A reliability-enhanced service deployment method for revenue optimization in edge computing

Abstract

The present invention proposes a reliability-enhanced service deployment method for benefit optimization in edge computing. First, the user's accessibility to the edge server is analyzed according to the user's maximum tolerance for service delay; then, the reliability of the user's access to the service is evaluated according to the availability of the accessible edge server, and the service reliability benefit of the application provider is defined by the number of users who meet the service reliability requirements; finally, the deployment cost and service reliability benefit are coordinated to model the service deployment benefit, and a benefit-maximizing edge service deployment model based on reliability assurance is constructed. The problem to be solved is reduced by using the classic non-monotone submodular function maximization problem under multiple knapsack constraints. Based on the characteristics of the submodular problem, an efficient and approximately optimal edge service deployment solution is proposed to improve the service deployment benefit and reduce the impact of potential failures on service availability.

Description

Technical Field

[0001] The present invention relates to the field of mobile edge computing technology, and in particular to a reliability enhancement service deployment method for revenue optimization in edge computing. Background Art

[0002] Compared with cloud servers, edge servers have limited computing, storage, and communication resources, making resource provisioning costly. In addition, each edge server can cover a small geographic area, depending on the collocated wireless infrastructure. Therefore, multiple service instances are required to provide services to geographically distributed users within a certain area. In addition, due to the limited coverage, resource constraints, and geographical dispersion of edge servers, potential failures, insufficient resources, or other uncertainties can easily lead to service interruptions or performance degradation, directly affecting the user's service experience. This makes the reliability of edge services critical, not only related to user experience, but also to the competitiveness and long-term sustainability of application service providers in a highly competitive market. Finally, due to resource and budget constraints, edge servers can usually only deploy a limited number of service instances, which requires the establishment of a reliable and high-yield edge service deployment method to reduce the impact of edge server failures on service availability and optimize the service deployment benefits, considering the high competition for limited edge resources for the deployment of multiple types of services and the heterogeneous service reliability and latency requirements of users. Unfortunately, previous studies have mainly focused on maximizing user coverage, minimizing service latency, or optimizing user-perceived service quality, but have failed to consider service reliability. As a result, service providers may not be able to provide continuous and reliable services to their users, thereby reducing service quality and ultimately harming the service provider's revenue. Summary of the invention

[0003] The present invention aims at the random failure of edge servers in edge computing environments, which may cause service failures and thus affect the service experience of users within the coverage area. Considering the limited heterogeneity of edge server resources, the service reliability and latency requirements of heterogeneous users, and the scenario of service deployment budget constraints, the deployment decisions of different types of services are optimized to maximize their overall benefits and reduce the impact of potential failures on service availability.

[0004] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0005] Step S1: Consider a service deployment scenario with multiple edge servers and different types of services, and build a quantitative representation model for service deployment, edge server accessibility, and service reliability benefits.

[0006] Step S2: Considering the limited service deployment budget, heterogeneous user service delay and reliability requirements, and heterogeneous and limited edge server resources, the service reliability benefits and deployment costs are coordinated to build a service deployment benefit maximization problem model based on reliability assurance.

[0007] Step S3: According to the characteristics of the problem, the service deployment benefit maximization problem based on reliability assurance is transformed into a set function optimization problem. The problem is analyzed to be a non-monotone submodular maximization problem, and the problem is reduced based on the non-monotone submodular function maximization problem under multiple knapsack constraints.

[0008] Step S4: Use an approximate optimization algorithm to solve and obtain an edge service deployment solution with theoretical performance guarantees.

[0009] Further, in step S1, consider a mobile computing system consisting of a set of n distributed edge servers Each edge server All have heterogeneous resources. Specifically, their computing resources Can be used to perform services on demand. Its storage resources Instances that can be used for hosting services. To be deployed on the system edge server Generally speaking, each service Both with deployment budget k , indicating the maximum cost of deploying its instance. Deploy service a on edge servers k The required computing resources and storage resources are respectively and express.

[0010] Given the total number of users requesting applications, m, it is expressed as Assume that each user Request only one by a j denoted by the service instance. Note that a user who needs multiple services can actually be converted into multiple users. j The maximum tolerable service delay is expressed as And u j The service reliability requirement is expressed as

[0011] Service deployment model:

[0012] At this point, the service deployment decision variable set is expressed as in

[0013]

[0014] Given an original service, multiple instances are usually deployed to provide services to users. Therefore, for each service a k , and its overall deployment cost is where c ik is in i Deploy a k The deployment cost of a k The total cost incurred by the instance should not exceed the given deployment budget b k ,Right now

[0015]

[0016] When deploying multiple services on the same edge server, the following resource constraints should be met. The first is the computing resource cr s Constraints, the second is storage resources sr s constraint

[0017]

[0018] Service accessibility model:

[0019] L ij Defined as user u j From the edge server i The perceived latency when receiving a service. Let acx ij Represents user u j Is it possible to access the edge servers? i Service instance on ij =1) or not (acc ij =0). Given user u j The maximum acceptable delay Service accessibility acc ij It can be expressed as follows.

[0020]

[0021] Quantitative representation model of service reliability benefits:

[0022] Given a deployment strategy User j In request a k The perceived service reliability during service (denoted as Rel j ) can be calculated as:

[0023]

[0024] Where Acc(u j ) is user u j The set of edge servers that can be accessed, i.e. r i Represents edge servers i The probability of availability, value 1-d ik r i Indicates u j From the edge servers i The probability of failure when retrieving a service. Based on this, Measure the probability u that all deployed instances are unavailable j Therefore, Rel j is u j Perceived service reliability, which is affected by R j ∈[0,1].

[0025] make Represents user u j service reliability requirements. In order to successfully serve u j ,Rel j The value should be no less than The value of Using a set of binary variables To indicate which users can be successfully served, where x j =1 indicates user u j The reliability requirement is met, that is, Otherwise x j =0 (i.e. Therefore, for users Can get

[0026]

[0027] The service reliability benefit of a service provider is defined by the number of users that meet the service reliability requirements (i.e., the number of users served). in Indicates request service a k The user set, that is

[0028] Furthermore, in step S2, a reliability-guaranteed edge service deployment benefit maximization problem model is constructed for the limited service deployment budget, different user requirements, and heterogeneous edge server resources. Indicates deployment of service a on the edge server k The deployment benefit obtained can be expressed as

[0029]

[0030] Where term1 represents the service reliability benefit, which is determined by the number of users that meet the service reliability requirements. Term2 represents the deployment benefit determined by the difference between the deployment budget and the cost incurred (i.e., the saved budget). It is a non-negative value.

[0031] The purpose of adding term2 is to avoid deploying too many instances while meeting the user's service reliability requirements. (i.e. x j =1), continuing to deploy more instances will not increase reliability benefits, but will only increase deployment costs.

[0032] The parameter γ>0 is an empirically chosen coefficient used to normalize the number of users served by the service deployment cost.

[0033] Generally speaking, a smaller γ means that term2 contributes more to the overall revenue, which can better control costs and ensure the profitability of the service provider. On the contrary, a larger γ will give priority to service reliability. Note that is a non-negative value. For any decision variable d ik =0 setting

[0034] At this point, considering the server resource constraints, the service reliability and latency requirements of heterogeneous users, and the service deployment budget constraints, the edge service deployment benefit maximization model based on reliability assurance can be constructed as follows:

[0035]

[0036] Furthermore, in step S3, problem P1 is transformed according to its characteristics.

[0037] In problem P1, x j and d ik are binary variables, subject to {0,1} and coupled to each other, i.e., d ik The value of x j When x j = 1, there is at least one d ik (a j =a k ) is 1 to ensure that user u is satisfied j In fact, it is not difficult to find that if d ik The value is inappropriate, x j The value of is always 0, which means that it is impossible to provide users with satisfactory service reliability. Further transform the problem. Note that the variable Rel j It is continuous, with a value between 0 and 1, that is, Rel j ∈[0,1]. jConvert to Rel j It not only simplifies the problem-solving process, but also makes it more accurate. Solve the problem well. Reflect the change of decision variables d ik The service reliability gain when the value of is. It is worth noting that Rel j is monotonically increasing. Therefore, when Rel j The value increases to no less than When it is mapped to x j =1, otherwise x j =0.

[0038] Then, by expanding the Rel given in Eq. j The service reliability benefit (i.e., Λ j ):

[0039]

[0040] It is easy to see that due to Rel j ∈[0,1], The value of can be 0 or 1. Therefore, Λ j ∈{Rel j ,1}, which is also a continuous variable. In this case, Λ j The value of will never increase by deploying more service instances. Otherwise, j =Rel j ≤1.

[0041] Transform the discrete variable x in the equation j The continuous form Λ j . Then reconstruct the statement of benefits as follows:

[0042]

[0043] At this point, the problem is reframed as:

[0044]

[0045] Note that by After assigning values ​​so that all constraints are satisfied, a feasible solution to problem P2 can be found. Intuitively, each d in problem P2 ik = Only if at least one user The service reliability requirement is valid only when the variable x in problem (P1) j More specifically, in problem (P1), if x j =1, then user u j have Therefore Λ j= 1. In this case, there is at least one decision variable d in the problem (P2) ik is set to 1. Therefore, if Λ j <1, it is equivalent to setting x j = 0. Therefore, problem (P2) is equivalent to problem (P1).

[0046] Problem (P2) can be reformulated as a set function optimization problem. represents a set of potential edge server-service pairs, where each pair (s i ,a k )∈ is mapped to the deployment decision variable d ik . The set of edge server-service deployment pairs for problem P2 is represented as Each pair obey therefore, Expressed as

[0047] For the feasible set definition As a service k A set of edge server-service deployment pairs, such that

[0048] Given Types of services include and For any a k ≠a h . Then you can get:

[0049]

[0050] make To request an application k A group of users who can access I k At least one instance of Then, the return can be rewritten as follows

[0051]

[0052] Given a set of services to be deployed on edge servers Will Defined as the total revenue generated by the set of edge server-service deployment pairs The expression is as follows

[0053]

[0054] At this point, problem P2 is constructed as

[0055]

[0056]

[0057] Further, in step S4, an approximate algorithm is used to solve P3 and obtain a service deployment solution with performance guarantee. The key idea of ​​the approximate algorithm is that, under the constraints of deployment budget and edge server resources, it iteratively selects a candidate edge server with the maximum marginal benefit from all potential pairs in Ω to determine a new service deployment.

[0058] make is the deployment pair of the edge server-service pair candidate set problem P2. Definition Indicates Adding an Edge Server-Service Pair The marginal benefit generated when . The calculation is as follows:

[0059]

[0060] The update rules are:

[0061]

[0062] Thus, any j =1 candidate deployment pair will be selected as the final deployment pair and included in the final solution set middle.

[0063] The specific process of the algorithm is as follows

[0064] Step S4-1: Starting from initialization, the algorithm creates two empty sets and Save candidate service deployment solutions and final solutions separately. In addition, initialize each application solver a k Potential edge server-service pairs Ω k and candidate configuration decision variables d ik , d ik Set to 0 and mapped to Ω k Each pair in .

[0065] Step S4-2: Initialize an empty set The edge server-service pair with the maximum marginal benefit is used to store each service.

[0066] Then, it calculates for each service a k The marginal value of all valid pairs of Among them, effective (s i ,a k ) must meet the budget b k and edge serversi Constraints on available resources.

[0067] Step S4-3: Next, for each a k , find the deployment pair with the maximum marginal value (s i′ ,a k ) and add the pair to the collection If no valid pair is found, a k Will be from Delete it.

[0068] Step S4-4: Next, the algorithm starts from Select the pair with the largest marginal value as a candidate deployment pair and add it to At the same time, the deployment decision variables of the candidate pairs Updated to 1.

[0069] Step S4-5: After that, it updates the service The service reliability benefit of the served users. Given a candidate solution and a pair for For each user u j , if its reliability requirements are met, the user's service reliability benefit is updated to 1, that is, Λ′ j = 1. This ensures that deploying more instances will not continue to increase its service reliability advantage.

[0070] Step S4-6: Then, of Each pair in contains And merge into a set Store the final deployment plan

[0071] Step S4-7: Repeat the above process until there are no services to be deployed.

[0072] Step S4-8: Finally, return to the collection As the final service deployment solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic diagram of an edge computing network in an example of the present invention;

[0074] Figure 2 It is a specific flow chart in an example of the present invention;

[0075] Figure 3 Schematic diagram of the approximate algorithm flow in the example of the present invention. DETAILED DESCRIPTION

[0076] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0077] Step S1: Consider a service deployment scenario with multiple edge servers and different types of services, and build a quantitative representation model for service deployment, edge server accessibility, and service reliability benefits.

[0078] like Figure 1 As shown in , consider how to deploy various types of services on edge servers with limited resources to meet the service needs of users. Consider a mobile computing system consisting of a group of n distributed edge servers For each edge server Its computing resources are Its storage resources are make To be deployed on the system edge server Each service Both with deployment budget k , indicating the maximum cost of deploying its instance. Deploy service a on edge servers k The required computing resources and storage resources are respectively and express.

[0079] Given the total number of users requesting applications, m, it is expressed as Assume that each user Request only one by a j Represents a service instance. j The maximum tolerable service delay is expressed as And u j The service reliability requirement is expressed as At this point, the service deployment decision variable set is expressed as in

[0080]

[0081] Given an original service, multiple instances are usually deployed to provide services to users. Therefore, for each service a k , and its overall deployment cost is where c ik is in i Deploy a k The deployment cost of a k The total cost incurred by the instance should not exceed the given deployment budget b k ,Right now

[0082]

[0083] When deploying multiple services on the same edge server, the following resource constraints should be met. The first is the computing resource cr s Constraints, the second is storage resources sr s constraint

[0084]

[0085] Let L ij Defined as user u j From the edge server i The perceived latency in receiving a service. Let acc ij Represents user u j Is it possible to access the edge servers? i Service instance on ij =1) or not (acc ij =0). Given user u j The maximum acceptable delay Service accessibility acc ij It can be expressed as follows.

[0086]

[0087] Given a deployment strategy User j In request a k The perceived service reliability during service (denoted as Rel j ) can be calculated as:

[0088]

[0089] Where Acc(u j ) is user u j The set of edge servers that can be accessed, i.e. r i Represents edge servers i The probability of availability, value 1-d ik r i Indicates u j From the edge servers i The probability of failure when retrieving a service. Based on this, Measure the probability u that all deployed instances are unavailable j Therefore, Rel j is u j Perceived service reliability, which is affected by R j ∈[0,1].

[0090] make Represents user u j service reliability requirements. In order to successfully serve uj ,Rel j The value should be no less than The value of Using a set of binary variables To indicate which users can be successfully served, where x j =1 indicates user u j The reliability requirement is met, that is, Otherwise x j =0 (i.e. Therefore, for users Can get

[0091]

[0092] The service reliability benefit of a service provider is defined by the number of users that meet the service reliability requirements (i.e., the number of users served). in Indicates request service a k The user set, that is

[0093] Step S2: Considering the limited service deployment budget, heterogeneous user service delay and reliability requirements, and heterogeneous and limited edge server resources, the service reliability benefits and deployment costs are coordinated to build a service deployment benefit maximization problem model based on reliability assurance.

[0094] make Indicates deployment of service a on the edge server k The deployment benefit obtained can be expressed as

[0095]

[0096] Where term1 represents the service reliability benefit, which is determined by the number of users that meet the service reliability requirement. Term2 represents the deployment benefit determined by the difference between the deployment budget and the cost incurred (i.e., the budget saved). It is a non-negative value. The parameter γ>0 is an empirically selected coefficient used to normalize the number of users served by the service deployment cost. For any decision variable d ik =0 setting

[0097] At this point, considering server resource constraints, user reliability, resource and latency requirements, and service provider budget, the revenue maximization problem can be formulated as

[0098]

[0099] Step S3: According to the characteristics of the problem, it is converted into a set function optimization problem. The problem is analyzed to be actually a non-monotone submodular maximization problem, and the problem is reduced based on the non-monotone submodular function maximization problem under multiple knapsack constraints.

[0100] In problem P1, x j and d ik are binary variables, subject to {0,1} and coupled to each other, i.e., d ik The value of x j When x j = 1, there is at least one d ik (a j =a k ) is 1 to ensure that user u is satisfied j In fact, it is not difficult to find that if d ik The value is inappropriate, x j The value of is always 0, which means that it is impossible to provide users with satisfactory service reliability. Further transform the problem. Note that the variable Rel j It is continuous, with a value between 0 and 1, that is, Rel j ∈[0,1].

[0101] Then, by expanding the Rel given in Eq. j The service reliability benefit (i.e., Λ j ):

[0102]

[0103] It is easy to see that due to Rel j ∈[0,1], The value of can be 0 or 1. Therefore, Λ j ∈{Rel j ,1}, which is also a continuous variable. In this case, Λ j The value of will never increase by deploying more service instances. Otherwise, j =Rel j ≤1.

[0104] Transform the discrete variable x in the equation j The continuous form Λ j . Then reconstruct the statement of benefits as follows:

[0105]

[0106] At this point, the problem is reframed as:

[0107]

[0108] Note that by After assigning values ​​so that all constraints are satisfied, a feasible solution to problem P2 can be found.

[0109] Step S4: Use the approximate optimization algorithm to solve and obtain the edge service deployment solution with theoretical performance guarantee. The specific algorithm flow is as follows: Figure 3 shown.

[0110] make is the edge server-service candidate set deployment pair of problem P2. Definition Indicates Adding an Edge Server-Service Pair The marginal benefit generated when . The calculation is as follows:

[0111]

[0112] The update rules are:

[0113]

[0114] Thus, any j =1 candidate deployment pair will be selected as the final deployment pair and included in the final solution set middle.

[0115] The algorithm flow is as follows

[0116] Step S4-1: Starting from initialization, the algorithm creates two empty sets and Save candidate service deployment solutions and final solutions separately. Initialize each application solver a k Potential edge server-service pairs Ω k and candidate configuration decision variables d ik , d ik Set to 0 and mapped to Ω k Each pair in .

[0117] Step S4-2: Initialize an empty set The edge server-service pair with the maximum marginal benefit is used to store each service.

[0118] Then, it calculates for each service a k The marginal value of all valid pairs of Among them, effective (s i ,a k ) must meet the budget b k and edge servers i Constraints on available resources.

[0119] Step S4-3: Next, for each a k , find the deployment pair with the maximum marginal value (s i′ ,a k ) and add the pair to the collection If no valid pair is found, a k Will be from Delete it.

[0120] Step S4-4: Next, the algorithm starts from Select the pair with the largest marginal value as a candidate deployment pair and add it to At the same time, the deployment decision variables of the candidate pairs Updated to 1.

[0121] Step S4-5: After that, it updates the service The service reliability benefit of the served users. Given a candidate solution and a pair for For each user u j , if its reliability requirements are met, the user's service reliability benefit is updated to 1, that is, Λ′ j = 1. This ensures that deploying more instances will not continue to increase its service reliability advantage.

[0122] Step S4-6: Then, of Each pair in contains And merge into a set Store the final deployment plan

[0123] Step S4-7: Repeat the above process until there are no services to be deployed.

[0124] Step S4-8: Finally, return to the collection As the final service deployment solution.

[0125] The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed by the present invention should be included in the protection scope recorded in the claims.

Claims

1. A reliability enhancement service deployment method for revenue optimization in edge computing, characterized by: The method comprises the following steps: Step S1: Consider a scenario with multiple edge servers and multiple different service deployment scenarios, and build a quantitative representation model of service deployment, edge server accessibility, and service reliability benefits; Step S2: In view of the limited service deployment budget, the heterogeneous service delay and reliability requirements of users, and the heterogeneous and limited edge server resources, the service reliability benefits and deployment costs are coordinated to build a service deployment benefit maximization problem model based on reliability assurance; Step S3: according to the characteristics of the problem, convert it into a set function optimization problem, analyze that the problem is actually a non-monotone submodular maximization problem, and reduce the problem based on the non-monotone submodular function maximization problem under multiple knapsack constraints; Step S4: using an approximate optimization algorithm to solve and obtain an edge service deployment solution with theoretical performance guarantee; In step S1, the service deployment model is defined: Consider a mobile computing system consisting of a set of n distributed edge servers Each edge server Computing resources Can be used to execute services and store resources on demand Instances that can be used for hosting services, To be deployed on the system edge server services, each service The deployment budget is b k , deploy service a on the edge server k The required computing resources and storage resources are respectively and Indicates that, given the total number of users requesting an application, m, is expressed as Assume that each user Request only one by a j Represents a service instance, which will j The maximum tolerable service delay is expressed as And u j The service reliability requirement is expressed as The service deployment decision variable set is represented as in For each service a k , and its overall deployment cost is where c ik is in i Deploy a k The deployment cost of deploying a k The total cost incurred by the instance should not exceed the given deployment budget b k ,Right now When deploying multiple services on the same edge server, the following resource constraints should be met. The first is the computing resource cr s Constraints, the second is storage resources sr s constraint In step S1, the accessibility model of the edge server is defined: L ij Defined as user u j From the edge server i The perceived delay when receiving the service, let acc ij Represents user u j Is it possible to access the edge servers? i Service instance acc ij =1 or cannot acc ij = 0, given user u j The maximum acceptable delay Service accessibility acc ij It can be expressed as follows: In step S1, the service reliability benefit quantification model is defined: Given a deployment strategy User j In request a k The service reliability perceived during service, denoted as Rel j , can be calculated as: Where Acc(u j ) is user u j The set of edge servers that can be accessed, i.e. r i Represents the edge server S i Probability of availability; make Represents user u j service reliability requirements, in order to successfully serve u j ,Rel j The value should be no less than The value of Using a set of binary variables To indicate which users can be successfully served, where x j =1 indicates user u j The reliability requirement is met, that is, Otherwise x j =0 Therefore, for users You can get: The service reliability benefit of a service provider is defined by the number of users who meet the service reliability requirements. in Indicates request service a k The user set, that is In step S2, let Indicates deployment of service a on the edge server k The deployment benefit obtained can be expressed as: Where the parameter γ>0 is an empirically selected coefficient used to normalize the number of users served according to the service deployment cost, term1 represents the service reliability benefit, and term2 represents the deployment benefit determined by the difference between the deployment budget and the incurred cost. At this time, considering the server resource constraints, user reliability, resource and latency requirements, and the service provider budget, the maximization benefit problem can be constructed as: In step S3, let Rel j ∈[0,1], when Rel j The value increases to no less than When it is mapped to x j =1, otherwise x j = 0, by expanding the Rel given in Eq. j The expression is to reformulate the service reliability benefit, that is, Λ j : Transform the discrete variable x in the equation j The continuous form Λ j , and then reconstruct the statement of the return as follows: The parameter γ>0 is an empirically chosen coefficient used to normalize the number of users served according to the service deployment cost. At this point, the problem is restructured as:

2. The reliability enhancement service deployment method for benefit optimization in edge computing according to claim 1 is characterized in that: The specific process of the algorithm in step S4 is as follows: Step S4-1: Starting from initialization, the algorithm creates two empty sets and Save the candidate service deployment solutions and the final solutions separately, and initialize each application solver a k Potential edge server-service pairs Ω k and candidate configuration decision variables d ik , d ik Set to 0 and mapped to Ω k Each pair in ; Step S4-2: Initialize an empty set The edge server-service pair with the largest marginal benefit for each service is used to store Then, it calculates for each service a k The marginal value of all valid pairs of Among them, the effective pair (s i ,a k ) must meet the budget b k and edge servers i Constraints on available resources; Step S4-3: Next, for each a k , find the deployment pair with the maximum marginal value (s i′ ,a k ) and add the pair to the collection If no valid pair is found, a k Will be from Delete from; Step S4-4: Next, the algorithm starts from Select the pair with the largest marginal value as a candidate deployment pair and add it to At the same time, the deployment decision variables of the candidate pairs Updated to 1; Step S4-5: After that, it updates the service The service reliability benefit of the served users, given a candidate solution and a pair for For each user u j , if its reliability requirements are met, the user's service reliability benefit is updated to 1, that is, Λ′ j =1; Step S4-6: Then, of Each pair in contains And merge into a set Store the final deployment plan; Step S4-7: Repeat the above process until there are no services to be deployed; Step S4-8: Finally, return to the collection As the final service deployment solution.

Images