A collaborative edge server deployment and resource scheduling method, storage medium and system
Through collaborative edge server deployment and resource scheduling methods, the deployment and resource mapping of edge servers are optimized, and the problems of high energy consumption and long response time caused by improper deployment of edge servers in mobile edge computing are solved, achieving lower energy consumption and Response time, adapt to fluctuations in the task arrival rate of users moving.
Patent Information
- Application Number
- CN202111483496.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-07
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-12-07
AI Technical Summary
The prior art ignores the scientific deployment of edge servers in mobile edge computing, resulting in high energy consumption and long response time, which cannot meet users' latency and security needs.
A collaborative edge server deployment and resource scheduling method is proposed. By establishing cost and energy consumption optimization models, combining improved loser elimination firework algorithms and genetic algorithms, dynamically adjusting the deployment and resource mapping of edge servers, and optimizing the deployment of edge servers and Resource scheduling to reduce energy consumption and response time.
It effectively reduces the deployment cost and overall energy consumption of edge servers, improves resource utilization, reduces task response time, improves user experience quality, solves single point of failure problem, and adapts to fluctuations in task arrival rate caused by user movement.
Smart Images

Figure CN114363962B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of mobile edge computing edge server deployment, and in particular to collaborative edge server deployment and resource scheduling strategies. Background Art
[0002] With the rapid development of mobile Internet, applications such as remote office, telemedicine, augmented reality, interactive games, virtual reality, and Internet of Vehicles have proliferated. The massive data generated by them all rely on cloud platforms for processing. However, the centralized data processing model can no longer meet their requirements in terms of latency, security, and reliability. Mobile edge computing is a technology that provides IT services and computing power to users at the edge of the network. It has the characteristics of proximity, low latency, and high bandwidth, so it is regarded as a key technology to solve the above problems. Many Internet companies and scientific research institutions have invested in the research of mobile edge computing to promote its rapid development. In mobile edge computing, most studies assume that edge servers have been properly deployed or that there are edge servers on each base station. They only focus on the impact of task offloading, resource allocation, and service migration strategies on latency, energy consumption, and other targets, ignoring that scientific deployment of edge servers is also an important influencing factor. The edge server deployment problem refers to selecting a suitable location to deploy edge servers based on the network status of an area. It is a prerequisite for dealing with problems such as computing offloading, resource allocation, and service migration. When deploying edge servers, considering that most edge nodes are not suitable for deploying edge servers and the high cost of random deployment, edge servers are generally deployed on base stations. Some studies also consider the mobility of users and install edge servers on drones or intelligent robots. Scientifically deploying edge servers can not only reduce system energy consumption and reduce the expenditure costs of service providers, but also reduce task response time and improve user experience quality.
[0003] The deployment of edge servers involves many factors, including capital expenditure, operating expenditure, service quality, and experience quality. Comprehensively weighing all factors and combining reasonable deployment of edge servers with resource allocation strategies can effectively improve the resource utilization of edge servers, increase the revenue of service providers, reduce the overall energy consumption and task delay of edge servers, meet the personalized needs of users, and improve service quality and user experience. Therefore, it is crucial to study a reasonable edge server deployment strategy.
[0004] After searching, the application publication number CN109819046B, a method for scheduling virtual computing resources of the Internet of Things based on edge collaboration, belongs to the field of virtualized wireless networks, and particularly relates to computing resource scheduling in edge collaboration for Internet of Things applications. This invention designs a virtual computing resource scheduling architecture based on edge collaboration, which makes full use of the idle virtualized resources of edge horizontal intelligent IOT devices, vertical sensor nodes and infrastructure, and significantly improves resource utilization and QoS of intelligent Internet of Things applications. In addition, the proposed algorithm can select the optimal computing offload path, minimize the computing resources occupied by the application while optimizing the data transmission delay, enable more Internet of Things devices to obtain computing resources, and ensure the normal operation of the application.
[0005] Difference: The first part of the computing resource scheduling in this paper is based on each task component as a unit, and the appropriate edge server is selected for computing offload. The second part considers the task relevance, transmits the results of the previous part of the task to the subsequent task computing node, and starts the task computing process of the subsequent node. The task component is a task that cannot be split for computing. This means that each task in the base station will participate in dynamic scheduling.
[0006] The resource scheduling scheme of the present invention is based on the total task volume of the base station within a period of time, indicating that the tasks within this period can only be unloaded to the edge server mapped by me. My should be used as the first part of his edge server selection, indicating that within this period of time, unloading tasks to the two edge servers mapped by me can meet the capacity constraint. Then use his first part to determine which of the two mapped edge servers each task should be unloaded to for execution.
[0007] If they are different, the solution of the present invention will have a shorter running time, but a slightly higher response time, because the present invention considers the base station as a whole; while his solution will have a longer running time because it considers the offloading problem of each task. The present invention considers the whole, while he focuses more on details. My solution connects two edge servers during mapping, and can offload a task to the two edge servers at the same time, while his only considers the one-to-one mapping problem. Summary of the invention
[0008] The present invention aims to solve the above problems of the prior art. A method for reducing energy consumption and cost under delay constraints is proposed. Taking into account the situation where the total task arrival rate of the edge server caused by user movement exceeds its execution capacity, a dynamic resource scheduling strategy is proposed to reduce the expected response time of the collaborative edge server deployment and resource scheduling method, storage medium and system. The technical solution of the present invention is as follows:
[0009] A collaborative edge server deployment and resource scheduling method comprises the following steps:
[0010] Based on the access latency and the capacity of edge servers, a cost and energy consumption optimization model for collaborative edge server deployment is established;
[0011] Obtain the initial solution of the cost and energy optimization model, map the base station to multiple edge servers; adjust the existing mapping according to capacity and delay constraints, and obtain the first optimized deployment plan for edge servers. The parameters used in the deployment plan include: the task arrival rate of the base station The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations connected to the edge server R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ).
[0012] The fireworks algorithm based on improved loser elimination is used to adjust the number of deployed edge servers and the mapping relationship to obtain the second optimal deployment plan for edge servers. The parameters used in this plan include: the task arrival rate of the base station The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations connected to the edge server R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ). This paper improves the calculation method of the explosion range of fireworks. The size of the explosion range in this paper represents the number of edge servers modified in the iterative optimization. The explosion range in the original algorithm is related to the explosion range in the previous iteration. However, after multiple iterations of the algorithm, the fitness value often remains unchanged, resulting in the explosion range shrinking continuously and eventually becoming 0, falling into a local optimal solution. Therefore, this paper improves the calculation method of the explosion range of fireworks to be only related to the number of edge servers deployed by the fireworks. When the fireworks produce sparks with a fitness value better than itself in this generation, the explosion range will be reduced, and the calculation method is half of the number of edge servers deployed by the fireworks multiplied by the reduction coefficient; otherwise, the explosion range will be expanded, that is, half of the number of edge servers deployed by the fireworks multiplied by the magnification coefficient.
[0013] According to the edge server deployment results, a model for minimizing the expected response time of dynamic resource scheduling is established;
[0014] A hybrid algorithm combining the bare-bone fireworks algorithm and the genetic algorithm is used to dynamically adjust the mapping between edge servers and base stations.
[0015] Furthermore, the cost and energy consumption optimization model for collaborative edge server deployment is established based on the access delay and the capacity of the edge server, specifically including:
[0016] Based on the base station data set, the base station locations are obtained and the base station set B = {B1, B2, ..., B M} and the edge server set S = {S1,S2,...,S N}(M≥N), the topological structure of the base station and edge server is represented as an undirected graph G=(V,E), V=BUS represents the set of base stations and edge servers. Assume that the edge server is located at the same location as the base station, so it has a total of M possible deployment locations; obtain the parameters of the base station and edge server, including the task arrival rate of the base station The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and build a cost and energy minimization model based on these parameters.
[0017] Furthermore, the deployment cost and energy consumption model specifically includes:
[0018] The energy consumption of the edge server consists of two parts: runtime energy consumption and idle energy consumption. The formula is:
[0019]
[0020] P(u(t))=KP max +(1-K)P max *u(t) (2)
[0021]
[0022] Where Energy represents the overall energy consumption of the system, P(u(t)) represents the system power from time t1 to time t2, t1 represents the time when task offloading starts, t2 represents the time when all tasks are completed, n represents the nth edge server, and P max represents the power of the edge server under full load, u(t) represents the CPU utilization from time t1 to time t2, Represents the edge server S n The total task arrival rate, p represents the rate at which edge servers process tasks, K represents the percentage of idle state to full load state of edge servers, and u represents CPU utilization. The deployment cost of edge servers is positively correlated with the number of deployed servers, so the deployment cost of edge servers can be expressed as:
[0023]
[0024] Among them, x m Indicates whether in B m Deploy edge servers at m Set to 1 to indicate that B is deployed m Deploy edge servers at B, m It is set to 0, and the z-score normalization method is used to normalize the energy consumption and cost respectively.
[0025] Furthermore, the initial solution of the cost and energy consumption optimization model is obtained, specifically including:
[0026] First, a mapping relationship is established. If the base station and the edge server are in the same location, only the edge server is mapped. Otherwise, two edge servers need to be mapped. Then, the existing mapping is adjusted according to the capacity and delay constraints, and the base station task arrival rate is allocated to the edge server to obtain the initial deployment plan for the first optimization.
[0027] Furthermore, the constraints of the deployment cost and energy consumption model specifically include:
[0028] Mapping constraint: When a base station and an edge server are located at the same location, it can only be mapped to the edge server, and the edge server is the proximal edge server of the base station; otherwise, the base station needs to connect to a proximal edge server and a distal edge server, and the formula is:
[0029]
[0030]
[0031] The binary variable Indicates base station B m Whether to connect to the near-end edge server S n , a value of 1 indicates connection, and a value of 0 indicates no connection; Indicates base station B m Whether to connect to the remote edge server S n , a value of 1 indicates connection, and a value of 0 indicates no connection. m represents the mth base station, B represents the set of all base stations, and S represents the set of all edge servers.
[0032] Deployment location constraints: Regardless of how the base stations are mapped, each edge server is co-located with a base station, allowing The base station set S mapped to the edge server n ∈S ε , then the formula is:
[0033]
[0034]
[0035] in represents a set of base stations, and these base stations are co-located with the edge server, S ε represents a set of edge servers with the locations of deployed base stations, S n Indicates the nth edge server.
[0036] Delay constraint: Each edge server has a fixed coverage range. When the Euclidean distance between the base station and the edge server exceeds this threshold, the base station's tasks cannot be offloaded to the edge server for execution. Let R cover represents the coverage of the edge server, then the formula is:
[0037]
[0038]
[0039] Constraint on the number of base stations connected to the server: The number of base stations connected to each edge server cannot exceed the threshold. When there are too many base stations mapped to the same edge server, the number of tasks arriving at the edge server at the same time will increase, which will increase the task response time. Let R ln represents the maximum number of base stations that the edge server can serve simultaneously, and the formula is:
[0040]
[0041] Where R ln Indicates the upper limit of the number of base stations that the edge server can connect to. When this limit is exceeded, the service quality of the edge server will be greatly reduced.
[0042] Capacity constraint: The arrival rate of tasks on each edge server cannot exceed the rate at which the edge server processes tasks. Represents the edge server S j The total task arrival rate, constant p represents the execution capability of the edge server, then the formula is:
[0043]
[0044] Furthermore, the fireworks algorithm based on loser elimination is adopted to adjust the number of deployed edge servers and the mapping relationship to obtain a second optimized deployment plan of the edge servers, which specifically includes:
[0045] (1) Initialize each parameter ( β, C a , Cr ,γ,τ), where is the number of fireworks, is the total number of sparks, β is the parameter controlling the spark distribution, C a is the magnification factor, C r is the shrinkage coefficient, γ is the number of losers in each iteration, and τ is the maximum number of iterations;
[0046] (2) Obtain the initial deployment plan for each firework, which contains four data: the number of edge servers, overall energy consumption, edge server location, and mapping plan. Calculate the cost and energy consumption z-score standardized values according to the deployment plan as the fitness value;
[0047] (3) According to the formula Calculate the number of sparks produced by each firework, where SN rank Indicates the number of sparks produced by fireworks with a fitness value ranking of rank. Represents the total number of sparks produced by all fireworks, Indicates the number of fireworks, and rank is the fitness ranking of the firework. At the same time, the explosion range of each firework is calculated, that is, the number of edge servers that need to be modified when each firework produces sparks. The formula is:
[0048]
[0049] Among them, Am g is half the number of edge servers in the g-th generation Fireworks. represents the explosion range of the g-th generation fireworks, f(X g ) is the fitness value of the g-th generation of fireworks;
[0050] (4) Randomly delete some deployed edge servers according to the explosion magnitude, and then redeploy some servers according to the mapping, delay and capacity requirements to meet the base station needs;
[0051] (5) Update the fireworks and reinitialize the eliminated fireworks. The algorithm ends when the maximum number of iterations is reached. Otherwise, the number of iterations increases by 1 and execute (3).
[0052] Furthermore, the expected response time minimization model of dynamic resource scheduling is established according to the edge server deployment result, which specifically includes:
[0053] Get the parameters of the base station and edge server, including the arrival rate of the task that the base station participated in the previous calculation Current task arrival rate The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max, coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and a model minimizing the expected response time is constructed based on these parameters.
[0054] Furthermore, the hybrid algorithm combining the bare bones fireworks algorithm and the genetic algorithm specifically includes:
[0055] (1) Initialize each parameter ( R mut ,τ′), where is the total number of sparks, R mut is the probability of mutation, τ′ is the maximum number of iterations;
[0056] (2) Regularly monitor whether there is any fluctuation in the task arrival rate of the base station. If not, keep the original mapping unchanged. Otherwise, use a hybrid algorithm that combines the bare-bone fireworks algorithm and the genetic algorithm to adjust the mapping;
[0057] (3) Calculate the expected task arrival rate λ of each edge server avg , when the resources allocated to the edge server exceed λ avg It will not allocate resources to any base station, where is the total task arrival rate of the base station during runtime;
[0058] (4) Obtain the initial mapping plan, randomly select unassigned base stations to map to edge servers, give priority to edge servers with more remaining capacity during mapping, and consider whether there are edge servers with a total task arrival rate exceeding λ after the allocation is completed avg ; Until all base stations are assigned; the initial mapping scheme is executed times, the best solution is the firework, and the rest are sparks;
[0059] (5) Perform a crossover mutation process on each spark; for the mapping scheme of each base station in the spark, if a mutation operation occurs, the mapping scheme is deleted; otherwise, a crossover operation occurs, and the base station has a 50% probability of selecting the mapping scheme of the corresponding base station in the fireworks, and the other 50% probability of keeping the mapping scheme unchanged;
[0060] (6) Find and remap base stations that violate mapping and capacity constraints;
[0061] (7) Update the fireworks. If the maximum number of iterations is reached, the algorithm ends. Otherwise, the number of iterations increases by 1 and the algorithm executes step 5.
[0062] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the collaborative edge server deployment and resource scheduling method as described in any one of the above is implemented.
[0063] A scheduling system based on any one of the methods, comprising:
[0064] Cost and energy consumption optimization model building module: used to build a cost and energy consumption optimization model for collaborative edge server deployment based on access latency and edge server capacity;
[0065] First optimization module: used to obtain the initial solution of the cost and energy consumption optimization model, base station mapping multiple edge servers; adjust the existing mapping according to capacity and delay constraints, and obtain the first optimization deployment plan of edge servers. The parameters used in the deployment plan include: base station task arrival rate The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations connected to the edge server R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ).
[0066] Second optimization module: Using the fireworks algorithm based on improved loser elimination, adjust the number of deployed edge servers and the mapping relationship to obtain the second optimization deployment plan for edge servers. The parameters used in this plan include: base station task arrival rate The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations connected to the edge server R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ). This paper improves the calculation method of the explosion range of fireworks. The size of the explosion range in this paper represents the number of edge servers modified in the iterative optimization. The explosion range in the original algorithm is related to the explosion range in the previous iteration. However, after multiple iterations of the algorithm, the fitness value often remains unchanged, resulting in the explosion range shrinking continuously and eventually becoming 0, falling into a local optimal solution. Therefore, this paper improves the calculation method of the explosion range of fireworks to be only related to the number of edge servers deployed by the fireworks. When the fireworks produce sparks with a fitness value better than itself in this generation, the explosion range will be reduced, and the calculation method is half of the number of edge servers deployed by the fireworks multiplied by the reduction coefficient; otherwise, the explosion range will be expanded, that is, half of the number of edge servers deployed by the fireworks multiplied by the magnification coefficient.
[0067] Dynamic adjustment module: According to the deployment results of the edge server, a model for minimizing the expected response time of dynamic resource scheduling is established; a hybrid algorithm combining the bare-bone fireworks algorithm and the genetic algorithm is used to dynamically adjust the mapping between the edge server and the base station.
[0068] The advantages and beneficial effects of the present invention are as follows:
[0069] 1. The present invention takes into account the collaboration of edge servers when deploying edge servers, which is more in line with reality. At the same time, it can solve the single point failure problem in the task offloading process and effectively reduce the deployment cost and overall energy consumption.
[0070] 2. Based on the reasonable deployment of edge servers, a dynamic resource scheduling strategy is introduced. According to the fluctuation of base station task arrival rate caused by user movement, the mapping between base stations and edge servers is dynamically adjusted, thereby reducing the expected task response time.
[0071] Innovation 1: The collaboration of edge servers is considered in both the deployment and resource scheduling stages of edge servers, allowing base stations that are not in the same location as the edge servers to map two edge servers. Advantages: Server collaboration can solve the problem of reduced user experience caused by single point failure of the server. Secondly, in a one-to-many mapping mode, the base station can offload multiple tasks to multiple servers at the same time, or offload different parts of a task to different servers to reduce task response time. The formulas defined based on server collaboration mainly include: (1), (8-15), (20), (21-25), among which (8), (10), (12), (15), (21), (23), (25) are existing formulas for single mapping, and (1), (9), (11), (13), (14), (20), (22), (24) are newly added formulas for dual mapping, which are defined for themselves. The reason why this point is difficult to think of is that when connecting two edge servers, we do not know how to reasonably distribute the base station load to the edge server. This article uses the relevant knowledge of queuing theory to distribute the load according to the probability of the cache queue of the proximal server being full. The relevant formula comes from "Operations Research Tutorial (Fifth Edition)".
[0072] Innovation 2: In the deployment phase, the mapping process is separated from the load distribution process when obtaining the initial solution. In the mapping process, in order to reasonably deploy edge servers, the maximum number of base stations R mapped by edge servers is introduced. ln , making the mapping and deployment more reasonable, and then distributing the load according to the mapping relationship according to formula (1). The reason is not easy to think of: because in the original single mapping, it is easier to obtain the deployment plan, and whether the mapping process and the load distribution process are separated has little effect on the deployment effect. In the dual mapping, there is a connection between the servers, and often the establishment of a mapping requires modifying the load of multiple edge servers. Therefore, it is easier to obtain a good solution by splitting the two processes.
[0073] Innovation point 3 uses the fireworks algorithm based on improved loser elimination in the deployment phase. The reason for using this algorithm is that in the iterative optimization process, I found that obtaining a better solution often requires a better initial solution, and it is difficult to obtain a better initial solution. Therefore, the fireworks algorithm with a loser elimination mechanism is introduced. Several fireworks will be reinitialized in each iteration. If the initialized solution is better, it will replace the original fireworks solution to improve the efficiency of obtaining good solutions. Secondly, the calculation method of the explosion range of fireworks is improved. The size of the explosion range in this article represents the number of edge servers modified in the iterative optimization. The calculation method is: when a firework produces a spark with a fitness value better than itself in this generation, the explosion range will be reduced. The calculation method is half of the number of edge servers deployed by the firework multiplied by the reduction coefficient; otherwise, the explosion range will be expanded, that is, half of the number of edge servers deployed by the firework multiplied by the amplification coefficient.
[0074] Innovation point 4 uses a hybrid algorithm combining the bare-bone fireworks algorithm and the genetic algorithm in the resource scheduling stage. Resource scheduling requires high algorithm efficiency and good results, while the fireworks algorithm based on improved loser elimination has low operating efficiency and gives priority to access delay factors when mapping, which is not suitable for solving resource scheduling problems. Genetic algorithms are often used to solve this problem, but there are problems such as large populations and complex selection processes. This paper uses the bare-bone fireworks algorithm to overcome these problems. At the same time, the crossover mutation mechanism in the genetic algorithm is improved and used for iterative optimization of the bare-bone fireworks algorithm. The improvement is as follows: For the mapping scheme of each base station in the spark, if a mutation operation occurs, the mapping scheme is deleted; otherwise, a crossover operation occurs, and the base station has a 50% probability of selecting the mapping scheme of the corresponding base station in the fireworks, and the other 50% probability of keeping the mapping scheme unchanged. Advantages: It can be well applied to the resource scheduling model of this paper, improves the operating efficiency of the algorithm, and still obtains a good solution. Reasons that are not easy to think of: It may be because of different scenarios, and different scenarios correspond to different improvements. BRIEF DESCRIPTION OF THE DRAWINGS
[0075] Figure 1 The present invention provides a preferred embodiment of the present invention as a flow chart;
[0076] Figure 2 A flowchart for establishing a mapping when obtaining an initial deployment plan;
[0077] Figure 3 A flowchart for distributing load when taking the initial deployment plan;
[0078] Figure 4 A flowchart of the collaborative edge server deployment algorithm of the present invention;
[0079] Figure 5 It is a flow chart of the dynamic resource scheduling algorithm of the present invention; DETAILED DESCRIPTION
[0080] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.
[0081] The technical solution of the present invention to solve the above technical problems is:
[0082] The collaborative edge server deployment and resource scheduling strategy includes the following steps:
[0083] S1, based on the Shanghai Telecom base station data set, obtain the location of the base station and construct the base station set B = {B1, B2, ..., B M} and the edge server set S = {S1,S2,...,S N}(M≥N). The topological structure of the base station and edge server is represented as an undirected graph G=(V, E), where V=BU S represents the set of base stations and edge servers. Assume that the edge server is located at the same location as the base station, so it has a total of M possible deployment locations. Obtain the parameters of the base station and edge server, including the task arrival rate of the base station The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and build a cost and energy minimization model based on these parameters;
[0084] S2, obtain the initial solution, first establish a mapping relationship, randomly select a base station to deploy an edge server, if the base station and the edge server are in the same location, only map the edge server, otherwise map two edge servers until all base stations are allocated; then adjust the existing mapping according to the capacity and delay constraints, when the total task arrival rate of the edge server is greater than its service capacity, remap some of the base stations linked to it to the edge server, if no suitable edge server is found, deploy a new edge server on the base station until all base stations are allocated;
[0085] S3, improves the fireworks algorithm based on loser elimination and applies it to solving the cost and energy minimization problem;
[0086] S4, based on the edge server deployment results, determines the edge server deployment location. In a mobile edge computing environment, users are mobile, which will cause the total task arrival rate of the edge server to exceed its service capacity, so resource scheduling is required. This solution requires obtaining the parameters of the base station and edge server, including the task arrival rate of the base station that participated in the previous calculation. Current task arrival rate The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and build the expected response time minimization model based on these parameters;
[0087] S5, the hybrid algorithm is based on the bare bones fireworks algorithm, and the iterative process uses the crossover mutation mechanism of the genetic algorithm instead, and is applied to solve the problem of minimizing the expected response time;
[0088] Preferably, the step S1 includes:
[0089] (1) Using the m / m / 1 / n queue model to describe the edge server S n The task processing process. When the base station and the edge server are not in the same location, the base station will map two servers. We define the edge server closer to the base station as the near-end server (NEES) and the other as the far-end server (FEES). Secondly, when the base station and the edge server are in the same location, the edge server mapped by the base station is also considered to be a near-end server. When the base station is connected to two edge servers, it first considers offloading tasks to the near-end edge server. When there is no remaining space in the cache queue of the near-end edge server, it will offload tasks to the far-end edge server. Therefore, the edge server S n The total task arrival rate and the probability α that the cache queue is full n The calculation formula is:
[0090]
[0091]
[0092]
[0093] in, and Binary decision variables, for When base station B m The near-end edge server connected is S n It is set to 1 when , otherwise it is set to 0; When base station B m The remote edge server connected is S n It is set to 1 when , otherwise it is set to 0.
[0094] (2) Deployment cost and energy consumption model: The energy consumption of the edge server consists of two parts: runtime energy consumption and idle energy consumption. The formula is:
[0095]
[0096] P(u(t))=KP max +(1-K)P max *u(t) (5)
[0097]
[0098] Where K represents the percentage of idle state of edge servers to full load state, and u represents CPU utilization. The deployment cost of edge servers is positively correlated with the number of deployed servers. Therefore, the deployment cost of edge servers can be expressed as:
[0099]
[0100] Among them, x m Indicates whether in B m Deploy edge servers at m Set to 1 to indicate that B is deployed m Deploy edge servers at B, m is set to 0. The z-score normalization method is used to normalize energy consumption and cost respectively.
[0101] (3) Constraints; Mapping constraints: When a base station and an edge server are located at the same location, it can only be mapped to the edge server, and the edge server is the proximal edge server of the base station; otherwise, the base station needs to connect to a proximal edge server and a distal edge server. The formula is:
[0102]
[0103]
[0104] Deployment location constraint: Each edge server is co-located with a base station regardless of how the base stations are mapped. The base station set S mapped to the edge server n ∈S ε , then the formula is:
[0105]
[0106]
[0107] Delay constraint: Each edge server has a fixed coverage range. When the Euclidean distance between the base station and the edge server exceeds this threshold, the base station's tasks cannot be offloaded to the edge server for execution. Let R cover represents the coverage of the edge server, then the formula is:
[0108]
[0109]
[0110] Constraint on the number of base stations connected to the server: The number of base stations connected to each edge server cannot exceed the threshold. When there are too many base stations mapped to the same edge server, the number of tasks arriving at the edge server at the same time will increase, increasing the task response time. Let R ln represents the maximum number of base stations that the edge server can serve simultaneously, and the formula is:
[0111]
[0112] Capacity constraint: The arrival rate of tasks on each edge server cannot exceed the rate at which the edge server processes tasks. Represents the edge server S j The total task arrival rate, constant p represents the execution capability of the edge server, then the formula is:
[0113]
[0114] Preferably, the application of the fireworks algorithm based on loser elimination in step S3 includes:
[0115] (1) Initialize each parameter ( β, C a , C r ,γ,τ), where is the number of fireworks, is the total number of sparks, β is the parameter controlling the spark distribution, C a is the magnification factor, C r is the shrinkage coefficient, γ is the number of losers in each iteration, and τ is the maximum number of iterations;
[0116] (2) Obtain the initial deployment plan for each firework according to step S2. Fireworks contain four data: the number of edge servers, overall energy consumption, edge server location, and mapping plan. Calculate the cost and energy consumption z-score standardized values according to the deployment plan as the fitness value;
[0117] (3) According to the formula Calculate the number of sparks generated by each firework, where rank is the fitness value ranking of the firework. Calculate the explosion range of each firework, that is, the number of edge servers that need to be modified when each firework generates sparks. The formula is:
[0118]
[0119] Among them, Am g is half the number of edge servers in the g-th generation of Fireworks, f(X g ) is the fitness value of the g-th generation of fireworks;
[0120] (4) Randomly delete some deployed edge servers according to the explosion magnitude, and then redeploy some servers according to the mapping, delay and capacity requirements to meet the base station needs;
[0121] (5) Update the fireworks and reinitialize the eliminated fireworks. The algorithm ends when the maximum number of iterations is reached. Otherwise, the number of iterations increases by 1 and executes 3.
[0122] Preferably, the expected response time model in step S4 includes:
[0123] (1) The expected response time of the base station is composed of communication delay and task execution delay. Secondly, since each base station in this paper is connected to two edge servers, and we calculate the task response time based on a single task of each base station, considering that the greater the probability of the base station offloading tasks to the edge server, the more resources the base station obtains from the edge server, and the task response time generated between the base station and the edge server has a greater impact on the expected task response time of the base station, therefore, the expected response time formula of the system is:
[0124]
[0125]
[0126]
[0127]
[0128] Among them, d(B m ,S n ) represents base station B m and edge server S n The Euclidean distance between and a non-negative variable D m Represent the electromagnetic wave propagation rate and base station B respectively m The average amount of task data, Indicates base station B m To the edge server S ncommunication capacity, Indicates the edge server S per unit time n The average number of tasks in the edge server S, the constant μ represents the average number of tasks in the edge server S per unit time. n The number of tasks that can be completed.
[0129] (2) Constraints. Mapping constraint: ensure that the base station at the same location as the edge server maps one edge server, and other base stations map two edge servers. Deployment location constraint: ensure that each edge server is accurately deployed on the base station. Capacity constraint: ensure that the total task arrival rate of each edge server is less than its computing capacity. The formula is:
[0130]
[0131]
[0132]
[0133]
[0134]
[0135] Preferably, the application of the hybrid algorithm in step S5 includes:
[0136] (1) Initialize each parameter ( R mut ,τ′), where is the total number of sparks, R mut is the probability of mutation, τ′ is the maximum number of iterations;
[0137] (2) Regularly monitor whether there is any fluctuation in the task arrival rate of the base station. If not, keep the original mapping unchanged; otherwise, call the dynamic resource scheduling algorithm to adjust the mapping;
[0138] (3) According to the formula Calculate the expected task arrival rate of each edge server. When the resources allocated to the edge server exceed λ avg It will not allocate resources to any base station, where is the total task arrival rate of the base station during runtime;
[0139] (4) Obtain the initial mapping plan, randomly select unassigned base stations to map to edge servers, give priority to edge servers with more remaining capacity during mapping, and consider whether there are edge servers with a total task arrival rate exceeding λ after the allocation is completed avg ; until all base stations are assigned. The initial mapping scheme is executed times, the best solution is the firework and the rest are sparks.
[0140] (5) Perform a cross-mutation process on each spark; for the mapping scheme of each base station in the spark, if a mutation operation occurs, the mapping scheme is deleted; otherwise, a cross-operation occurs, and the base station has a 50% probability of selecting the mapping scheme of the corresponding base station in the fireworks, and the other 50% probability of keeping the mapping scheme unchanged.
[0141] (6) Find and remap base stations that violate mapping and capacity constraints,
[0142] (7) Update the fireworks. If the maximum number of iterations is reached, the algorithm ends. Otherwise, the number of iterations increases by 1 and the algorithm executes step 5.
[0143] Specifically, the present invention discloses a collaborative edge server deployment and resource scheduling strategy, such as Figure 1 As shown, the following steps are included:
[0144] Step 1: Based on the Shanghai Telecom base station data set, obtain the location of the base station, obtain the location information of all base stations, and obtain the parameters of the base station and edge server, including the number of base stations M and the task arrival rate of the base station The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and build a cost and energy minimization model based on these parameters.
[0145] In this implementation scheme, the above acquired data information is used to establish a mixed integer linear programming model to obtain a reasonable edge server deployment plan.
[0146] Step 2: Get the initial solution. First, establish a mapping relationship. If the base station and the edge server are in the same location, only map the edge server. Otherwise, map two edge servers. Then adjust the existing mapping according to the capacity and delay constraints. When the total task arrival rate of the edge server is greater than its service rate, remap some of the base stations linked to it to the edge server. If no suitable edge server is found, deploy a new edge server on the base station.
[0147] In this implementation, the mapping process is separated from the load distribution process, and the edge server closest to the base station is given priority during mapping. This deployment solution has lower costs and energy consumption.
[0148] Step 3: Improve the fireworks algorithm based on loser elimination and apply it to solve the cost and energy consumption minimization problem.
[0149] In this implementation, according to the initial deployment plan obtained in the second step, the standardized values of the cost and energy consumption z-score are calculated as the fitness values, the number of sparks and the explosion range generated by each firework are calculated, and some deployed edge servers are randomly deleted according to the explosion amplitude. Then, some servers are redeployed according to the mapping, delay and capacity requirements to meet the needs of the base station. When remapping, the edge server closest to the base station is given priority, the fireworks are updated and the eliminated fireworks are reinitialized. The algorithm ends when the maximum number of iterations is reached, otherwise the number of iterations is increased by 1 and the optimization continues.
[0150] Step 4: According to the deployment results of the edge server, the deployment location of the edge server is determined. In order to solve the problem that the total task arrival rate of the edge server exceeds the service rate caused by user mobility, it is necessary to obtain the parameters of the base station and the edge server, including the task arrival rate of the base station that participated in the calculation last time. Current task arrival rate The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and a model minimizing the expected response time is constructed based on these parameters.
[0151] In this embodiment, the data information obtained above is used to establish a mixed integer linear programming model, the optimization target is the expected response time, the constraints include mapping constraints, deployment location constraints and capacity constraints, and a reasonable dynamic resource scheduling solution is obtained.
[0152] Step 5: Based on the bare-bone fireworks algorithm, the hybrid iterative process uses the crossover mutation mechanism of the genetic algorithm and is applied to solving the problem of minimizing the expected response time.
[0153] In this implementation, it is regularly monitored whether there is any fluctuation in the task arrival rate of the base station. If not, the original mapping is kept unchanged. Otherwise, the dynamic resource scheduling algorithm needs to be called to adjust the mapping. The dynamic resource scheduling process first calculates the expected task arrival rate λ of each edge server avg , when the resources allocated to the edge server exceed λ avg , it will not allocate resources to any base station; secondly, obtain the initial mapping scheme, randomly select unassigned base stations to map to edge servers, give priority to edge servers with more remaining capacity when mapping, and consider whether there are edge servers with a total task arrival rate exceeding λ after the allocation is completed. avg , until all base stations are assigned; the initial mapping scheme is executed times, where the best solution is used as fireworks and the rest are sparks; a crossover mutation process is performed on each spark; for the mapping scheme of each base station in the spark, if a mutation operation occurs, the mapping scheme is deleted; otherwise a crossover operation occurs, and the base station has a 50% probability of selecting the mapping scheme of the corresponding base station in the fireworks, and the other 50% probability of keeping the mapping scheme unchanged; find and remap base stations that violate the mapping and capacity constraints, update fireworks, and the algorithm ends when the maximum number of iterations is reached, otherwise the number of iterations is increased by 1 and the optimization continues.
[0154] The present invention is further described below using a specific example:
[0155] The data set selected is the 2014 Shanghai Telecom base station data set. The base station data obtained includes the latitude and longitude of the base station, the user, and the start time and end time of the service for the user.
[0156] Since this paper studies the deployment and resource scheduling scheme of collaborative edge servers, the location information of 1,500 base stations is selected. It is assumed that the coverage range of the base stations is 10 km. Other information is randomly selected within a reasonable range according to the mobile edge computing environment.
[0157] Assume that the task arrival rate of each base station is [4×10 6 ,6×10 8 ], the average task size of the base station is 60Kb, the communication capacity of the link is [100,1000], and the propagation speed of electromagnetic waves is 2×10 5 km / s.
[0158] We chose Inspur NP3020 as the edge server, which is equipped with 4 cores, each with an operating frequency of 3.4GHz. The cache queue length of the server is 4, the power in full load state is 300W, and the idle power of the edge server accounts for 60% of the power in full load state. Assuming that the task execution time of the edge server follows a normal distribution, with an average of 20ms and a variance of 5, μ is 0.05. Based on the average task volume of the base station and the computing power of the server, we set the maximum number of base stations served by the edge server to 30.
[0159] Initialize the parameters of the loser elimination fireworks algorithm ( β, C a , C r ,γ,τ), assuming parameter values β=0.75,C a =1.2, C r =0.9,γ=2,τ=100.
[0160] Randomly select base stations to deploy edge servers, establish a mapping between edge servers and base stations, and then allocate base station loads according to the mapping to obtain an initial deployment plan (fireworks) that meets the mapping, delay, and capacity requirements. Calculate the sparks and explosion range generated by each firework, and adjust the number of deployed edge servers and the mapping relationship according to the explosion range to obtain a new solution (spark). Update the fireworks according to the sparks generated by each firework, and reinitialize the two fireworks with the worst fitness values until the maximum number of iterations is reached and the algorithm ends. Otherwise, the number of iterations is increased by 1, and sparks are continued to be generated for iterative optimization.
[0161] At this time, a total of 153 edge servers are deployed, with an overall energy consumption of 43246 kWh, while the deployment scheme of the benchmark algorithm requires the deployment of 192 edge servers, with an overall energy consumption of 50086 kWh. Obviously, the present invention effectively reduces the number of deployed edge servers and energy consumption, and correspondingly reduces the deployment cost.
[0162] Initialize the parameters of the hybrid algorithm ( R mut ,τ′), assuming the parameter value R mut =0.05,τ′=20.
[0163] Monitor whether the base station task arrival rate has changed. If not, keep the original scheduling strategy unchanged; if it has changed, call the dynamic resource scheduling algorithm. Calculate the expected total task arrival rate λ of each edge server avg Randomly select a base station to map to the edge server with the highest remaining capacity, distribute the base station load, and determine whether the total task arrival rate of the edge server exceeds λ avg If it exceeds, the edge server will not map the base station, and the process will be repeated until all base stations are allocated. The above process obtains the initial resource scheduling plan, of which the optimal solution is fireworks and the rest are sparks. The crossover mutation process is performed on the sparks, and the optimal solution is greedily selected as the fireworks until the maximum number of iterations is reached and the algorithm ends. Otherwise, the number of iterations is increased by 1, and the spark iterative optimization continues.
[0164] At this time, the expected task response time is 60.10ms, while the experimental result of continuing to use the fireworks algorithm based on loser elimination is 62.81ms, and the experimental result of using the traditional genetic algorithm is 60.25ms. Obviously, the present invention effectively reduces the expected response time.
[0165] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
[0166] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0167] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0168] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.
Claims
1. A collaborative edge server deployment and resource scheduling method, characterized in that: The following steps are involved: Based on the access latency and the capacity of edge servers, a cost and energy consumption optimization model for collaborative edge server deployment is established; Obtain the initial solution of the cost and energy consumption optimization model, and map the base station to multiple edge servers; adjust the existing mapping according to the capacity and delay constraints, and obtain the first optimized deployment plan of the edge server. The parameters used in the deployment plan include: the task arrival rate of the base station The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations connected to the edge server R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ); The improved fireworks algorithm based on loser elimination is adopted to adjust the number of deployed edge servers and the mapping relationship to obtain the second optimal deployment plan of edge servers. According to the edge server deployment results, the expected response time minimization model of dynamic resource scheduling is established. Using a hybrid algorithm that combines the bare-bone fireworks algorithm and the genetic algorithm, the mapping between edge servers and base stations is dynamically adjusted; The cost and energy consumption optimization model for collaborative edge server deployment is established based on the access delay and the capacity of the edge server, specifically including: Based on the base station data set, the base station locations are obtained and the base station set B = {B1, B2, ..., B M } and the edge server set S = {S1,S2,...,S N }(M≥N), the topological structure of the base station and edge server is represented as an undirected graph G=(V,E), V=BUS represents the set of base stations and edge servers. Assume that the edge server is located at the same location as the base station, so it has a total of M possible deployment locations; obtain the parameters of the base station and edge server, including the task arrival rate of the base station The rate p of edge server processing tasks, the length of cache queue len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and build a cost and energy minimization model based on these parameters; The deployment cost and energy consumption model specifically includes: The energy consumption of the edge server consists of two parts: runtime energy consumption and idle energy consumption. The formula is: P(u(t))=KP max +(1-K)P max *u(t) (2) Where Energy represents the overall energy consumption of the system, P(u(t)) represents the system power from time t1 to time t2, t1 represents the time when task offloading starts, t2 represents the time when all tasks are completed, n represents the nth edge server, and P max represents the power of the edge server under full load, u(t) represents the CPU utilization from time t1 to time t2, Represents the edge server S n The total task arrival rate of the edge server is p, p represents the rate at which the edge server processes tasks, and K represents the percentage of the idle state of the edge server to the fully loaded state. The deployment cost of the edge server is positively correlated with the number of deployed servers, so the deployment cost of the edge server can be expressed as: Among them, x m Indicates whether in B m Deploy edge servers at m Set to 1 to indicate that B is deployed m Deploy edge servers at B, m are set to 0, and the z-score normalization method is used to normalize energy consumption and cost respectively; The fireworks algorithm based on loser elimination is adopted to adjust the number of deployed edge servers and the mapping relationship to obtain the second optimized deployment plan of the edge servers, which specifically includes: Step 1.1: Initialize parameters in, is the number of fireworks, is the total number of sparks, β is the parameter controlling the spark distribution, C a is the magnification factor, C r is the shrinkage coefficient, γ is the number of losers in each iteration, and τ is the maximum number of iterations; Step 1.2, obtain the initial deployment plan of each firework, which contains four data: the number of edge servers, overall energy consumption, edge server location, and mapping plan. Calculate the cost and energy consumption z-score standardized values according to the deployment plan as the fitness value; Step 1.3: According to the formula Calculate the number of sparks produced by each firework; where SN rank Indicates the number of sparks produced by fireworks with a fitness value ranking of rank. Represents the total number of sparks produced by all fireworks, Indicates the number of fireworks, and rank is the fitness ranking of the firework. At the same time, the explosion range of each firework is calculated, that is, the number of edge servers that need to be modified when each firework produces sparks. The formula is: Among them, Am g is half the number of edge servers in the g-th generation Fireworks. represents the explosion range of the g-th generation fireworks, f(X g ) is the fitness value of the g-th generation of fireworks; Step 1.4: Randomly delete some deployed edge servers according to the explosion magnitude, and then redeploy some servers according to the mapping, delay and capacity requirements to meet the base station needs; Step 1.5, update the fireworks and reinitialize the eliminated fireworks. The algorithm ends when the maximum number of iterations is reached, otherwise the number of iterations increases by 1 and execute step 1.3; The hybrid algorithm combining the bare bones fireworks algorithm and the genetic algorithm specifically includes: Step 2.1: Initialize parameters in, is the total number of sparks, R mut is the probability of mutation, τ′ is the maximum number of iterations; Step 2.2, regularly monitor whether there is any fluctuation in the task arrival rate of the base station. If not, keep the original mapping unchanged. Otherwise, use a hybrid algorithm combining the bare-bone fireworks algorithm and the genetic algorithm to adjust the mapping; Step 2.3: Calculate the expected task arrival rate λ for each edge server avg , when the resources allocated to the edge server exceed λ avg It will not allocate resources to any base station, where is the total task arrival rate of the base station during runtime; Step 2.4: Get the initial mapping plan, randomly select unassigned base stations to map to edge servers, give priority to edge servers with more remaining capacity during mapping, and consider whether there are edge servers with a total task arrival rate exceeding λ after the allocation is completed. avg ; Until all base stations are assigned; the initial mapping scheme is executed times, the best solution is the firework, and the rest are sparks; Step 2.5, perform the cross-mutation process for each spark; for the mapping scheme of each base station in the spark, if a mutation operation occurs, the mapping scheme is deleted; otherwise, a cross-operation occurs, and the base station has a 50% probability of selecting the mapping scheme of the corresponding base station in the fireworks, and the other 50% probability of keeping the mapping scheme unchanged; Step 2.6, find and remap base stations that violate mapping and capacity constraints; Step 2.7: Update the fireworks. If the maximum number of iterations is reached, the algorithm ends. Otherwise, the number of iterations increases by 1 and step 2.5 is executed.
2. A collaborative edge server deployment and resource scheduling method according to claim 1, characterized in that: The initial solution of the cost and energy consumption optimization model is obtained, specifically including: First, a mapping relationship is established. If the base station and the edge server are in the same location, only the edge server is mapped. Otherwise, two edge servers need to be mapped. Then, the existing mapping is adjusted according to the capacity and delay constraints, and the base station task arrival rate is allocated to the edge server to obtain the initial deployment plan for the first optimization.
3. A collaborative edge server deployment and resource scheduling method according to claim 2, characterized in that: The constraints of the deployment cost and energy consumption model specifically include: Mapping constraint: When a base station and an edge server are located at the same location, it can only be mapped to the edge server, and the edge server is the proximal edge server of the base station; otherwise, the base station needs to connect to a proximal edge server and a distal edge server, and the formula is: The binary variable Indicates base station B m Whether to connect to the near-end edge server S n , a value of 1 indicates connection, and a value of 0 indicates no connection; Indicates base station B m Whether to connect to the remote edge server S n , a value of 1 indicates connection, and a value of 0 indicates disconnection; B m represents the mth base station, B represents the set of all base stations, and S represents the set of all edge servers; Deployment location constraints: Regardless of how the base stations are mapped, each edge server is located at the same location as a base station, allowing the base station cluster to Mapped to edge server S n ∈S ε , then the formula is: in represents a set of base stations, and these base stations are co-located with the edge server, S ε represents a set of edge servers with the locations of deployed base stations, S n represents the nth edge server; Delay constraint: Each edge server has a fixed coverage range. When the Euclidean distance between the base station and the edge server exceeds this threshold, the base station's tasks cannot be offloaded to the edge server for execution. Let R cover represents the coverage of the edge server, then the formula is: Constraint on the number of base stations connected to the server: The number of base stations connected to each edge server cannot exceed the threshold. When there are too many base stations mapped to the same edge server, the number of tasks arriving at the edge server at the same time will increase, which will increase the task response time. The formula is: Where R ln Indicates the upper limit of the number of base stations that the edge server can connect to. When this limit is exceeded, the service quality of the edge server will be greatly reduced; Capacity constraint: The arrival rate of tasks on each edge server cannot exceed the rate at which the edge server processes tasks. Represents the edge server S j The total task arrival rate, constant p represents the execution capability of the edge server, then the formula is:
4. A collaborative edge server deployment and resource scheduling method according to claim 1, characterized in that: The method of establishing a model for minimizing the expected response time of dynamic resource scheduling according to the edge server deployment result specifically includes: Get the parameters of the base station and edge server, including the arrival rate of the task that the base station participated in the previous calculation Current task arrival rate The computing power of the edge server p, the cache queue length len, and the maximum number of base stations served at the same time R ln , full load power consumption P max , coverage R cover , the distance between the base station and the edge server d(B m ,S n ), and based on these parameters, a model for minimizing the expected response time is constructed.
5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the collaborative edge server deployment and resource scheduling method according to any one of claims 1 to 4 is implemented.
6. A scheduling system comprising: Cost and energy consumption optimization model establishment module: first optimization module: second optimization module: used to execute the method described in any one of claims 1-4.
