluster is divided into eight parts.

[0076] Figure 7 The average number of tasks of each grid in the twelve clusters in a day is shown. This makes it more intuitive to see the clusters, i.e. the center and suburbs of the city.

[0077] Figure 8 The change of the number of tasks of each cluster in a day is shown. Figure 9 The change of the number of tasks of each cluster in a week is shown. This illustrates the effectiveness of the previous clustering of grids in the frequency domain. In the city center, the number of tasks of the grid is closely related to human activities, and the number of tasks is more during the day and less at night. In the remote suburbs, the number of tasks of the grid tends to be flat, and the change is not particularly obvious. It can be found that the maximum and minimum values of the number of tasks in each cluster appear at different times, which is closely related to people's life, such as industrial areas, residential areas, transportation areas and entertainment areas. We propose a scenario that fixed sites are deployed in remote suburbs and unmanned aerial vehicle-based servers are deployed in areas closely related to human activities.

[0078] Unmanned aerial vehicle-mounted mobile edge servers are deployed for the grids of the first cluster, the second cluster, the third cluster and the fourth cluster, and fixed site edge servers are deployed for the grids of the fifth cluster.

[0079] Experiments prove that the clustering effect in the frequency domain is effective, and through the time domain analysis of the clustering results, it can be decided to select fixed site edge servers or mobile edge servers for deployment.

[0080] The above-described embodiments are only the preferred embodiments of the present application, and are not intended to limit the scope of the present application. Any changes made in shape and principle according to the present application should be covered within the scope of protection of the present application.

Claims

1. A method for making deployment decisions for edge server types, characterized in that, include: Rasterize the maps of cities where edge servers need to be deployed; Perform a discrete Fourier transform on the discrete variation of the number of tasks for each raster to obtain the sample dataset. This represents the result of performing a discrete Fourier transform on the discrete change in the number of tasks in the m-th grid. K-means clustering is performed on the results of discrete Fourier transform of each grid cell to divide the multiple grid cells of the map into K clusters; Perform time-domain analysis on K clusters to obtain the edge server type deployment decision results for each cluster; The number of clusters, K, was determined by using the elbow method after plotting the residual sum of squares curve.

2. The edge server type deployment decision method according to claim 1, characterized in that, K-means clustering is performed on the results of the discrete Fourier transform of each grid cell, including: Input sample dataset Number of clusters K, maximum number of iterations N; Step 1: Randomly select K samples from the sample dataset D as the initial K centroids: {μ1,μ2,...,μ K }; Step 2: Step 2-1: With iteration count n = 1, initialize the cluster partition C to... Step 2-2: Calculate the sample data x i and each centroid μ j Distance: Where i = 1, 2, ..., m, m is the total number of grid cells obtained after rasterization of the map; j = 1, 2, ..., K; Based on the principle of proximity, the sample data x i It is divided into the cluster containing the nearest centroid. And update Step 2-3: For C j Recalculate the new centroid for all sample data. Step 2-4: If none of the K centroids have changed or n equals N, then go to Step 3; otherwise, go to Step 2-2. Step 3: Output cluster partition C = {C1, C2, ..., C} K } 3. The edge server type deployment decision method according to claim 1, characterized in that, Perform time-domain analysis on K clusters. For clusters where the number of tasks changes significantly between day and night, deploy mobile edge servers. For clusters where the number of tasks does not change significantly between day and night, deploy fixed-site edge servers.

Images